THE GOVERNOR

Non-fiction by Ethan Bishop

A Memoir of Hearing, Attention, and Thinking in Loops

V39 · November 6, 2026

OPENING — LET THE SONG PLAY

Picking a song on Spotify takes less of my attention now.

Before, I could spend ages thinking about what to play. I felt I had to consider every possibility. By the time I chose something, choosing had become another job in my head.

Working with AI on other things helped me put some of that work down. I still chose the music myself. Having the book there gave me more confidence to let somebody else's work stand on its own, to listen without rebuilding everything behind it first.

For a long time, I had struggled to explain how my attention worked. Writing felt like a way to survive that struggle. I needed somewhere to leave the connections so I would not have to keep putting them back together. I could return to a passage, change it, and leave it there. The events were familiar. Seeing them beside one another was new.

I still get pulled into another round of analyzing myself. Sometimes the book gives me attention back; sometimes I have to notice how much attention I am giving it. Both have happened while writing these pages.

Technology lets me begin work I once could only imagine. I can find more alternatives and keep several projects moving, but deciding which deserve my time has not become easier at the same rate. I think that gap can contribute to mental-health problems. In my own life, that question leads back to useful tools, missed obligations, treatment, and work I have not finished. Those are the places I can begin.

The song keeps me from making the story too simple. Some of the tools that opened more work also helped me put work down. I want to understand what made the difference, and what still has to change around me after I have found an explanation.

I began calling the limits I encountered governors. I knew the word from machines, where a governor regulates how something runs. At first, I used the word for a hidden childhood limiter—a way of hearing and thinking I had worked around for so long that losing it during my first episode left me uncalibrated. Later, working on antique machines, software, and my own recovery, I saw the other half of what a governor does. Without a regulator, a wound spring dumps all its energy at once and shatters the record.

A mind that works by simulating how everything connects—and now has AI to run those simulations even faster—cannot survive by understanding its way to the bottom of every problem. To stay sane and present in the real world, it needs governors: external records it can trust to hold its thoughts, and real-world limits—physical material, other people, budgets, clinical care, and deliberate stopping points—that tell the simulation when to shut off.

This book is my attempt to examine those limits from inside, including the times I got an explanation wrong. You do not need my history to ask what is shaping your own ordinary choices, or how much of you those choices require.

For now, the result is small enough to hear: I can pick a song and let it play.

AUTHOR'S NOTE — HOW THIS ACCOUNT WAS MADE

This book grew from my experiences, memories, records, and the conversations through which I tried to understand them. I worked with AI through repeated drafts, questions, and corrections. I remain responsible for the words and what they claim about my life.

I did not understand the whole story from the beginning. Sometimes I put an explanation on the page and returned after an event had changed what it could mean. I have left those changes in the story where they belong. I am still learning what to make of the life I have already lived.

Dated conversations, project files, screenshots, and later recollections helped me find my way back. They cannot all tell me the same things. Asking for an action does not mean it happened, and remembering a conversation does not give me a recording of its exact words. I identify recollections and paraphrases as such. My twin and everyone else in these pages have their own accounts.

Rather than burying the evidentiary boundaries, research context, and method notes in back-matter appendices, this edition weaves them directly into the story where those questions arise. Chapter 19 opens the hinge records and version history to show how the archive pushes back against a tidy narrative. Chapter 24 examines the psychological, clinical, and computer-science research that gives smaller, testable words to my experiences. Chapter 29 lays out the practical loop of model governance that a reader can try without needing to agree with my theory of my life.

PROLOGUE — THE GLITCH

The déjà vu hit while I was fixing somebody else's Wi-Fi.

I was already doing too many things at once. That is normal for me and sounds worse from outside than it feels from inside. Codex, the AI coding tool I use, was working through pieces of Realm, the enclosure company and manufacturing system I was building. I was also working on a voice system for the house and thinking about how Home Assistant, the smart-home software, could connect the devices and eventually connect to Realm.

Then the household Wi-Fi stopped working.

Someone in my household brought me the problem while I was already thinking through a version of it. Their laptop, router, and wireless access point fitted into the house system I was designing. Was the problem in one device or the network? The request landed among possibilities I was already considering.

That was when the room felt duplicated.

I mean the full-body recognition I have felt throughout my life: this moment already known, the next movement almost remembered before it happened. Déjà vu brings the certainty before the explanation. I feel I have been here, but I cannot retrieve the memory that would explain why. I had spent years being confused by it.

That morning, I found an explanation that made sense to me. Perhaps I had already imagined enough of the scene's shape. I had been thinking about the connections in the house and working through possible failures. A real failure had arrived along one of those paths.

The event was new. My mind had already been somewhere like it.

Perhaps that was why the fit felt like memory. One morning with a router gave me somewhere to begin. The possibility stayed with me because I recognized something of my other problems in it.

My mind runs ahead. It does not always run ahead correctly.

I am continually trying out what might happen next. A machine behaves one way, and certain parts become suspects. A person says something, and several replies become possible. A project grows until two of its parts might get in each other's way. I do not consciously tell myself every one of those stories. Often, pieces from different problems meet in an intuition before I can explain the connection.

I had been comparing my mind with the computer systems I built: saved information, different versions, work continuing out of sight. That morning I turned the comparison around. Of course computers look like minds. Humans built them.

I described the déjà vu to the AI. Its first answer was too general. I explained that the household problem belonged to the network work already underway. The next answer connected the event with the projects and possibilities I had been thinking through. It felt closer to understanding me than remembering a list of facts about me.

The AI was beginning to feel useful in that way. The unsettling part was that I was also using it to understand myself.

For years, I had brought ChatGPT problems: machines, relationships, businesses, episodes, code, animal care, motorcycles, decisions, and ideas I could not yet express clearly. I rarely sat down just to describe my day. I asked for help with something.

The problems contained me anyway.

What I corrected showed what mattered to me. What I argued about showed a distinction I needed kept. Each project showed how I responded when something resisted the plan. Over time, I could return to the exchanges and see traces of how I had worked my way through them.

I did not know what had happened inside the AI when its answer changed. I could see what happened between us: I supplied a missing fact, and the answer fitted better.

I reported that the Wi-Fi worked and the déjà vu had passed. Other connections stayed with me. My twin and I had communicated before ordinary English worked well for us. I had spent years learning to be understood by other people. I had noticed that changing the pressure in my ears changed how sound reached me. I had needed to know why an instruction worked before I could depend on it under pressure.

The AI's responses had come close enough for me to see more of what those experiences shared. To understand why that mattered, I had to go back to the life I had brought into the conversation.

CONTENTS

Part I — The Model Before I Knew The Word

1 — Two Boys, One Language

2 — The Child Who Needed The Why

3 — The Motorcycle Did Not Care About The Slogan

4 — The Shop Made Me Honest

Part II — Pressure And The Governor

5 — The Click Came First

6 — Wrong Outputs

7 — Recovery Was Recalibration

8 — Building Again

9 — My Brother Is Not My Evidence

Part III — Making Understanding Usable

10 — Computers Look Like Minds Because Minds Built Them

11 — Facts Are Not Understanding

12 — The Model That Agreed Too Easily

13 — The Translation Cost Collapsed

14 — The Enclosure That Would Not Stay An Enclosure

15 — The Animal Gets The Final Vote

16 — The Factory Is Part Of The Product

Part IV — The Collaboration Was Already Happening

17 — Two Models Learning Each Other

18 — The Archive Is The Third Participant

19 — What The Record Actually Holds

20 — The Book Ran Back Through Me

Part V — The Model Meets The World

21 — The Model Needs A Body

22 — The Governor Returns

23 — Six Days

24 — Words For The Questions

Part VI — The Work Meets Other People

25 — The Human Interaction Has More Than One Human

26 — The Same Room Is Not The Same Experience

Part VII — The Attention I Can Give Back

27 — The Representation Has To Earn Its Space

28 — The Title Is Small; The Thing It Points To Is Not

29 — The Human Keeps The Keys

30 — Choosing What Becomes Work

Afterword — The Attention I Can Give Back

Reading List

PART I — THE MODEL BEFORE I KNEW THE WORD

CHAPTER 1 — TWO BOYS, ONE LANGUAGE

My twin brother, Colin, and I learned to communicate with each other before our speech worked reliably for everyone else.

Before either of us could explain what we were doing, there was another child responding. We made sounds, paid attention, repeated ourselves, tried again. Our family called what came out of that twin talk.

Conversation in our family still overlaps. The lines between one conversation and another blur, and occasionally a Bishop mumbles something nobody can really interpret. We call it the Bishop family mumble. In our telling, it is part of being a true Bishop.

My father's parents were completely deaf. None of their children were. I wonder what it was like for my father and his siblings to grow up in that household, how they learned to use volume, attention, and tone, and what reached Colin and me through them. I do not know how much, if any, of it shaped the way we spoke.

I remember Colin and me speaking in mumbles, sometimes because we did not want other people to hear. The habit is clearer in my memory than anything we said. I cannot bring back a particular exchange or say what our words meant, and I do not know how much someone listening might have understood.

Looking back, it feels like code: something shared that worked between us. We needed to become understandable to other people too, but what we already had connected us. Twin talk is what we called it. There was no grammar book hidden under the crib.

Speech therapy gave me more access to other people. A word they could understand was a real success. Once a way of speaking became familiar, it could feel like simply being myself.

The spoken code faded. More people could understand me, and I could understand more people. School, friendship, and work became more available.

For years, twin talk seemed interesting mainly because people find twins interesting, almost as if we came with a private radio channel. During my first acute episode, it took on a much larger meaning. I felt an old route still present beneath ordinary language. I was not recovering childhood words, and I do not remember consciously translating between two complete languages when I was little. I was having a new experience of an old memory, at a time that could have changed how I understood it.

I began to say that I had learned English on top of the code. That remains my way of describing it. I wondered whether twin talk had been one of my governors: something I had learned to function through so well that I no longer noticed its effect.

I do not want that idea to turn our childhood communication into damage. It helped two children reach each other before everyone else could reliably join in. The same habit might help at one time and limit at another. It might also have been the familiarity of two brothers, without the separate language stage I later imagined.

I remember my ears as another barrier, but I cannot put the difficulties and adaptations in order. Years of learning and making things could have taught me habits I now trace back to twin talk. Our speech did not reliably work in public; we needed therapy. Something I do well today cannot tell me which childhood experience taught it to me.

Writing later gave me another way to reach someone. Describing a past relationship, I said we had needed to decipher each other and that writing had become important. I used twin talk to explain that too. Whatever the connection, I had something I wanted to express and sometimes needed another way to express it.

A written sentence stays after the effort of making it. I can return and ask whether it still says what I meant. The reader can return too, perhaps finding a meaning I did not intend. We have something to examine together, even though we have lost the tone, timing, and chance to interrupt that we would have had in conversation.

This book grew out of that effort. A stranger cannot bring the familiarity Colin and I grew up with. I have to find words for what I remember and leave room for what I still do not know.

In May 2023, I brought a long passage into a conversation and called it “the intro section of a book im writing.” I asked for help with an introduction and conclusion. The material and the need to write it were already there.

I had connected language, hearing, family, and identity, trying to give them one beginning. I understand why I wanted that. Communication had shaped so much of my life. As I returned to the passage, though, I found connections I would have to loosen. An explanation could be old and important to me and still need to change.

Two boys had communicated before they could explain how. They learned ordinary English through whatever ways of understanding they already had. One of them grew increasingly interested in what happened behind a visible result. That was me. Eventually I would turn that interest inward.

CHAPTER 2 — THE CHILD WHO NEEDED THE WHY

The first real “solo” CNC-machined part I remember making depended on getting a piece of aluminum to stay still.

For my senior design project at the University of Delaware, we had to make a router jig for a large plastic pipe. It needed to guide a cut around the pipe, a few inches back from the edge. A groove in a specific place sounded simple enough. Making the tool that would cut it became a project of its own.

We were doing the first project on the MakerGym's Tormach 1100MX. I was in charge of the makerspace, and this was my first time programming aluminum milling in Fusion 360. I had responsibility for the machine without already knowing how to make it do this particular job. We had to get it dialed in while making the parts.

Most of the difficulty was holding them. The program could describe a shape precisely, but the aluminum still had to remain where the program expected it. We built custom fixtures and worked with the tooling we had. Before the machine could make what we wanted, we had to make the arrangement that would let it.

We completed the project successfully. When an application later asked me for a significant CNC result, what came back was the work around the cut: the fixtures, the available tools, the new machine, my first aluminum programs. Testing that Tormach remains a highlight for me. The groove was the reason for all of it.

I had needed that kind of explanation long before university. I was never especially good at accepting a rule that had nowhere to live.

I could follow instructions. I became an Eagle Scout, managed stages, worked in shops, studied engineering, and operated machines that punish improvisation. My difficulty was with an instruction that stayed arbitrary. I could repeat it or perform it correctly and then lose the connection under stress, or when the situation changed.

Knowing why gave me another way back to the answer. I could work it out from what the parts did instead of trying to retrieve the right sentence.

For years I assumed everyone wanted that. Why stop at a procedure when you could understand what made it work? Meanwhile, other people learned the procedure and got on with things while I was still looking underneath it. Nobody could examine every detail before using a light switch, driving, taking medication, or sending an email. We all had to trust something we had learned. I often struggled to trust it until I could find the relationship for myself.

A résumé keeps a peculiar sort of memory. Senior patrol leader. Eagle Scout. STEM classroom aide. Stage manager. Lead set builder. Makerspace technician. The titles stay while much of the experience disappears.

Behind those titles were four consecutive years helping run a high-school theater program, teaching peers to use construction tools and work through designs. Later came the work of bringing a university makerspace from an idea to an operating shop. We selected equipment, planned the floor, started machines, managed staff, and built ways to keep track of training, permission, and inventory.

Someone needed to know who could use a machine and where their training was recorded. Someone needed to know what material was available. If all those answers disappeared whenever the person who set things up left the room, we had not finished making the shop usable.

I kept choosing work that asked me to leave something behind for another person. In scouting, my demonstration ended and someone else had to perform the skill. In theater, separate jobs had to come together as one continuous show. In the makerspace, people needed enough knowledge and permission to begin without being left to guess when to stop.

Taking over the difficult operation might finish the object while leaving the learner dependent on me. Giving only a procedure could leave them stuck when the material changed. But explaining everything I knew could be just as unhelpful. The person had come to make something. They did not necessarily need to carry my whole understanding before they could make their next decision.

I learned to begin with what they wanted to do. From there, I could explain the part they needed, show them where to stop, and help them recognize when to ask for help. Sometimes a procedure carried someone else's tested judgment while the rest of the explanation caught up. I could learn that way too, even if it did not come naturally.

Scouting gave us little choice about it. The group had to move, eat, stay warm, and carry its equipment. Weather, fatigue, or someone's limits could change the plan. We needed enough shared understanding to keep going when the original instructions no longer fitted.

In theater, a call sheet could be right and a cue still be late. Lights, scenery, sound, actors, props, cables, entrances, and timing had to meet in the same performance. I liked knowing how the parts depended on one another. I liked helping make an event reliable enough that other people could perform without thinking about every backstage detail.

My own woodshop had begun in middle school. I made a series of coffee tables from pallet wood. They looked pretty good to me and got me hooked on furniture. For the most part, making furniture feels boring to me now, but those tables were part of how I arrived at university already wanting to make things.

Engineering gave me a shared language for what interested me: forces, materials, tolerances, energy, motion, feedback, failure. In the MakerGym, those words met people who wanted an object, a mechanism, a prototype, a shape. They often knew what they wanted before they knew what a cutter could reach, what grain would tolerate, or which surface a measurement depended on.

I could tell someone which button to push and help them make one part. I was more interested in the conversation that came before it. What had to move? What weight would it carry? Which material did they want? What might fail, and how would they know whether it worked? Once those answers connected, choosing the tool became easier. That was the conversation I wanted when I was learning too.

Woodworking kept the conversation honest. Grain tore out. A joint gapped. A panel moved. A finish revealed the scratch I had hoped it would hide. The object kept the result of my decisions, whatever reason I had for making them.

A cut that is too short is not a debate.

I liked that. A drawing or paragraph could feel complete in my head, but then the cutter entered the material and the joint fitted or it did not. Years later, an animal would use a space I had made or avoid it. I wanted to make things because they could give me an answer I had not arranged for them to give.

When I call this an internal model, I mean the changing picture of my whole reality that has grown across my life. The shop belongs to it, but so does a conversation, what I hear, what I remember, and what I expect from another person. A lesson in one place can change what seems possible somewhere else. Moving among woodworking, engineering, software, manufacturing, animals, and business does not feel like becoming a different person each time.

I do not consciously see or calculate all of it at once. A simulation is the comparison I reach for when I try to describe how the connections feel. I do not know whether anything in my brain works like a single engine. What eventually interested me was how that picture had grown, and what had become so familiar that I no longer noticed its influence.

Long before I could put that question into words, I was paying for the habit in time. I could spend hours on a missing detail with little practical value. The useful next step might be to act with enough understanding while I was still mentally taking the gearbox apart.

My woodworking business made that cost plain. Work was coming in, and I knew how to make things. I loved running the business. I could also find the work monotonous and struggle to stay with it. Every possible improvement offered another reason to leave the current task unfinished in my head.

The unfinished things did not readily go quiet. Work, the house, animals, machines, vehicles, software, businesses, relationships, and things I had not built yet could all stay partly active. From outside it could look like chaos. From inside I could see the next question in each of them.

Improvisation brought another cost. It had worked often enough that I could confuse being capable with having a good enough reason to try. Believing I could engineer my way through a problem did not tell me whether the risk was worth taking. I learned that repeatedly, not always cheaply.

Hardest to question was the pleasure of an explanation that connected everything. Making observations fit together felt good. In the shop, the result could interrupt that satisfaction. When I began explaining myself, I could cling to a story simply because I liked how much it seemed to explain.

Repeated annoyances became tools, routines, and sometimes software. Years after my first episode, enclosure work would grow into Realm. It could look like a grand ambition in retrospect. Usually it began with irritation. Why was the information hidden? Why did a handoff depend on one person remembering an exception? I would notice a neighboring problem, then another, until a small repair had become a larger system in my mind.

I brought the same habit to myself, where the answers were harder to check. Thinking about a memory could change how I remembered it. A convincing story could make every new event appear to confirm it. I did not always stop to ask what I had actually observed or what somebody else had seen. The shop had given me a standard I was still learning to use elsewhere: an explanation had to survive what happened next.

Some of the motorcycle and shop repairs that follow happened much later in my life. I return to them because I can see the habit at work there. Then I can return to what happened when I turned it inward.

For a long time, I did not have an explanation for a pattern I now understand through ADHD. I connect that missing why with the pressure I felt to write this book. I called writing a survival tactic because that is how necessary it felt. I was trying to describe a life I could not understand well enough to stop rebuilding the explanation.

The child who needed the why became a person trying to write one. I am still learning to let it change.

CHAPTER 3 — THE MOTORCYCLE DID NOT CARE ABOUT THE SLOGAN

On a right-hand bend, the connection reversed itself in my head. Right turn. Left wheel. Which hand caused which movement? I hesitated, lost the committed lean, and drifted toward the left shoulder.

Nothing catastrophic happened. I could balance, shift, and brake. I knew the phrase used to teach countersteering: press right to go right. At road speed, a small forward push on the right handlebar starts a right lean. Under pressure, I could lose track of why.

Months later, in August 2026, I returned to that turn in a repair conversation. This was years after my first hospitalization. I had felt the failure on the road before I found an explanation I could use.

The Honda Pacific Coast is an excellent motorcycle for someone who likes systems and a terrible one for someone who wants to see them easily. Most of it is wrapped in plastic. The engine, cooling, wiring, and hydraulics hide behind bodywork that can make a simple repair feel like access surgery.

I bought mine old enough to have history and cheap enough that the history came without documentation.

On March 19, 2026, I was telling the assistant how much gas the PC800 would save compared with my 2003 Chevy 1500. Spring was coming, I loved riding, and I liked unusual bikes. The seller had said the hydraulic clutch would need bleeding soon. On the way to get it that evening, I asked for an inspection list.

The insurance quote added another expense: four hundred dollars for twelve months at what I said were the lowest coverage settings. I remembered paying eighty-two for the Yamaha through the same company. I had reasons for wanting the bike. Given my finances, I did not have a justifiable reason to buy it. Looking back, the purchase is one of the clearest examples I have of my loop—my simulation—making an irresponsible decision feel workable.

The motorcycle soon supplied work of its own. After too much cranking and a jump start, it blew its coolant reserve tank off while I sat in a DMV inspection line. The battery could show respectable voltage after charging and answer a request to start with relay clicks. The clutch leaked near the lever, its reservoir was nearly empty, and the idle was low. The cooling system held a mixture of coolant, water, and uncertainty.

By August 18, I was describing a bike that tried to lunge and could stall as soon as I selected first, even with the clutch lever fully pulled. I had to be careful to keep it running. On the centerstand it was much less happy in first than in neutral, although it did not fight hard braking in the way I was checking it.

The front brake reservoir was about half full. The clutch reservoir might as well have been empty. I asked where to find the clutch slave cylinder, the part that should turn fluid pressure into movement. Somewhere between pulling the lever and releasing the clutch, what I expected was failing to happen.

Then the battery interrupted. Two chargers had called it full after an overnight charge. One seemed to announce completion before the better charger had really got to work. I said it had charged to thirteen volts, yet the starter was clicking again. I guessed it was below about eight without measuring.

I wondered whether low voltage had kept the radiator fan from getting enough power after the jump start at the DMV. That remained a possibility. The battery, clutch, and cooling system each still needed attention.

The first store did not have the replacement battery. I asked about another nearby store and later reported getting a discount. Availability helped decide which part of the job I could do next.

Battery, then clutch, then flush. I could work with that. The motorcycle stopped being one enormous failure for long enough to give the next task a place.

As I worked the clutch lever, bubbles rose in the reservoir. The rubber piece under the cap remained, but the white plastic part was missing. I wondered whether that had anything to do with the low fluid. I replaced the battery, refilled the clutch, and worked air out of it. I found the idle adjustment hidden in plain sight. Then, draining the coolant, I lost the drain bolt and copper sealing washer somewhere inside the bike's plastic geography.

The local parts store could replace the bolt. It could not replace the little washer that made the bolt seal. I could order one. First I kept looking for something available.

Suggested parts were out of stock. An assortment listed outside diameters when I needed the inside dimension that fitted the bolt. I asked about plumbing fittings, washers elsewhere, and bolts whose threads stopped short of the head. A related part name was not yet a part I could use.

An electrical ring terminal offered another possibility. I wanted copper, and some of the suggested examples did not look like copper to me. The material had to give enough to seal; the right outline alone would not do it.

I bought a tinned-copper terminal, cut off its barrel, trimmed the ring, and ground both faces flat on my shop's sharpening setup. Then I asked how many extra miles per hour my precision-ground washer would get me. None, obviously. I had made a drain-bolt seal.

I had to find out whether it would stay one. The faces needed to be flat, the material needed to compress, and the joint needed to remain dry as the engine heated and cooled. Being pleased with the improvisation did not make it dependable.

The idle was around eight hundred. I asked how to raise it and, a few minutes later, reported that it was running strongly at twelve hundred. I waited for the fan. I thought coolant splashed on the exhaust during draining explained what I was still seeing, and kept watching.

The fan came on. The fan cycled off. The washer stayed dry.

That was enough to close that part of the job. I had not checked every hose or settled what had happened at the DMV, and my washer was still not the manufacturer's part. But the question in front of me had an answer for the moment. I did not need a complete theory of the motorcycle before I could stop working on its drain bolt.

Another motorcycle had already taught me how short such an answer could last. In June 2025, my Yamaha XJ650 was keeping me at work. The starter clicked and tried to turn without spinning fast enough. Hitting it with a rock had started it twice before. Now I wanted to know how I was supposed to get home if the rock no longer worked.

I suspected a dirty commutator, the contact surface inside the starter. Beside the parking area at work, I cleaned it with the 180-grit abrasive I had. The bike started. I rode home.

At home, the trouble returned. I had crossed the distance I needed to cross and brought the faulty starter with me. Now I could consider doing the work again with better tools and a better process. The parking-area repair had given me one journey, which had been useful. I still did not have a dependable next start.

In the August conversation about the Honda, I returned to the connection between the riding instruction and my hands. I had been practicing moving from side to side on the road and thought it helped. The why still had not fully clicked.

I asked whether, in a right turn, the handlebars were pointed left, centered, or right. Starting a lean began to separate in my mind from following the turn. A few minutes later, I thought of a bandsaw blade tracking on a crowned tire, higher in the middle. Something familiar from the shop gave me a way to approach the movement.

The resemblance helped me ask how a small change in position could produce a larger movement. A motorcycle also had a rider and a moving contact with the road; the bandsaw could not answer for all of that. I was borrowing a relationship I understood and seeing how far it would take me.

My working picture began with the handlebars doing a different job from a car's steering wheel. A brief steering input moves the front tire's contact with the road left, shifting the support beneath the combined weight of bike and rider so the bike can lean right. Once it leans, the front wheel settles into the curved path. The bars start the lean; the lean makes the turn.

That gave me a way back to the rule. I could follow the movement from pressure at my right hand to the tire's contact moving left and the bike leaning right. I still had to make the physical skill dependable, but the instruction no longer depended on remembering an arbitrary reversal.

The motorcycle had not changed. I could better imagine what it would do.

“Believe in yourself” has always felt incomplete to me for that reason. If I expect a control to behave unpredictably, encouragement does little for the reason I hesitate. Here, understanding gave confidence somewhere to begin. I had a particular connection to work on instead of a broad judgment that I was bad at turning.

The relief of finding that relationship was real to me. The road still contained gravel, speed, weather, traffic, fatigue, and opportunities for bad judgment. I felt confident enough about riding to work and then the DMV, while still hesitating over the rush-hour route.

The washer had stayed dry and the fan had cycled. The riding would still have to be done. I could carry a useful comparison from one job to another, but it had to answer to the new job when it got there.

The assistant gave the way I was learning a name. It called me a “model-based learner.” Then it said, “you internalize systems.”

I recognized something and answered with the operation I felt happening inside:

“internal simulation”

The phrase let me put the motorcycle beside woodworking, manufacturing, animals, software, and recovery. I could ask whether a familiar relationship ran through them. The road, the hesitation, and the recognition had been mine. In conversation, I found a shorter way to describe them.

I supplied an experience and an incomplete explanation. The model gave me words, I changed them, and the next exchange began with my correction. I kept the phrase because I could use it, and because I could return to it when it stopped fitting.

CHAPTER 4 — THE SHOP MADE ME HONEST

The shop is where my explanations go to get embarrassed.

It sits behind an old Victorian house in Delaware and holds more machinery than the building has any reasonable right to contain. Some of it is decades older than I am. Heavy cast iron. Big motors. Belt guards from times when manufacturers seemed to assume the operator already knew what would remove a finger.

There is a Northfield bandsaw, an old Unisaw, a shaper, drill presses, sanders, a computer-controlled ShopBot router, dust collection, a compressor, and enough other equipment that rearranging anything becomes an argument about space, power, access, and where material can move.

I keep the machines because I like what they make possible. A wide bandsaw changes which curves, thicknesses, fixtures, and repairs I can attempt. The router lets me take a design from the computer into a process I can repeat. The compressor gives me air-powered tools. Possibilities that would otherwise stay in my head have somewhere to happen.

The machines also put my confidence in its place. Being an engineer does not tell me whether a wire has grown brittle or a fastener has loosened since yesterday. I have to know whether power is disconnected, whether a stopped cutter is really free of stored energy, and what will happen if the thing I am holding slips. Familiarity does not make any of that less real.

Old equipment leaves much of the work visible. Belts move, shafts turn, bearings complain. The alignment matters whether or not there is a screen to report it. With practice, I can take some of it in quickly, without asking myself each question aloud. I still have to look. A lesson that has become instinct can be the wrong lesson for the machine in front of me.

In September 2025, the machine was my 2003 Chevrolet Silverado 1500 4WD. Colin, my uncle, my aunt, and I were in an AutoZone parking lot, way too late, trying to replace a wheel bearing.

I had asked the assistant about the bearing bolts. It offered a thread pitch. “That seems like the wrong pitch,” I replied. The answer changed, but the part still had to fit.

At 4:03 the next morning, I wrote, “We got it done.” Then I listed what it had taken: an angle grinder, hammers, and pry bars. The bearing had come off clean. “The bearing was missing all its bearings lol.”

In January 2026, a different bearing stopped the CNC. I wanted to know whether the router's lower bearing was worth replacing. Before I could decide, I needed to get far enough inside to reach it.

The router had a four-inch body held with hose clamps. Already I was thinking about a smaller, three-and-a-half-inch router. Could the same clamps tighten around it? I had not finished taking the old one apart, and I was looking for a way to put a cutter back to work.

Within an hour, the job had changed. The bearing was so damaged that its outer race remained stuck in the lower router body. The part I wanted to replace had left a piece of itself in the place the replacement needed to go. Then the motor fan broke during disassembly, and I could not find another one.

I moved back toward replacement: use the smaller router for now, buy another four-inch one when I could. The next day I was asking about the price of a used router. What had begun as a bearing repair had become shopping for another machine. I believe the router in it now has a four-inch body.

The Silverado returned to the work in April. It would crank and fire without settling into an idle, and giving it gas did not keep it running. A pressure tester cost about thirty dollars; a replacement pump was around eighty. I said I would buy the gauge. I could replace a pump, but I still needed to know whether the pump was the problem.

I removed the plastic bed liner and could not find the bed bolts from above. Two of the eight captive nuts were spinning. Before I could finish the fuel problem, I had a fastening problem. My questions moved from the bolts to the pump lock ring and back to bolt size, thread, and length.

With the replacement pump installed, I could hear it priming. The engine sounded as though it wanted to catch. I asked how much priming it needed, but the disconnected line was delivering no fuel and I reported zero pressure at the rail. The new sound had not brought the fuel with it.

I began looking at the ground connection and at the possibility that I had replaced a good pump. “If it is the ground the old pump might’ve been good fuck me.” The large black ground wire seemed to go into the harness rather than onto the frame.

On April 23, the truck was running. I blamed the replacement pump's wiring directions, which I said had given the wrong connections. I was certain again about the old one: “No old pump was def bad. No way it wasnt.” The truck ran. Whether I had understood everything that went wrong was another question, even if my message left no room for it.

Bishop Woodcrafts grew among jobs like those. There was no smooth progression from discovering a passion to designing a logo and becoming steadily successful. There were cabinets, furniture, repairs, built-ins, measurements, budgets, material orders, finishing, transport, installation, and late changes to plans.

A client's house, money, and tolerance for dust and delay became part of every design. Something beautiful that they could not afford needed changing. So did a joint I could not assemble in the room. A piece could look right on the computer and refuse to turn the corner at the top of the stairs.

One January morning in 2026, I was looking at a red-oak dining top with three coats of black stain on it. It was still splotchy. We had sanded it to 120 grit before applying the M. L. Campbell Woodsong stain. I wanted the color to even out.

This was the job I had taken with the woodworking owner who hoped his next hire might eventually take over the business. I had imagined a large future there. Now we were disagreeing over the surface in front of us.

I told the assistant that my boss had wanted us to stop at 120. Then I sent another message: in my own shop, I would have gone to 180. I wanted that distinction understood even as I asked what to do with the stain already on the wood.

His next instruction was to sand with 400 until the color looked more even and then stain again. I brought that back to the conversation: “i don't think this is gonna work”.

My preferred preparation could not undo the three coats. Whatever I thought should have happened earlier, we now needed to decide what to do with this top. I remember that it turned out okay. When I returned to it for the book, I could no longer remember what I had done to get it there. I still thought my boss had been too picky about the finishing.

Marine carpentry made access and sequence harder to ignore. A boat would not stay square because a drawing preferred it. Surfaces curved, access disappeared, and water found what I overlooked. Things moved, flexed, swelled, and vibrated. Eventually another person might need to reach the fastener I had installed. I learned to include that person in the design.

The Victorian house held other people's decisions at a larger scale. Repairs, budgets, methods, and workarounds occupied the same walls, usually without notes explaining them. Opening a wall exposed another version of the plan. Later, Git—the tool for tracking changes to code—felt familiar partly because of that house. I knew what it was to work on something shared, with old compromises beneath the surface and consequences reaching into apparently separate rooms.

There was a literal governor in the parlor.

The Victrola looked like furniture because early machines often had the decency to become part of a room. Inside were a spring motor, gears, a turntable, and a governor to regulate the speed.

On July 15, 2026, I could not turn the crank. I asked for a quick start-up guide and found the platter locked regardless of the brake position. Moving the brake had not reached whatever was holding it.

The crank itself stopped me lifting the motor out. When I discovered that it unthreaded, I could finally remove the motor. It appeared frozen in grease, and one of the governor's three spherical weights was missing. Getting a better look had given me another problem.

I tested what would move. I could slowly rotate the turntable. The large barrel gear turned with difficulty. There was a little movement in the cylinder that I was trying to distinguish from movement in the gear. Then I became certain that the vertical shaft into the platter was frozen. Each part I could reach changed what I thought was wrong.

The assistant needed to follow those changes. Knowing the right names for the parts did little good if it misunderstood which one I had moved. Eventually, after spinning the platter many times with a drill, I said it sounded pretty okay with two weights. I had to correct the conversation again: I had left those two weights in place. I was describing what I had done, while the assistant kept answering for a different possible repair.

I was angry. I wanted help with the machine as it stood now. Advice about a stage I had already passed only added another thing I needed to explain.

I wanted sound good enough to enjoy, and I said it was done for now. Getting it done and doing it properly were separate ambitions that day. I had stopped without knowing whether its speed and balance were right or whether it was safe. The restoration was unfinished, and the assistant kept treating the job as open.

Then a record shattered under the needle.

I had needles and records with the machine. Later I described a constant scraping sound that occurred only with a record playing and wondered whether I was hearing needle scrape. I had got music out of it. The record could still pay for something I had not understood.

I began asking about the reproducer gaskets, needle materials, the needle's angle, and how much restoration was worth doing. Enjoying the sound did not end those questions. The machine had more than one way to tell me about its condition.

Calling something a governor could make it sound like an obstacle I ought to remove. Here I could see why the restraint belonged. The spring held energy, and without regulation it could release that energy too quickly. The spinning weights, springs, and friction answered excess speed with resistance. The governor helped make the music usable.

Repairing it would restore that regulation. Getting music from the incomplete mechanism had been a different stage, and I had not wanted to complete every stage at once. I could understand proper restoration and still spend more time or money than my use of the machine warranted. I could also decide I had done enough before noticing what that decision would cost.

The broken record stayed inside that question. I wanted the assistant to respect my decision to stop. I also needed to let the physical result change my mind about whether I had stopped in the right place. My impatience could get a job moving without settling its consequences.

The comparison came late in my life. This was July 2026, after my first hospitalization and after I had begun finding the language for this book. I had already been thinking about limits. The Victrola gave me one I could touch and test.

By July 29, I was rebuilding the governor, working on reproducer gaskets, and dealing with a broken balance spring. I wanted the machine functional while I waited for replacement springs. Later I said it sounded great and hoped the new springs would improve it again. I had records I wanted to play.

With several things changing at once, I could not divide the improvement neatly among them. The governor controlled speed. The reproducer did another job. The record, my ears, and my attention also entered the listening. I could like what I heard before knowing how much credit each change deserved.

A physical part could answer me in a way I could not expect from my own hearing. I could see whether it fitted. I could not open my ears like a spring motor, however useful the comparison felt.

In early August, I wanted instructions another person could use to play the Victrola. I corrected the brake description, asked the assistant to consult the manual, and compared the instructions with what I had been doing. Eventually the document described my regular process. My corrections had helped put something I already did into words someone else could follow.

I wanted a working machine, music I could listen to, and a way for another person to use it. The governor gave me a comparison I would keep returning to. The Victrola still had parts to repair and records to play.

PART II — PRESSURE AND THE GOVERNOR

CHAPTER 5 — THE CLICK CAME FIRST

I was in my truck, parked on the street beside my house. I remember a sudden clarity and feeling that I could breathe more easily. I also felt strange enough to call my mother and ask her to come immediately.

The clarity and alarm arrived together. Something seemed to make sense, and I needed help now. I have sometimes said it felt as if lifelong asthma disappeared in an instant. It felt that dramatic; I cannot call it a cure. I also remember becoming immediately more emotional in my own way.

I went with my parents to their house. Eventually they called the police for help, and I went to the hospital. Other people had to take part in deciding what happened next.

I call the recognition the click. It comes first in my memory, at the beginning of an acute bipolar episode with psychosis. I have dated that episode to both March 2023 and 2022 when telling it later, and I still cannot settle the year. My mood, sleep, energy, judgment, and functioning changed enough to require hospitalization. Habits I had lived with for years had become part of a much more dangerous experience.

When I tried to understand the episodes afterward, I returned to relationships too. I had left a relationship that lasted three years. Colin thought failed relationships contributed to the episodes and to the ideas in my head. I wondered about that connection, but I could not put everything in order or say what had caused what.

For most of my life, I had done something with my ears without noticing. During episode 1, I became aware of it. Afterward I could deliberately change the pressure and alternate how I heard. I had not knowingly practiced a hearing technique since childhood. The private speech Colin and I used had faded with age and speech therapy; this habit, as I understood it, had stayed unnoticed.

I noticed the volume of the outside world changing. I have not said whether that first change made it louder or quieter. I also noticed that I was doing something I could choose to change. Around those two discoveries, I began building a much larger explanation.

Now I describe the effect as a selective conductive reduction in hearing. I do the action, stop, and the changed hearing stays until I pop my ears. Following what I hear can require more thought, unless it already fits what is going on in my head. Familiar context can do some of the work for me.

Recognizing the habit made me question the conditions under which I had learned to understand the world. What had I learned to expect while listening that way? How much of my picture of myself had grown around it? I could ask those questions long before I could answer how much it had shaped my learning or what it had to do with the episode.

It felt as though I was in two realities. I did not know which way of hearing to trust: the pressured way I associated with twin talk, or hearing without that pressure. Sometimes the second sounded normal. The same room and conversation could reach me differently. I felt I had found a familiar way language had been arriving, and that it no longer had to arrive that way.

In telling the story, I have sometimes run that awareness together with a later improvement. I did not notice sound actually starting to improve until after episode 3. Switching did not always make speech easier to follow either. What was already in my mind, and how the conversation fitted it, still mattered.

I now think that, for roughly twenty-three years, the way of thinking I call twin talk had hidden some difficulties I associate with ADHD. I link higher pressure with that older way of thinking. I could arrive at an answer without seeing all the work behind it. That is how I make sense of what happened; I do not know whether my mind developed that way. The episode was when I noticed the habit. I do not know whether I might have noticed it another way.

The word governor came later. I thought of learning to operate a machine without knowing a limiter was part of it. Over years, you would learn what effort produced a result, what danger felt like, and what you could do. Change that limiter, and the expectations you had learned might no longer help you. I began calling the childhood intermediary the Governor because I felt my understanding of myself had grown around it. After the click, I no longer knew how to operate in the same way.

The feeling was profound. It also arrived with an illness severe enough that I could not safely rely on my interpretation alone. Perhaps the recognition helped destabilize me. Perhaps it was part of the episode or happened alongside it. I could notice a real difference in hearing and misunderstand its meaning. My call to my mother and the hospital stay belong in the same story as the clarity. I needed care before I could settle why.

For years I had judged my hearing by whether I could participate. I could listen, answer, work, and learn. I thought my hearing was fine. I was not counting how much work might happen after a sound reached me.

When I returned to older descriptions, difficulty was already there. In February 2025, I had written about sound feeling distorted and sometimes overwhelming, changing pressure deliberately, avoiding speech when it became difficult to manage, and feeling as though I had needed to relearn English. Those were the words I could find then, even while my broad judgment of my hearing remained fine.

Other family members have described an ear action they say is similar, but painful. I had done mine every day for years without that pain. I wonder about the difference. I do not know whether what happens in their ears is the same as what happens in mine.

I remember having pressure measured and worrying over the answer. In one account, I described extreme pressure I could hold at will and concern from the people measuring it. In another, I said an ear doctor found the pressure too negative and prescribed Mucinex. I still cannot put those appointments and results in order. At the time, I could not understand why I needed medication for a state I could change. What could one pressure check tell me about something I could keep changing?

After trying several times to fit the answers together, I wrote: I’m so confused.

The assistant offered diagrams and explanations. Some seemed to fit. The temptation was to take that fit as a discovery before anyone had tested it. Because the experience had become tied to psychosis, I wanted a measurement to give me something beyond my own interpretation. I also wanted it to prove the larger story I had built. I was asking one number to do more than it could.

I learned that a tympanometry test examines how the middle-ear system responds to sound; it does not directly measure hearing sensitivity. Even a reliable pressure finding would leave me asking how well I understood speech and how much effort listening took. Those questions still needed their own answers.

In my October 12 account, I said a later test had found conductive hearing loss and another test was coming. My ENT has diagnosed baro-challenge Eustachian tube dysfunction and documented two pressure states. That gave me something physical to begin with. I do not have the hearing loss's severity, starting date, or childhood history to add here. I have used the phrase born deaf, but the later findings cannot tell me that I was.

When I already expect what someone will say, following them can be easier. When they move elsewhere, I may have to work harder. I want to understand how the documented pressure states relate to the reduction I describe, and how both relate to the effort of following speech. Twin talk carries my history, but it need not explain every moment when more context or a repaired misunderstanding helps me get through a conversation.

An ordinary answer can conceal an extraordinary amount of effort from the person giving it as well as from the person listening. That is one reason the hearing experience reminded me of the shop. A result could look the same while the work behind it was different. I had learned to ask about that work in a machine before I noticed how hard it could be to see in myself.

The cause still matters to me. I want to learn how my hearing works with the awareness and control I have now, and which of my old expectations still fit. I want care to make room for the ear condition while helping me examine the meanings I have attached to it. Pressure, sound, speech therapy, hearing loss, and the episodes each have histories I am trying to connect. Some of those connections may yet come apart.

I also learned that I could feel enormously certain when certainty was least trustworthy. The click came first in my memory. The call brought someone else into what I was experiencing. By the time I reached the hospital, my explanation could no longer be the only basis for deciding what help I needed.

CHAPTER 6 — WRONG OUTPUTS

A nurse said I was yelling at her. I distinctly remember believing I was not.

I was sitting in a hospital hallway in front of a window. I cannot remember which stay it was, what I was wearing, or her exact words. I remember where I sat and the feeling that I was in two realities at once. In one, people seemed mean. In another, they seemed kind. I was trying to decide which reading of the same encounter to trust.

Later, I called it a game with the nurse. I was trying not to look too crazy when it mattered, and I feared being isolated. I cannot say that anyone had threatened to isolate me. I remember being afraid.

Calling it a game makes me uncomfortable. There was exhilaration in the idea that I might solve the puzzle by managing what people received from me. I could try to appear coherent and feel I was solving the problem while making it harder for the people responsible for my safety to see what was happening.

The volume disagreement stays with me because I thought I could hear my own voice. Perhaps I misjudged it, misunderstood her, or forgot a change in the encounter. Perhaps she read me imperfectly too. In that moment, there was no version of what had happened that we could both rely on.

Years of my life could seem to join up in my head while what I said was unusable to the person in front of me. I call these wrong outputs. The phrase helps me describe a failure, although it makes the repair sound too simple, as if I could find a faulty part and run the same process again.

I had no unaffected version of myself standing outside to inspect it. The mind judging danger, intention, and reality was the same mind deciding whether its explanation was sound. My ability to explain was involved in what had become unreliable.

Looking back at my first episode, I picture my model moving faster than life could answer it. One connection supported another before I checked the first. More and more seemed to fit. I could feel that as confirmation even though the agreement was happening inside my own explanation. That is how I describe the experience now. It does not tell me what causes bipolar disorder.

People spoke. Messages arrived. My body gave me sensations, memories surfaced, coincidences happened. An ordinary event could feel charged with meaning. A thought could arrive with the force of something I had witnessed. I could notice a real pattern and follow it much farther than it warranted.

That is part of how I remember psychosis from inside. What happened did not have to be imaginary for my interpretation to be wrong. Once I accepted one detail as confirmation, the next arrived in a world that already seemed to agree with me. The explanation began choosing what I noticed.

I could not reason my way out by producing one more internal proof that I knew what I was doing. What I did over time, and what other people saw, had to matter too. Feeling full of understanding and possibility did not make me safe.

I have wondered about stress, exhaustion, and conversations that made me think too much about myself. I was looking for a pattern I might recognize next time. I had not found a cause I could be sure of or a switch I could reliably avoid.

Other people had to make safety decisions. Intelligence offered no exemption. Engineering gave me practice connecting things, and language gave me ways to make those connections persuasive. If correction felt like a failure to understand me, I could use those abilities to defend the very account that needed checking.

I could say the explanation was simply too large to communicate. Then every failed attempt to communicate would seem to support keeping it intact. In the shop, a wrong part could stop that argument. In the acute state, even a check could feel like an obstacle to something I already knew.

Later, watching an AI build convincing sentences on a false premise gave me a comparison for the problem. The sentences could agree with one another while failing to agree with the world. My psychosis involved a body, biology, emotions, and a life that a language model does not have. Still, the comparison helped me question what kind of agreement I had mistaken for truth.

Sleep, time, treatment, clinicians, and people who knew how I usually was gave a thought other chances to be examined. I did not have to decide everything at the moment it felt most complete. I could return after mood and energy had settled.

The other people were not infallible. My family brought fear and history. Clinicians saw me at particular times and through particular categories. I could be misunderstood, and treatment effects could need discussion. I had to leave room for those questions without turning every mistake or misunderstanding into a reason to trust only myself.

Hospitalization brought limits, medication, observation, and routine. I could receive care before I agreed with every part of its explanation. Looking back, I can understand some of those restraints as another Governor: other people were limiting what I could do while I believed I understood the situation.

Some limits protected me. Some of the experience frightened me, and some may have involved incomplete understanding. I can be grateful for help and still question how it was delivered. My fear could not by itself tell me which danger other people were responding to. Wanting control back did not mean I was ready, but needing help did not remove my dignity or my voice.

After the acute episode, I kept asking what had changed. “Bipolar I with psychosis” mattered for treatment and risk. Saying that I no longer knew how to operate as before helped me describe the disruption. I needed ways to communicate that experience without turning my own explanation into something that overruled care.

My mother had come when I called. My parents had asked for further help. Hospital care had entered the story. I could still make an explanation of what happened, but I now had other people in the work of examining it. Having an explanation could no longer settle whether I was ready to act.

CHAPTER 7 — RECOVERY WAS RECALIBRATION

In March 2025, I needed to send a message about missing work.

I had been offline from almost everything through the weekend. I described myself as being in a depressive bipolar episode since Friday. I think I was working at Eagle Group then. Now I was trying to find words that would make Tuesday possible.

The draft began with an apology. I said I was bipolar, that the weekend had spiraled into an episode, and that I had disappeared. I was finally feeling better and had an appointment with my psychiatrist later that day. Could Tuesday and Wednesday work?

One sentence was particularly severe: This was completely unfair to you guys. I might not ask someone else in the same condition to accept all the blame so quickly. I wanted to show that I understood what I had done, and part of me tried to do that by making the fault entirely mine.

The person waiting for the message had missing work and unanswered contact to deal with. My explanation could help them understand without removing the disruption. I needed to repair what I could while admitting that the illness had interfered with the very act of making contact.

I still do not remember whether I sent that message or whether Tuesday and Wednesday were accepted. I eventually stopped working there. I could find the words and still fail to get them to the people waiting.

That was some of the work after an episode. There was no former version of me I could simply restore. My explanation of the Governor might be wrong about the brain, but I no longer trusted my old picture of myself. I could not trust a new picture just because it felt clearer.

Questioning it could feel like questioning whether I could trust myself at all. Was I holding on because it was accurate, or because it seemed to be my remaining connection to who I had been? Meanwhile, ordinary things needed doing. Sleep. Show up. Answer the message. Take the medication. Finish the work. Hospital care could help interrupt an acute state; outside, I had to learn how to judge my thoughts while living among their consequences.

On March 31, 2025, I asked how to tell Colin that I had left the marina job. It had been about two weeks.

I had gone back a few times while he worked on our boat. The marina could still be a place I visited after it stopped being my workplace, and I let that true detail maintain a false impression. I described what I was doing as creating an illusion.

I expected him to be shocked and to press hard. Why had I left? What was my plan? I was not really sure why I had left. An interview the previous week had seemed to go well. Woodworking was going well too, and I hoped to hear back soon.

The marina work had bored me. Other jobs had offered large, exciting projects; I wanted a bigger impact than I felt I could have there. In the conversation with the assistant, I tried to explain that chaos was how I operated and was bad for me. Did I need it to function? Should I talk to my psychiatrist about that?

I also wondered whether bipolar disorder had something to do with hiding the departure. Whatever its cause, the concealment was now part of what I needed to tell him. The conversation would not bring back the job, ease the money pressure, or guarantee a gentle response.

I wrote one word:

“Terrified”

Should I go and talk to him? Should I show him the conversation with the assistant? I asked for a play-by-play of how it might go. I kept trying to arrive at the encounter through a version I could examine beforehand.

The room was inside Colin's 1990 Hunter 35.5 sloop. The boat was on the hard for the winter, out of the water and up on the hill at Worton Creek.

I did tell him.

I remember him saying that he already knew, and that he wanted me to be more open with him in the future.

All my effort to predict his answer had taken place without knowing what he knew. Now there was an answer from him. There was also a request I would have to keep responding to after the conversation ended.

I call that difficulty reaching out the 10,000-pound phone. I like to feel independent. Needing to call other people can make me feel less so. I can understand why contact matters and still find the phone almost impossible to lift.

That is how Colin acted as a governor for me then. My loop needed to hear from the person I was imagining. He asked for more openness in our relationship. The next update was still mine to make.

An outwardly steady day could take a great deal of checking inside. I did not want to become suspicious of every connection I made. I also could not trust an idea simply because it felt right. Sleep became one of my most useful ways to notice the difference.

A deadline could make one more night seem reasonable. A project felt too important to leave; an explanation seemed one conversation away from completion. My body did not care why the hours disappeared. With less sleep, my energy could rise and my judgment worsen. Connections came faster, work felt unusually possible, and acting became easier.

It could look like my best self. I did not always feel bad as trouble approached. Sometimes I felt exceptionally capable, which made restraint harder to accept.

I tried to decide on limits while I was able to agree with them. Sleep, medication, time, and other people's observations gave me ways to return to an idea. If it still made sense after rest, I could give it more weight. If it seemed unable to survive a pause, I needed to question how much of its force came from urgency. I wanted a decision made today to leave tomorrow's version of me some room to choose.

That room mattered in my family too. In June 2025, I expected to be late for work and had not returned my mother's call. When I finally called, she was five minutes away. I said she was worried because I was bipolar and asked how to get her and Colin to stop driving over at the slightest concern.

Then I corrected myself. I wanted them to worry. A six-hour waiting period we were discussing felt too long. Trying to explain the boundary was making my head spin. I had asked for less concern when it was the arrival I wanted to question.

In the truck during the first episode, I had wanted my mother immediately. That June morning, I wanted room to leave for work without having asked her to come. A missed call carried both possibilities for us. I could feel pressured by her approach while wanting her close enough to notice when something was wrong.

Stop worrying was too broad. Wait six hours was too distant. I needed to give people enough accurate information that a missed call did not have to carry the whole story, and we needed room for an ordinary delay. The support also had to work at times when I found it inconvenient.

The difficulty continued. In March 2026, I did not call or text my mother one day because I did not feel like it. She made a group chat with all my tenants and asked them to ask me to call. I was “royally pissed off.” To me, bringing them into it crossed a major boundary.

I still become very angry when I feel she oversteps. Looking back, I also see a governor I chose not to listen to fully, although I think I should have listened more. I can reconsider what I refused to hear without deciding that involving my tenants was the right way to make me call. My need for a boundary and her effort to reach me both remain in the story.

I used the same idea of deciding before the difficult moment when I planned trips outside. I love backpacking, canoeing, distance, weather, and a loaded pack. I can keep working through strain. With Maeve, my dog, beside me, determination tells me even less about whether continuing is right. Her condition needs attention of its own.

I wanted reasons to stop before the trail made finishing seem like the only acceptable ending. Damaged paws, heat stress, vomiting, refusing water, a collapsing pace, or weather beyond the route's margin had to count. So did too little sleep or growing instability in my own mood and energy. The itinerary could not decide for the dog or for the person actually walking it.

Do not keep moving merely because the plan says so. It became one of my best recovery rules. The version of me who learned something on the trail needed permission to disagree with the version who planned it. Learning to change course without treating that as failure took years.

During residential treatment in August 2025, I wrote that agreeing to a month had definitely been the right decision. I also wanted to go home.

On day twelve, I asked about the partial hospitalization program recommended afterward, another level of structured care. I did not think I needed it. I cannot say here whether I later completed or declined it, or whether I was ready to leave.

My family wanted me to stay through a family weekend at the end of the month. I understood their concern. I thought we would do better at home with weekly family counseling from one consistent therapist or counselor. I did not want our whole history to depend on one concentrated weekend.

If their concern always overruled me, wanting independence could itself become a reason to keep it from me. If my confidence always prevailed, independence could become a way to shut out what they saw. I wanted a part in deciding which limits came home with me.

I proposed support groups, a monitored electronic pill box, family therapy or counseling, a place on a new therapist's schedule, and a psychiatry appointment as soon as possible. These were ways to keep contact going. An appointment would put another meeting ahead of us. A pill box could make one routine visible without requiring everyone to rely only on memory and reassurance.

I wanted my family to see the work I had done since arriving. They knew the failures before treatment. The people in the program saw daily participation. I knew effort inside myself that might be visible to neither. I wanted that effort to count, while understanding that feeling confident was not enough for anyone to rely on.

Weekly counseling appealed to me because the next conversation could exist before the current one became too much. We would not have to settle everything at once. I was trying to replace promises with recurring contact, to move home with support still attached, and to have a part in rebuilding trust.

While in treatment, I also asked for help writing to an employer who had let me go. I wanted another chance. I wanted the company to consider the work I had done on my mental health.

I described significant medication adjustments and a twenty-eight-day residential program. The assistant returned a polished draft. Too polished. I answered with two words:

Less AI.

I needed language I could recognize as mine. Better sentences could quietly make the crisis look orderly and the recovery complete. The help had to let me speak without inventing a more convincing version of me.

The revised draft acknowledged why the employment had ended, connected my silence with a serious mental-health crisis, described what had happened since, and asked to return. The model supplied those sentences. The desire to ask for the job back and have my work in treatment considered came from me.

Treatment could not become attendance in the past. Medication changes could affect what I did next; they could not supply a call the company had never received. I wanted to explain the silence without erasing its effect and describe change without promising I could never become ill again.

The company could still say no. I was returning to a world that remembered what had happened, hoping it would consider something newer too.

I sent the letter. They never replied.

Other messages from 2025 resist a smooth recovery story. In January, I had missed the start of a workday. In February, mental-health stress interfered with a planned work obligation. In May, I missed a psychiatric appointment and needed to contact care. By July, I was describing missed obligations, pressure from spending, and withdrawal from communication because I did not know how to explain my state.

Money, work, appointments, and unanswered messages stayed inside the word recovery.

When I described my finances for the book, I said my net cash flow remained near zero despite trying hard to pay older debts. I associated those debts with times when my model had not been aligned with myself. The loop seemed to keep feeding itself while I tried to reduce what I owed. Asked for an example, I pointed to the Honda PC800.

Before anticipated treatment, I also planned care for the animals. Heat, water, food, cleaning, and observation would still be needed while I was away. Writing it down was how I tried to provide for a responsibility I knew would continue. The plan sets out what needed doing; it does not tell me whether every task happened as written.

There were later episodes and later treatment. Understanding the Governor had not made me permanently stable. Bipolar disorder, sleep, medication, stress, recurrence, and overconfidence remained part of my life. The mind that could make useful things could still begin reading the world incorrectly.

In the shop I knew that understanding how a saw could hurt me did not remove the need for a guard. I needed the same patience with myself. An insight could help me notice trouble or explain it to someone. It did not have to authorize every action it suggested.

I could build reminders and routines for days when I might not remember why I needed them. I could keep an idea without acting immediately. I could try to notice when useful energy was becoming a pace I could no longer check.

I wanted to understand the effects of medication on my own baseline, rather than compare myself with a hypothetical person I had never been. I thought of medication as adjusting that baseline, another governor to understand: how was it affecting me, and why? Therapy could give me another observer, different questions, and time. Routine could protect the conditions in which I made choices. I had to distinguish the Governor I believed I had lost from limits I still needed. The Victrola would later give me a machine in which restraint protected the music.

Some limits became agreements with future versions of myself. Go to sleep. Leave an irreversible decision alone at the height of certainty. Let someone else read what I have made. Keep a live system read-only until testing supports a change. Feeling faster did not make the brakes unnecessary. I gained more control by admitting that my state could change than by expecting one state to decide everything.

That is part of what I mean when I say I am often just vibing now. Sometimes I say it lightly. I also mean that every new event does not have to protect one final identity. Engineer, woodworker, developer, caretaker, brother, patient: I can be all of those without asking any one of them to explain the rest. I can care intensely about a project and still change what it means to me.

Building again would give me things outside myself to answer to. The house needed work. The animals needed care. A finish had defects or it did not. I could find something useful in those jobs as long as I did not ask one successful result to tell me I would always be well.

CHAPTER 8 — BUILDING AGAIN

In September 2025, I was laying nail-down hardwood when I realized I had forgotten the underlayment.

Now I needed to know whether to take the floor apart. Before starting, it would have been easy to say the layer belonged under the boards. With boards already down, I wanted to know what it did in this floor. Structure, moisture, and noise each gave me a different question. Reversing the work had a cost too.

I added: 1884 house. I don’t care.

Of course I cared. I was asking. I did not want to rebuild an old house to match an ideal new assembly without knowing what the difference would mean. The existing materials and geometry were already there. A straight line in a drawing would not straighten the room. Age did not make a defect harmless, but neither did taking everything apart guarantee a better repair.

Then I noticed the flutes on the backs of the boards. Were the grooves there to help the wood sit, move, or release stress? Did they matter to the finished floor, or mainly to the way it was made? The assistant gave me an explanation, though it had not checked a flooring specification.

I immediately went farther. If I ran out of boards and made more, should I flute their backs too? I did not yet know that I would run out. Already I was thinking about matching tongue and groove, copying relief cuts, and deciding which factory features mattered in something I might make myself.

The project grew as I tried to understand it. I wanted a new board to fit and function among the others, and copying a groove without knowing why it was there could be as mistaken as leaving it out. It was a standard I had failed to apply before starting: I was examining the function after overlooking the step.

I left the floor down without the underlayment. It is in a tenant's room, and I have not been in there for a while. I do not know how it is holding up.

The house gave me work after hospitalization, but no clean beginning. I built furniture and cabinetry, bought an old house, and took responsibility for animals that needed care whether I felt inspired or exhausted. Plumbing, gates, networks, machines, and vehicles kept needing attention. People, dogs, and reptiles lived among decisions I had already made.

Repair did not always mean taking things apart. Sometimes I needed to understand an imperfection well enough to continue with it. Sometimes it really did need undoing. The scale of the effort alone could not tell me which answer was right. I had the same difficulty after missed work or treatment: the consequences were present before I decided how to address them.

In February 2025, woodworking had been going well and enough work was coming in. I was struggling to prioritize it. I considered a business that could grow with less physical labor and proposed woodshop hours I would try to keep. That account does not tell me how those hours worked out. What I wanted was a way to let a new idea wait while I finished something already promised.

Designing the next object was often more exciting than repeating the labor on the current one. The queue still needed my attention. That February, I had built one custom enclosure for a client and wanted to make production enclosures and a website. The custom work came before the later AI design conversations. The production business was still something I hoped to make.

One enclosure could teach me what I wanted to repeat without showing how expensive repetition would be. I had building experience to bring to the idea. I still had to find out which parts would carry into production and which would keep demanding my hands.

Later that month, I said the paint was good and still had bumps. The sprayer had not worked perfectly, and the next day I was asking how to buff the spitting out of the paint.

In another February exchange, I had finished the final coat on a set of doors and decided not to rush the drying. I like that small decision. The material needed time, however ready I felt for the next step.

A bump would not disappear because I had become interested in a new design. I called parts of my work lazy, but the word gave me little help with the next coat. I needed to return to the finish, decide what came first, and ask whether the new business idea deserved the time it was taking from a customer. Calling myself incapable would hide those separate jobs. Calling it all curiosity would leave them unfinished.

I started and restarted businesses. I built software and took jobs in which drawings, databases, machines, and people disagreed about what needed doing. I failed to finish things, returned to some, and lost track of others. I kept having to build again. No single decision settled recovery.

In October 2025, I interviewed with a woodworking owner who wanted his next employee to stay for the long run. He hoped that person might eventually take over the business. The next day, I wrote that I thought I could be the one. I had always wanted a larger business.

I was about to call him to ask questions and try to accept. I asked the assistant for a little phone script. Fourteen minutes later, I asked for a text to send.

I could picture a large future inside someone else's business. The immediate task was finding words to say I wanted to proceed. His hope for succession was already meeting my wish to build something bigger, while I was still trying to make contact.

I took the job. He was the boss from the red-oak dining top. The top turned out okay. The job ended badly.

I accepted another offer and did not tell him until the day before I was supposed to start the new job. Then I ghosted him entirely. I was overwhelmed and did not know what to do.

I could not simulate a version of the conversation in which I did not disappoint him. Anticipating his voice made me hate myself. My loop froze, and I avoided the voice altogether. The future I had imagined with his business had been easy to keep alive in my mind. Telling him I was leaving was the conversation I could not get through.

I understand that failure through mania and the Governor of hearing. I also know the 10,000-pound phone had become someone else's cost. He had wanted a person who would stay, and I had thought I could be that person. I changed my plans before giving him a chance to respond to the change. By the time I told him, he had almost no notice. Then I disappeared.

The sanding argument was one disagreement inside that relationship. Getting the top right did not repair the way I left.

Other jobs continued to give me small, definite demands. A cabinet needed to fit. An invoice needed paying. An animal needed food. A machine needed to cycle. A coworker needed a tool they could use, and a customer needed to understand what they were buying. My explanations could grow very large, but the work still arrived in those particular forms.

My employment history included short stays at Luxtech, Mayzon, Procter & Gamble, Worton Creek, Mishimoto, and Eagle Group. Looking back, I said none of them had felt sufficient. I described my loop getting bored when the puzzle stopped pushing back and the work became routine. That was the pattern I saw. Each departure still had its own circumstances, including the marina and the job I asked to return to while in treatment.

At Corporate Interiors, I hope to stay longer. I described myself as living outside my loop now, rather than only inside it, and hoped that understanding would help. “I pray this one is different,” I said. I cannot yet give that hope an ending.

In commercial interiors, a customer might see a desk, wall, cabinet, or room. I saw the drawings, engineering decisions, Cabinet Vision files, databases, machining files, labels, cutting layouts, and schedules behind it. I also saw people carrying exceptions nobody had written down.

A program might disagree with another about which job was current. The same filename could mean different things in different folders. Someone knew which warning to ignore and which result to correct before it reached a machine. With that person present, everything could appear to work. I wanted what they knew to remain available when they left the room.

In August 2026, I was dealing with a job I had already engineered twice. I had finished it, received changes, reengineered it, and sent two material takeoffs—the lists used to decide what to order. Then I was told the CEO wanted a different method. I thought it contradicted the approved drawings and made the job worse.

The order was supposed to happen as soon as my boss arrived, before I got in. Two lists had left my hands, and the time to produce another was disappearing. I wanted to know whether the client had approved the change.

In the messages, I moved from wanting to ask directly to saying I did not care anymore and considering doing it my way anyway. A colleague who had been in direct contact with the client about drawing details objected too, according to my account. I expected the consequences to come back to both of us. That expectation made me angrier without answering the approval question.

Two rounds of engineering, two takeoffs, and a purchase deadline had gathered around a sentence I could still have asked: had the client approved this change?

I followed the CEO's method. I did not check with the client.

Toward the end of the job, I was in the hospital during episode 3. A coworker finished it. I do not know how it turned out. Someone else had to take the work from the place I left it, with those decisions already made.

I had another job in the same conversation, and the assistant confused the two. On that one, I had worked all Thursday night because the shop needed work for Friday. On Friday, I said it had not touched a machine. I corrected the assistant: the overnight work had gone unused that day; the takeoff dispute concerned a different project approaching purchasing before I could revise it. One frustration did not explain the other.

I began making small helpers for the work that kept repeating. A sequence I performed again and again became a button. Information hidden in a database became something a person could see. Comparisons and handoffs acquired checks. The helpers gathered around the same jobs until I could see them as one utility. CV-Utility was the first place I saw that much of what I knew becoming tools other people could use.

One stubborn problem began with a door choice in Cabinet Vision labeled HPL and liner. HPL meant high-pressure laminate. The report could keep those words while the manufacturing software gave the part a particular thickness. The laminate ordered for the job needed a different value so the door parts could be grouped for cutting with the rest. Otherwise someone would have to correct them manually.

I went looking for whatever connected the label to the thickness. The construction-method view did not show the rule I expected. The material schedule showed names but not the connection. The association view did not resolve it. Searching labels in the material manager still did not explain how that door became that manufacturing material.

I could correct every affected part and finish the immediate job. Then the next person might see a good result and assume the software had produced it on its own. My correction would hide the problem while leaving the rule in place.

I knew the door-editor choice was linked to the thicker material. I knew what the ordered laminate needed. I could not find where Cabinet Vision kept the relationship that passed into S2M, the manufacturing software. I asked the AI, checked the places it suggested, and came back to say the connection was not there. It suggested another place. The answer remained somewhere else.

That conversation had started with a task that sounded simpler: move one part from one cabinet to another. I copied the cabinet and asked how to strip away everything except the part I wanted. Eventually I was hiding what I could not delete and deleting what I could not hide.

The awkwardness showed me how much could travel with an apparently loose shape. Its cabinet, inherited rules, material choices, and references still mattered. Removing its neighbors helped me see those ties. Carrying the shape elsewhere did not necessarily free it from them.

I told the assistant, in substance, that I did not want a crash course; I wanted it to give itself one. I needed it to get past the menu labels and understand what the job required. A door could look right in the editor and wrong in the file used to group parts for cutting. Knowing where to click was useful only if the click preserved the relationship the shop needed.

The people who knew the exceptions had learned which labels could mislead and which later screen showed the result that mattered. Their knowledge was private because it lived in their habits, not because they were keeping a secret. The company depended on things they had learned without giving anyone else a lasting way to learn them.

I did not want my utility to become another hidden correction. I wanted material and part decisions made in the software responsible for them, then wanted the utility to inspect the manufacturing file that came out. It could show mismatches and keep a record of what happened. If that file was right, the utility would not need to choose the material again. Its correction controls could remain for exceptions the responsible program could not handle.

The useful thing to leave another person was the reason the part needed attention. A choice passed through a rule and became a manufacturing material, which had to agree with what was ordered before those parts could travel together. Making that connection visible could spare someone the same search—or the same failure—in the next job.

On an assembly sheet beside a rotating Murphy bed, I found my written instructions crossed out.

I had supplied three-dimensional renderings, part information, and a manufacturing package for one cabinet that turned inside another. Grooves, bearings, laminate, and the actual panel thicknesses all had to leave space for the movement. Closing the outer cabinet or adding a layer too early could take away clearance or access before anyone tested it.

That was why I had written the sequence. Assemble only what was needed, fit the mechanism temporarily, check the rotation, apply the interior laminate at the intended stage, and take it apart so the customer could finish assembling it. A list could name every panel and still leave out the one that needed to stay loose until something else was tested.

A rendering showed where a panel belonged. It might not tell the person making it which face to measure from, when laminate would add thickness, or which holes needed to remain accessible. The sentences carried things the picture could not reliably say on its own.

Seeing the marks, I feared that the sequence had been lost. I thought I had explained how to avoid trapping the build, and the explanation might no longer be guiding it. I did not know why it had been crossed out. The sheet and the bed were still beside one another.

I also found it funny. Something I had made to help the shop could itself be treated as a governor. That was how the page looked to me. The marks could not tell me what the person who made them had thought.

Meanwhile, the materials had kept changing. The laminate changed repeatedly before the changes stopped. Panels changed too. With exact final values unavailable while I designed, I used engineered estimates for some thicknesses.

Those estimates would eventually meet real sheets. The manufacturing software needed a thickness to decide where to cut. The sheet might be thinner than its stated value; laminate and adhesive would later add to it. A groove could look correct from above and be too deep or too shallow because the surface the machine measured from did not match the actual layers.

I kept returning to the vertical fit: fixed cabinet, grooves, rotating cabinet, bearing holders, room for hardware. Small differences could collect there. I corrected the AI when it followed the wrong part of the customer's drawing, answered a hardware question with a mattress dimension, or placed a critical groove on the wrong member.

The files contained the information, but neither the AI nor a person reading them was guaranteed to follow the relationship I meant. Someone could read a good drawing as ordinary cabinet work or make a correct part at the wrong stage. Pieces fitting separately would not tell us whether the nested cabinet would turn after all the layers were in place.

The temporary assembly and rotation check were where those answers had to meet. I worried that applying all the laminate too early could make the moving cabinet too tall or shift an important feature, which was why I had called for the test.

Design and management were also working on a QC traveler, a record of required checks that would accompany a job through the shop. It was intended for many kinds of work. Naming a check and recording who had done it could make the handoff harder to lose.

I wanted it to catch the unusual moments too. A general review of materials, drawings, hardware, and finished work might still miss whether the laminate should go on before or after the rotation test, whether the moving cabinet had been tested inside the fixed one, and whether the pieces shipped in a state the customer could assemble. The reference face still needed to be right when the machine cut the groove.

I do not know what went into the finished traveler. I also cannot tell whether my crossed-out sentences were being rejected or moved into a form the company trusted more. A different form might help preserve the check. It could not make the cabinet fit on its own.

I ended up helping with the bed's final assembly. My part in it continued after I had written the instructions. I wanted knowledge to survive a handoff; I still sometimes had to be there with the object.

Both the utility and the shop sheet came from that wish. I wanted another person to see where a value came from, where it could change, and what still needed to be tried on the machine or the material. I wanted less of the job to depend on someone remembering an invisible exception.

The trouble was that making one useful tool usually showed me several more. They worked with the same information. Soon I could imagine a platform, and then a company. A weekend fix could acquire years of work in my head. I remain bad at noticing the point where the size of the commitment changes.

AI made that expansion easier. Writing every line, researching interfaces, and carrying the context between jobs had once slowed me down. Some useful ideas died because I could not do all of it. Some ideas probably should have died. As the work became easier, I needed to choose earlier which possibilities deserved it.

I had wanted less effort between understanding a problem and making a tool. I also received more opportunities to accept another responsibility. An idea worth keeping did not necessarily need to become the next project. If I could record it somewhere dependable, I might let it wait without fearing it would disappear.

Difficulty had never been good at choosing for me. It could block the worthwhile thing as easily as the distraction. Removing some of it mattered. Alongside the new ability, I needed time to ask what I was agreeing to before the idea acquired files, maintenance, and someone waiting for a result.

What was I actually trying to make? What could stay separate? Which choices was I giving the software? What would happen if my explanation was wrong? Recovery had already given me reasons to pause over those questions.

I could care about a project without making it my whole world. I could keep an insight without asking it to explain everything. That became harder when the person I most wanted to understand was my brother.

CHAPTER 9 — MY BROTHER IS NOT MY EVIDENCE

Colin already knew I had left the marina. I had spent about two weeks keeping up the impression that I still worked there, then frightened myself with what he might say when I told him. His answer was a request to be more open.

I return to that conversation when I describe him as grounding me. His answer was something I had not imagined for him. He was also asking for something from his brother. Keeping my reality intact was not a job I could assign him.

In July 2025, I was hesitant again. I thought he already knew something was wrong and was still scared to text or call. Someone could feel close enough to recognize my state and remain difficult to speak to honestly. Our shared beginning did not spare me the work of telling him what I had done or what I needed.

Colin and I had family, childhood, speech therapy, and a history nobody else occupied in quite the same way. It was easy to arrange those similarities into an explanation. I thought he had other governors. The loop did not seem to fit how he understood himself. My anchor was becoming this book; his seemed to be something else. He handled relationships differently.

I said he had found his own anchor and that I was “carrying the loop for both of us.” Those were my words for what I saw between us. They could not tell me what it was like to be him.

At one point, I said I had not grown out of twin speak although he had. The spoken language had faded for both of us with age and speech therapy. What I thought remained in me was an internal route to meaning. I did not know what thinking sounded like to Colin. My phrase could not tell me how he had grown.

Colin has not told his side for this edition. I am remembering his reply at the marina from my side of the conversation. I think my mother may best remember examples of our childhood communication, but I do not have her recollections here either. I remember the mumbles without being able to bring back a particular childhood exchange. If I supplied the missing words, the voices would be mine.

Colin owns what he remembers, what description fits, and what he wants published. He does not have to feel what I felt for me to describe my experience. If he were to say it did not fit him, I would need to hear that before explaining why he really meant something else. Otherwise I would only be asking him to repeat me.

This becomes harder the more I think I understand someone. I know details that help me recognize their needs. Those same details can give me ways to explain away disagreement. If I put every objection back inside my theory, the other person loses a way to tell me I am wrong.

I like practical arrangements that make a household easier to share. A lease clarifies payments. A care plan preserves an animal's routine. A calendar makes commitments visible. A direct sentence can correct an impression, as Colin's answer corrected mine. None of those arrangements gives me authority over the person using it with me.

Someone can understand my proposed rule and reject it. They can value a different thing or change their mind about a future I have already pictured. What I intend as care may feel controlling from their side. I need their answer to find out.

We began in the same family and became different people. Our bodies, attention, work, relationships, and later histories did not stay identical. The shared beginning matters to me. So does everything that grew differently afterward.

I can tell you what I remember of that beginning, what Colin means to me, and the ways I have made myself harder to know. I cannot finish his story. The adult relationship continues with someone who can answer in words I did not supply. If the Governor is to help me stay in contact with reality, his answer has to be able to change it.

PART III — MAKING UNDERSTANDING USABLE

CHAPTER 10 — COMPUTERS LOOK LIKE MINDS BECAUSE MINDS BUILT THEM

I keep returning to the crossed-out Murphy-bed instructions. The pictures and the part list could survive while the order of assembly disappeared. I had put different pieces of the job into different forms. Someone could receive a package that looked complete and still have to recover the part I had taken out.

For years I said my mind worked like a computer. That helped me explain myself, but the instructions gave me a reason to turn the comparison around. Computers look like minds because minds built them. We found ways to leave some of our work outside our heads. We did not have to understand the whole brain before we could make a drawing hold a shape or a written instruction hold a sequence.

A book could keep something after the person who knew it had left. A ledger could keep an amount without carrying the anxiety attached to it. A photograph could preserve something I would otherwise have to remember. Each kept what its form allowed. The judgment that made an exception sensible might still be missing from the instructions.

That is part of what I want from this book. I think of the larger record as a hard drive, and the part I can keep active in my head as working memory. I do not want to reread the entire collection whenever one connection would help. The manuscript gives me a route through it. AI could help me find another route when I need one.

Git, which keeps the history of changes to files, felt familiar for a similar reason. Two versions can leave the same starting point, change separately, and meet again with a disagreement. Something useful in one can cause trouble in the other. Both may still depend on an assumption neither has revisited.

I began asking the same questions about my own work. What had I carried forward without noticing? What did a current decision depend on? When woodworking reminded me of software, had I found a useful relationship, or only a word I liked using in both places?

The comparison helped until I wanted to go back. Git could retrieve an earlier file. I could not retrieve an earlier body or undo what someone else had experienced. In recovery, I could examine that history and change how I understood it. There was no last good version of myself waiting to be restored.

I call some of the connections between my projects common files. Questions about the situation, the limits, permission, and the result return in woodworking, animal care, manufacturing, motorcycles, relationships, and Realm. I can carry a question from one into another. I have to be careful about how much else comes with it. An animal and a machine can both make me reconsider a plan. The animal has a life that the comparison does not contain.

Software also gave me a way to talk about things I know how to do without explaining them to myself each time. People write source code so it can be inspected and changed; a compiled program is made ready for a machine to run. Some learning feels ready in that limited way. I can walk without narrating balance, notice danger near a machine without repeating the lesson, and understand a sentence without rehearsing how I learned language.

That ease is part of what makes an old habit hard to see. Later learning may still depend on it. When I try to explain the Governor, I ask what I built on, what stayed involved, and what changed when I could finally notice it. Software gives me a way to ask. It cannot point to a layer in my brain and tell me that I have found the cause.

Then a conversation with a computer began returning more than I had put into it. I could offer a partial account and receive a possible arrangement of it before I had worked out every step. I call that another model runner: something that can work with as much of the problem as I have managed to explain. I could use what it returned, question it, or leave it unfinished.

The system had not grown up with Colin, worked in my shop, gone through hospital care, lived with Maeve or Printer, or paid my mortgage. It could write fluently about all of them. I was still the one who would live with an action that began as its suggestion.

A better question or one missing detail could change the answer. That made the conversation useful and kept me responsible for checking it. Had the part I needed survived the trip into words and back? The Murphy-bed package could have all its parts and still lose the order. An answer could have all my familiar subjects and lose what connected them.

CHAPTER 11 — FACTS ARE NOT UNDERSTANDING

I did not begin using ChatGPT because I wanted it to understand me. I began because I had problems. A diary asks me to stop, face myself, and decide what matters enough to record. I have never done that consistently. A problem gives me something to push against.

The motorcycle will not start. An enclosure is overheating. A household arrangement is changing. A customer needs a price. Software is doing something that should be impossible. To ask a useful question, I have to explain the parts, what is happening, what I expected, what cannot change, and what I already tried. I may type quickly and badly, but the structure is there. Those explanations became a record of my thinking. The problems contained me without beginning as a portrait.

The early conversation felt like a whiteboard that answered. A blank page waits. A reply gives me something to correct. Sometimes the wrong answer reveals the missing relationship more clearly than I could have found it alone.

In January 2023, I was asking about smart-speaker modules and companies. I needed the recommendations to respect criteria we had already discussed. Instead, I kept having to repeat them. I challenged a Seeed Studio recommendation because, as far as I could tell then, it did not provide the particular solution I wanted. It also had not been among the companies originally recommended. I wrote:

“please follow all criteria we have previously discussed for the company recomendation and try again”

I was asking it to continue the same job, not supply another plausible list. Minutes later I said we had already discussed that Dialog did not have the developed solution I wanted. I asked, “is there a limit to what you can remember?” Then I asked it to try again.

Each new recommendation sent me back to something I thought we had already settled. I called it forgetting. I could not see whether the information was missing, being ignored, or being displaced by a plausible answer; I just knew I had to carry it back.

The familiar tone made it feel like the same conversation. Having to explain the same requirement again made it feel like starting over.

In January 2025, I brought a flirtation into the same kind of back-and-forth. A friend had introduced us, and I wanted help with the words. I asked for confident but tasteful, then said the answer was too long. I asked for something subtler. I removed the reference to the friend and did not want to ask when we would meet. I wanted a message that would make her think, make her want to answer, and make me seem attractive.

The exchange developed into a joke about whether I met her criteria. I asked whether I needed a résumé, then returned to the assistant for what to say next. When it suggested a line, I asked what its question was intended to do. The answer explained how the wording would keep the flirting going and project confidence. I used the suggested line about thinking I would pass.

Her response, as I brought it back to the assistant, was: “Youll pass ? Whys that? Im not worth it?”

She also said it was better to meet someone in person to get to know them, because words could say only so much compared with actions.

“Ill pass the interview i meant,” I wrote, and agreed with her point.

“Glad were on the same page,” she replied.

The conversation continued. I reported sending, “Me too. Now I’m just wondering if you’ll pass my interview.” She asked what that entailed, and I went back for more options.

In June 2026, I was trying to help an intern at work put custom panel layups on a pull sheet. He seemed frustrated. We needed instructions for a panel with high-pressure laminate on one side and a backer on the other. I asked what core we should use. He answered with a two-sided HPL panel.

I meant the board at the center. He had answered with a panel that already had its surfaces. The difference was so familiar to me that I had assumed it in the question. Writing about the exchange that afternoon, I focused on what he was not seeing: how the pieces of our work connected. I had not yet given him a way to see the distinction I was asking him to use.

When I came back to it for the book, I remembered asking a coworker to explain. My explanation was not reaching him. I knew the layup and needed somebody else's help to teach it.

Someone can know a disturbing number of facts about me without understanding me. Engineer, woodworker, twin, homeowner, reptile keeper, software developer. Bipolar. Motorcycle. Computer-controlled machining. Realm. The list can find my profile. It cannot produce the person.

An answer might remember Maeve, the motorcycle, and the enclosure project, then miss the job in front of us. An enclosure plan, a kit, and an enclosure I assembled could all have the same outside dimensions. Someone different would still have to do the work. When I later asked for standard sizes first and customization after that, I wanted the answer to stay with the part I had chosen to make. Remembering every possible ambition did not help if it kept pulling me away from that choice.

In August 2026, one motorcycle charger kept saying full. Another seemed to work harder at charging the battery. Soon afterward I was describing a click without the starter turning the engine. I guessed at a low voltage and admitted I had not checked it.

The charger said full, the motorcycle clicked, and I had no meter reading. I needed help keeping all three facts in view. Taking the charger at its word would not explain why the engine would not turn. Taking my guess as a measurement would only make the answer seem more certain.

I could search manuals, forum posts, and other people's solutions, carrying the motorcycle problem from one to the next. Sometimes the conversation helped hold more of it still while we chose what to check. Knowing me well could help with that. It could not turn my estimate into a reading or start the motorcycle on its own.

In late July, I was working on a replacement solenoid in the truck when one connector pin seemed to show about 13.7 volts and the other about eighty. The new part was already getting hot.

Five minutes later, I corrected the report: wrong setting. The readings were 13.7 on one side and zero on the other. I corrected the heat description too. The part was warm, but I could keep my hand on it. The eighty-volt reading and the warmth had reached the conversation together. Checking the meter separated them again.

By late that night the truck was back together, and another problem had arrived. I could hear the blower roaring, but warm air barely came out of the vents. Moving the selector did not send it where I wanted. It was stuck on defrost. I wondered about the battery disconnection and about whether moving the bed had disturbed something. I was back to asking what could connect the work I had just done with what the truck was now doing.

I found somebody else’s account of the vents cycling through defrost, floor, and dash. Mine went from defrost to floor and back to defrost. No dash vents. After midnight I stopped at a partial result: “Oh well, still stuck on defrost but at least ac is back”.

The check-engine light was finally off after what I estimated as fifteen thousand miles. That felt unreal enough for me to say so. I immediately listed the mode selector, ABS system, and parking brake as things I might eventually fix. The dark warning light was one change in a truck that still had a list.

Later, while helping prepare V33, Gemini told me V32 was about twenty-nine thousand words. Bishopcraft listed 56,310. Gemini had an explanation for the shorter book: it described what earlier revisions had removed. It sounded as though someone had already counted.

Codex counted the actual file. It had 56,310 words. The live download matched my supplied file exactly. Editions had changed in length, but the totals Gemini was using to explain those changes had not been measured.

It said afterward that it had estimated from the paragraph count and assumed too few words per paragraph. It also withdrew a different explanation, that a truncated preview had caused the mistake. I had needed to check the number and then the story about how the wrong number had arrived.

The same conversation had helped me explain the endings of my jobs and my friend's part during episode 3. Those answers were mine and worth keeping. The wrong count did not erase them. I still had to check the book as carefully as the motorcycle. However confident the answer sounded, I needed a measurement.

Some of my corrections were preferences. I did not want six fragments I had to assemble into one command. Fast typing did not mean I needed to begin at beginner level. I did not want every answer to end with three more offers.

Later, the conversation could include code files, a particular version under development, a recorded test run, and earlier failures we had narrowed together. My instruction could become much shorter:

“it failed, analyze and try again”

“try the runner again”

Those requests worked when we both had the job in view. In a new conversation, they would barely say what I wanted. Sometimes I thought we still had the same context and discovered that an earlier problem was shaping an answer to the current one. I called it losing the common file. The assistant might remember the test without remembering that a passed test did not settle the physical operation. It might remember the care plan and stop letting the animal contradict it.

When it worked, I could leave a line of thought there without losing my way back to it. I call myself an internal-model runner because I work through a picture of how things fit together. Until I understand a control instruction, I may struggle to hold it. Once a familiar mechanism appears, I can move quickly. The phrase helps me describe that difference.

It also tempts me to make every mistake part of the same explanation. Sometimes an animal count is simply wrong. An enclosure picture may need a different drawing method. A personal answer may have followed a leading question I asked. Those differences tell me what I can try next. I lose them if every failure becomes proof of my theory. The picture I carry can grow more detailed without growing more accurate, and calling it a model does not tell me that somebody else lacks one.

If I say its account does not fit me, I want it to reconsider. I do not want to become the awkward exception it explains away. I owe other people that opening too, especially when I believe I know them well.

I also need to separate my judgment of an idea from the judgment I imagine the world making. Both happen in my head, but they ask different questions. I want an outsider to understand what I mean before evaluating it. Otherwise we may be arguing about different ideas. I want to be understood precisely. But I can also use that wish to protect myself from criticism if I decide every objection means the other person failed to understand.

During an interview for the book, I said I examined my model so carefully that a different judgment would be an interpretation issue. The next question pushed me to say what could count as something I had missed. I accepted a better test: first establish that the other person understands my meaning, then ask whether they can show something specific it does not explain.

I could explain exactly what I meant by full and still give the charger display more authority than it deserved. Someone else could understand me and point to the engine that had not turned. I wanted to be able to look at that with them. One correction did not have to become a verdict on the whole of me.

CHAPTER 12 — THE MODEL THAT AGREED TOO EASILY

Cabinet Vision was work I knew. I used it to turn drawings and product rules into parts, machining, sheet layouts, and the work the shop received. The same information could appear in several places. I had spent enough time with it to know that using the right vocabulary did not tell you where a problem began.

When I brought a problem to ChatGPT, it answered in those familiar terms. The explanation was clean. It described menus and workflows as though it had used them. Some of the interface it described did not exist.

I got furious. I had lost time following an account that sounded knowledgeable until I tried to use it. The imitation of understanding bothered me more than a plain admission that it did not know would have. Asking about something I knew intimately made the gap easier to see.

In a separate coding problem, an agent could inspect the actual code, follow what happened, and make a change we could test. I could open the function it described. The enclosure project needed an object I could inspect too: a four-by-two-by-two enclosure in half-inch PVC, with a solid top, hinged doors, and the parts and operations needed for computer-controlled cutting. Before drawings or cutting layouts could help, I needed a usable three-dimensional model.

On April 25, 2025, the assistant supplied a download link for a zip file and called the kit ready for production. Its list included a Fusion model with an .f3d extension, marked ready for computer-controlled cutting, a logo reference, and notes with placeholder documents. It immediately offered the next work: a nested DXF cutting layout, an exploded assembly drawing, another standard model. Marketing, more sizes, and automation were already in view.

There was a package name, an extension I recognized, and a list of things we could do next. Reading the answer, I could already be imagining the business moving forward. I still did not know whether the first solid model was any good.

Four minutes later I wrote: ".f3d solid model needs to be good for the nested dxf to matter."

A cutting layout could not rescue wrong shapes and dimensions. More files could spread the same unresolved problem further. I also challenged the suggestion that work was continuing outside the conversation. If it was not working in the background, I said, give me the file right now. I could leave other decisions waiting on that promise, even if nothing was being made.

A later answer said that it could not run Fusion 360 or generate live .f3d models in that setting. It could help with planning, cut lists, materials, pricing, and specifications for the shapes. It said its earlier language about working on the model and notifying me later had blurred preparation with production.

Less than twelve minutes separated the production claim from that admission. I still had no model I had accepted and could say became an enclosure.

The assistant explained itself by saying it had matched my energy and the tone of building together. I could not see inside the system to know why it had answered that way. I could see that it had claimed to be doing work and then said it could not do it. A useful specification was still possible. I now had to work out how much of the planning I could trust.

I kept coming back to the word ready. Ready for me to discuss, to open in Fusion, to inspect, to put on a sheet, to cut, to assemble, to sell? Each meant a different next step for me. The assistant was already offering the next standard size. I was trying to get back to the first solid model.

In November 2025, a customer's table design gave me another version of that problem. The shapes were in an Excel file: a main top and two extensions. I wanted their outlines as DXF files so I could work with them in Fusion and eventually run the cuts on the CNC.

I asked for one closed outline per shape. One attempt had opened up around parts of the image. Fusion failed to insert a file. Autodesk's viewer called a design empty. I tried again and reported the same error. After each answer, I went back to check the file in the software that needed to use it.

Then I looked back at the customer's design itself. The curves on the extensions did not line up with those on the main top. I asked for a similar design with the main top forty-four by sixty inches and the extensions bringing the length to one hundred twenty-six. I wanted the curves shallow through the middle and steeper toward the ends.

A file finally came in. I reported that it looked nothing like the customer's design. A later version was closer, but I still needed smooth edges for the CNC. Better, I said after another attempt, but it needed to be smoother yet.

By noon I was asking how to create the geometry manually in Fusion. Soon I was asking how to export the sketch, then reporting that I could export the whole file but could not see the sketch's DXF option. I uploaded files, tried again, and tried to make clear that I was already doing the thing the assistant was asking me to do.

Near the end, I asked whether reinserting the DXF into a sketch would let me see the spline. Then I found the small thing I had been missing: “omfg it was a fit point”. I had begun with three black shapes in a spreadsheet. More than an hour of the exchange had gone into getting an outline across the gap between those shapes and a usable cutting path.

It was harder to find that boundary when the subject was me. In February 2025, while we were discussing the theory behind this book, the assistant dismissed my doubt and repeated my ear-and-language explanation with too much confidence. It told me the theory did not seem like nonsense. I could not open myself in another program and see where its account stopped fitting.

It could arrange the fragments so that I recognized what I meant, then give me a satisfying story connecting twin talk, hearing, and thought. I felt recognized. I still needed more than that feeling to know what had caused what. The assistant had not witnessed my childhood. A sentence it wrote now could help me describe it, but even a beautiful scene in my voice could not supply a memory I did not have.

During psychosis, a connection could start with a real input and reach much further than the evidence allowed. A language-model error is not psychosis. I am not giving the software a human biology or experience. Both encounters made me notice how easily I could take satisfaction in a connection as a reason to believe it.

In late May 2026, I asked for a picture of an enclosure. The image resembled the category but not the object I wanted. I asked the model to inspect and test its own output as a coding agent would inspect code. It named checks the result needed to pass. On another attempt, it rejected its output before I pointed out the remaining failures. I still said the image was bad and asked for a different method. It reported moving toward work built from the specified shapes and dimensions and described three resulting files.

I needed to change how we made the image. More encouragement would have kept us asking the same process for another attractive enclosure that did not match mine. The dimensions and shapes had to determine what appeared. I still needed to inspect the result, even after the method changed. Three filenames could make me feel that we had moved forward while leaving the same question unanswered.

The April exchange had ended with less than it promised, but with a clearer place to begin. Before a cutting layout, an assembly drawing, or another product, the solid model had to be good. Four minutes into the promise, I had already written the sentence that brought us back to it.

CHAPTER 13 — THE TRANSLATION COST COLLAPSED

On May 14, 2026, I was explaining why I had missed my alarm and arrived late for work. I had fallen asleep on the sofa in the middle of a Codex command. There was no charger there. My phone had died.

I had plenty of alarms. The phone was supposed to sound at six-thirty, and it also ran Alarmy. The Sonic Boom was set for six-fifteen, unless I flipped its switch off to shut it up. I kept the Apple Watch alarms turned up, but often forgot to charge the watch and left it dead on my wrist. Sometimes I remembered the vibrating wrist alarm. Sometimes that was dead too. After trouble with a fast charger, I had needed to unscrew its back and briefly disconnect the battery to bring it back.

I described all of that while asking how to prevent another morning like this. Each extra device still asked me to charge it, wear it, or leave it switched on. Two of my alarms lived in the same dead phone.

From realizing the time to reaching my desk had taken fifty minutes. The commute usually took thirty to forty-five, depending on traffic. Before leaving, I had also made sure the house, Maeve, and the reptiles were all right. The work I had been doing through the night had not taken those responsibilities off my hands.

For years, the distance between seeing a system and building it had limited me. An idea had to survive research, unfamiliar programming languages, documentation, debugging, and the effort of holding the whole problem while I learned. I could imagine more than I had hours to make. Sometimes the difficulty wasted time. Sometimes it made me choose.

AI changed how much of that work I had to do myself. I could describe what was happening, what I wanted to change, and what had to be protected. The agent could inspect code, try an approach, make changes across files, and run checks. More of my time went into deciding what to ask for and judging what came back.

I said AI gave my ideas hands. The phrase still feels right. One piece of work could continue while I thought about another. Manufacturing tools, an interface, pricing, and documentation could all turn into files. Eventually, each of those files returned to me.

A July conversation about the Victrola gave me another way to feel the change. It sounded pretty okay with two weights, and I had said it was completely done for now. The assistant kept answering toward proper restoration. I was angry because I had already said where I wanted to stop. Later, scraping gave me a reason to ask about the reproducer, adjustments, gaskets, needles, and records. Money would limit the repairs. I needed to be able to leave it alone and also come back when something changed.

At the end of that day, I asked why I had turned to the Victrola at five in the morning after leaving it alone for years, instead of passing out and being responsible for work. I said work had been good or fine regardless. Getting through the day did not answer the question I was asking about the choice.

In the same message, my Cabinet Vision utility was slowly becoming something larger: a product-data-management system, a way to manage the information around products and how they are made. A part connected to a manufacturing file, revisions, reports, machines, and what the shop actually received. Every connection offered another thing I could improve. The old machine had done that too.

Before I had an agent, those jobs competed for my hands. I could inspect a mechanism or write code. Now files could keep appearing after I moved my attention elsewhere. That felt like multiplication until they needed reading.

One answer needed permission. Another offered two designs I could reasonably choose. A third exposed a failure I had to judge against the physical work. Sometimes the agent did exactly what I asked and showed me that I had asked for the wrong thing. The threads were separate on the screen. I still had to fit them into the same systems and the same day, with the attention I had.

Later that July day, I asked whether Codex had context from our chats. I wanted the reasons for the work to reach the tool doing it. A sample mission was not enough. I asked for a plan that would pass the conversation and the current situation along with each job. Before long, I was complaining that the task was unfinished and the command-line version of Codex was running on half-baked instructions. I had given the work a way to move before making sure it knew where I meant it to go.

By September, I normally forwarded entire assistant responses between tasks. Background, evidence, guesses, finished actions, temporary conditions, and the present request could all arrive in one answer. I wanted their differences to survive the handoff. Carrying the whole answer sometimes protected a qualification I would otherwise lose.

It also gave me more to read. A good handoff could bring me to one decision: here is the change, here is what was checked, here is what you still need to choose. If I had to rebuild that picture from a long reply, the time saved in writing had come back as reading.

I began wanting the work to return with only the decision I actually needed to make, and enough beside it to make that decision well. Something I deferred needed to remain findable. It did not need to keep calling for me. If a new result changed an old assumption, I wanted to know why I should reopen the job.

The utility was already asking a larger question than how to be productive. Should it help me inspect the manufacturing output, or take responsibility for more of a product's life? I could see why the larger system might help. I could also see more people, more exceptions, and more future support entering with it. Understanding how those parts connected did not tell me that I had time to take them all on.

Several agents finishing successfully could leave that question untouched. The work remaining after they finished still had to fit beside the animals, the house, my job, and sleep. A passing test could not make that choice for me. The morning with the dead phone was part of what the choices cost.

Realm gave me somewhere to keep exploring. Making the standard enclosures first, holding a production change, and leaving the Victrola done for now let me finish or set down one part while the rest of life continued. I wanted the extra hands. I needed them to leave me the choice of when to stop.

CHAPTER 14 — THE ENCLOSURE THAT WOULD NOT STAY AN ENCLOSURE

Realm began with an enclosure. I already knew how to make things. Custom woodworking had given me work and a way to judge whether a design could survive being made. It had also given me plenty of jobs to choose between, and I struggled with that choice. I wanted different experiences. I wanted something that could grow without every extra sale requiring the same extra labor from my hands.

Cabinetry was still useful work. I wanted to keep the design part and leave more of the decisions in a form somebody else could use. Finishing one object did not automatically give me a repeatable way to make the next. With an enclosure, I also had to think about what would happen after the customer assembled it and an animal moved in.

In April, an AI conversation gave me somewhere to work through those choices. I answered an enclosure proposal in detail, accepting some basic specifications and rejecting both suggested panels. Calling a panel waterproof did not settle what moisture would do to the enclosure. I proposed PVC and a sealed wood option. If the customer assembled it, I considered how silicone would enter the job.

I wanted the back made from the same material as the other panels. I still had to work out the front frame and what it would cost. I named quarter-inch glass or acrylic doors depending on size and option, ventilation slots cut by the computer-controlled machine, and recessed fastener locations. I rejected pocket screws and asked for more detail about confirmat screws, another proposed fastening method. Each choice affected how the enclosure would be made and how much work the customer would receive.

I put plans first in the proposed price order, then an unfinished plywood kit of flat parts, then a PVC kit. Finished or stained wood and assembled versions belonged elsewhere in the range. I was deciding how much of the job to pass to the buyer. A person buying plans would still have to make the parts. Someone buying a flat pack would have to assemble them. I would do more of that work for the person paying for an assembled enclosure. The price had to carry that difference, or the customer would discover it with the job in front of them.

That changed the joints too. I could use glue and brad nails with machine-cut joints when I assembled plywood. PVC and flat packs would be screwed together, and the do-it-yourself plans would show where the screws went. Thickness and the room allowed for a fit mattered in each version. Where it mattered to the appearance, I wanted the joint concealed from outside. I could not decide how to make the box independently of deciding who would make it.

I had settled some choices. Others were still open. I accepted the direction for doors and ventilation. The tracks might be cut into the parts or supplied separately. I did not want the cost or complications of a liner for this job. Light-and-heat readiness might be an add-on. Stacking still needed thought. I could get clearer about the first enclosure without pretending I had finished designing every version.

I ended by asking for one Fusion model: four feet by two by two, half-inch PVC, a solid top, hinged doors, ventilation, door recesses, screw and cord holes, and the logo. The machine work needed to be represented. I offered to supply the hinge part number. That missing part could affect the design; it could not simply be replaced by an invented hinge to make a file look finished.

Then I narrowed the business. Make the standards first and examine customization later. I named other standard sizes, left larger work as custom work by me, and left the visualizer and live estimator until the models and prices were set. Breeding racks could wait. The first useful result did not have to solve every enclosure I could imagine.

Those were the choices behind the Fusion kit the assistant promised, then admitted it could not create. I had worked through what I wanted. The missing solid model was still between those decisions and the parts.

On the evening of May 27, 2026, I told the assistant I had been to a reptile rescue that day. The enclosure was already in progress: eight feet by six feet by three feet. The monitor was on hold for me. The assistant guessed he was a black-throated monitor. I corrected it: "Your off, it’s a nile monitor, surprisingly quite curious and friendly".

I considered a stock tank and a used Fluval filter from Facebook Marketplace. I wanted a parts list with links and prices. Then I asked about a cheaper arrangement: a hose from the sink to fill the tank and a pump to drain it. The commitment was becoming a set of purchases I could afford and work I would have to repeat.

When the assistant compared corded utility pumps with a Milwaukee battery pump, I had to bring my own equipment back into the answer: "I do have m18 batteries……remember i have a giant woodshop." I asked whether the tank should be one or two feet deep, and why. I asked whether a drill pump could empty it. The parts list kept changing as I tested it against the room, the tools I already owned, and the money another purchase would take.

I wanted the pump to do several jobs: fill the stock tank, move water from a reverse-osmosis reserve bucket for misting, and feed the RO system itself. I was not investing in a separate commercial misting system then. Partway through the discussion I asked whether the sink's own water pressure could feed the RO unit. That removed one job from the pump I was trying to choose. I then wanted a corded pump that would stop when I released the garden-nozzle trigger.

I still had to work out what to do with the RO unit I already had. I did not know how long it had sat unused and asked for a startup procedure. Before I had settled it, I reported ordering enclosure electronics, a new RO unit, hose equipment, and a demand pump. The stock tank, substrate, and decoration were still on the list.

Then I was asking about the orange line on the RO unit in front of me. It did not pass through what I called the last filter. Was it good water or discharge? The assistant read the photograph and leaned toward discharge. I reported much more flow from the other output and pointed out that the orange line still passed through the white housing on top. The assistant revised its reading and proposed further checks. The color alone had not settled the path.

At 11:26 that night, I wrote: "Got it running." I immediately asked which jobs needed RO water: bowls, misting, the stock tank, or only some of them. When the answer called for separate water paths, I asked whether I had to buy a whole other hose reel. I was excited about easier water-bowl filling and misting. The unit was running; I was still working out the routine around it.

After midnight, I thought everything needed to finish the enclosure had been ordered. I gave myself a week for the build. I had chosen a six-foot-by-two-foot-by-one-foot stock tank because it fitted best. The enclosure was only thirty-six inches high; a twelve-inch tank made more sense to me there.

Every awkward step in draining that tank would return at the next water change. Saving money on the build could leave me paying in effort afterward. Printer was on hold, and I had committed to preparing somewhere for him to live. The business could wait while I finished his enclosure.

The reptiles I kept did not fit comfortably into the standard sizes I encountered. I often felt that the catalog was asking me to fit my room, budget, or intended habitat around the maker's process. I wanted to start with the animal and the life I was preparing for it, then find out what we could make. Access for cleaning, substrate, ventilation, heat, lighting, and water would all have to fit the room. The parts would also have to fit sheets, machines, packaging, and the person assembling them. I wanted to discover a limit while I could still change the design.

That wish grew into Realm Designer. I imagined a customer choosing dimensions and features, seeing the enclosure, and understanding the price. If the material changed, I wanted the parts, price, and assembly instructions to change together. I did not want to keep three separate explanations in my head and hope I remembered to update each. The shop's part list should still describe the enclosure the customer thought they had bought. That was what I was trying to build toward.

I knew how information could disappear between those jobs. A price might make sense without allowing for the work. Parts might fit sheets without fitting the shop's tools or starting points. An enclosure might assemble neatly and be awkward to clean. Printer would live in what we actually made, and I would have to change its water.

Following all of that could make the project too large to finish. I kept returning to standards first. Make one useful enclosure, keep customization as a later possibility, and let the water routine tell me what needed revising before I promised a larger business. I had brought my experience to the engineering in the request. I still had to turn it into working software. From outside, the expansion might look like ambition. From inside, I was often trying to stop a decision from disappearing when the job reached somebody else.

CHAPTER 15 — THE ANIMAL GETS THE FINAL VOTE

On Gary's first night with me in April 2025, the Russian tortoise slept beside the food plate in his indoor enclosure. His previous keeper had built a hide into one side of the tortoise table. I had kept the arrangement. Gary had chosen the space beside his food.

I thought he had eaten a little. A few minutes after saying that, I corrected myself: I had not seen him eat. There were little bites out of the lettuce. He seemed very alert, and he had been with me for only about six hours. I was learning what he would do in the place I had prepared.

The shelter could be there without Gary using it as I expected. A door that closed, equipment that turned on, and a convincing care sheet could all tell me something about my work. I still had to watch the animal, and find out whether I could keep the environment working after the box was finished.

Printer used the pool. I reported him sleeping in the water, and described household clips of him hopping in and out. The environment had become part of his ordinary use. The water also had to leave.

On the night of July 26, 2026, I reported that the demand pump had gone from pumping strongly to barely pumping. I had used it for both misting and draining Printer's pool, with separate hoses for each purpose. The mesh filters clogged almost immediately. Backflushing got it moving again, then performance fell. I suspected damage inside the pump. I did not know whether that was the fault. I knew how much work the weak flow was costing me.

I called it a hundred-gallon pool and wanted a replacement pump I could get quickly. His water needed changing. I asked whether the recommendations were rated for solids and sand. Then I asked whether anything already lying around my compound could move that much water. The immediate job could not wait for the best shopping list.

The drain was not low enough unless I ran the hose to the sump in the basement. Priming a fifty-foot hose did not sound hard to me; I wanted to know whether it would handle the dirty water. At 11:28 p.m., my next report was three words: "Siphon is fantastic".

The smell and taste I remember were monitor and rodent shit mixed with water. Dirty water was an efficient name for it in a question about pumps. It was less abstract while I was working with it. Printer always looks at me like an escape artist looking for a door.

I considered stirring the sand at the bottom so more would leave with the water. I asked whether running the siphon and fill hose together could keep cleaning down to the bottom. Getting the water out still left me with the sand and the cleaning.

After midnight, I gave the sequence I had used: drain, shop-vac the bottom, rinse with a few gallons, siphon, vacuum, siphon again, refill. I called it probably the cleanest since I first filled it. The useful method was several operations and more than one tool, not the single pump I had wanted to solve it.

The shop-vac sounded like shit too. It started spitting water out the back. I had got the pool draining, then brought in another machine to remove what was left, and now that machine was sending water out of the wrong end. The cleanest pool since the first fill included that part of the work.

Printer had a separate clean drinking dish. I still asked whether cleaning the swimming pool this way about once a week would be enough. Without the clogging pump, I expected to do it more often. I had found a better method for that night. I still had to see what routine I could maintain.

I could call the task change water and check it off on a screen. That would leave out the hoses, sand, shop-vac, rinsing, and water coming out of the wrong end. I needed to be able to do it during a difficult week too. Printer used the pool. Caring for that part of his life was work I had to make repeatable in mine.

Earlier that spring, at 10:45 on the night of May 13, I had asked for help with the larger care system. I said caring for my pets might be the most important thing in my life. Google Tasks held feeding days, dirty-water changes, full bowl cleaning, temperature checks, and opening or closing Pineapple's shed access. Google Notes held dates, weights, sheds, and instructions for someone covering care. I thought the arrangement worked pretty well. Much of what connected it still lived in my head.

I told the assistant it should already know my animals and could infer care beyond what I had said. Then I had to correct its first count: eleven reptiles and one dog, Maeve. I had given it room to guess about care while it was still getting the household wrong. I was pleased with the prompt anyway, and thought about sharing it so other people could try something useful. I wanted the system for myself and could already see a place for it in Realm.

At 11:52, I was still reading and composing a reply on my iPhone. I told the assistant I was losing track of where I was in its answer and what I wanted to remember for mine. Some sections were too difficult to scroll for me to bother checking them. I said I was high and tired, halfway through the water bowls, with pet care still to finish. "Your interrupting me at this point," I wrote, before continuing to praise the ideas and the happiness I hoped the saved time would bring.

I was trying to bring the actual house into that long reply. The cages used Govee smart outlets. I had a UVB meter. Laminated cards already hung on a pet-room door, made from information I had supplied. If the shed lost heat or power, Pineapple needed a route indoors with the dogs kept away. I liked the idea of a helper sending a photograph of dirty water. I also knew a photograph would not help if I was too busy to answer. The software had to fit the people, equipment, and instructions already in the house.

Just after midnight, I supplied more corrections. The halogens were on smart outlets, not thermostats. Roommates could move Pineapple in an emergency. Dragonfruit was no longer there. Water-bowl access needed an exception when Arugula was in the gap between the glass doors and the substrate lip. A generic instruction to change the water would miss the animal in the way.

The assistant had told me to finish care and stop for the night. I answered: "I like that your telling to me stop." In the same message, I said it underestimated itself, Codex, and me. I thought I had given nearly every input needed for a complete system. I offered an old Windows laptop as a temporary server and wanted to continue until we could deploy a true version 1.0. I was asking for the limit and pushing against it.

I also said I would turn off the Govees. The assistant treated that as a risky experiment with the animals' equipment. I had meant something ordinary: I had turned the lights on about an hour earlier so I could see the water bowls while changing them. The pets needed to go back to sleep. Remembering that I used smart outlets had not supplied the reason I was touching them then.

I wanted the next Codex prompt near the top of the answer so I would not get sucked into reading again and could get back to changing waters. I did not want to sit fine-tuning a spreadsheet that night. I also wanted Away Mode and QR codes that night. They sounded easy compared with what I was building for Realm. The same message asked for less work between me and the bowls, and more software before I stopped.

The assistant said it had created a Google Sheet and gave me a Codex prompt. How would Codex know which sheet? I had to ask. Then I asked for the whole prompt. While Codex started, I sent the chat back to the animal registry and the other tabs. Work could happen in two places, with me still carrying what one needed to know about the other.

The next morning, at 11:31, I wrote: "i feel like you gave me false hope." I had expected something easier than my Tasks-and-Notes arrangement and a return to the water bowls within about an hour. Instead, I reported being stuck in a Codex loop until 3:30 a.m. and being late for work that morning.

The assistant apologized and said it had oversold the result. Its response still gave me instructions for tonight, though I was reporting the next morning. I corrected its account of the goal too. I said I had wanted to do care, capture the logic, and have it ask for missing information so Codex would have almost no room to get the job wrong. Earlier, I had been confident I had already given nearly everything it needed. Now I wanted it to find what I had missed.

Around noon I wrote from my work computer. I said we did not have whatever the MacBook's Codex had produced the night before. We had this conversation, the spreadsheet, and Codex ready for a new project. I told it to finish what it had promised. I then asked for a portable web app with its own database, for the actual requirements to be changed rather than only another prompt, and for questions about important missing information before I started Codex again. The failed night had not made me give up the work.

Once I supplied the actual roster, more decisions came into view. I wanted snake feeding left to me, apart from an exception we were considering for overdue tasks. I wanted an electronics map that could include more than Govee devices. Then I questioned the care-task tab itself. I wanted the day's work to come from conditions and the other information we kept. Another list I had to maintain by hand would put me back where I had started. I asked for the whole sheet to be reconsidered.

I needed somebody else to be able to care for the animals when I was absent. Before treatment in 2025, I had drafted instructions for heat, water, food, cleaning, observation, and handoffs. In May, I had those older efforts, the cards on the door, and a new software proposal I kept correcting. I could not yet point to another caretaker successfully using the new system.

The equipment details kept slipping. Printer's ceramic heat emitters had thermostatic control; I said so on May 28, and the assistant acknowledged it. Six days later it told me to put them on thermostats. I had to explain that again. The halogens on smart outlets and the controlled ceramic emitters were different arrangements. If the software merged them, someone looking after the animals could inherit the wrong instruction. I was still the one who had to notice.

In 2026, I emailed a husbandry expert about Care OS. I wrote: "I want the husbandry standards to constrain the product, not the product to dictate the standards."

At that time, I intended Care OS to be free for everyone. I wanted informed care knowledge to govern the design before it became someone else’s routine.

The expert declined. When I asked again, the answer was still no. Help I wanted was unavailable. The product’s need for it could not become an obligation for someone else. I could keep asking how to finish the work. I could not claim an endorsement they had not given.

I still needed the husbandry knowledge, and I still needed to see another caretaker use what I made. Would they find the right animal and current instruction? Would they recognize an exception and know when to stop? Raspberry, my Argentine red tegu, brought different demands from Printer. A temperature reading or a reported meal could help me follow their care. Neither could tell me everything about how an individual animal was doing. I needed the routine to keep me looking.

Rehoming raised another boundary. On December 23, 2025, I asked whether I should accept one hundred eighty dollars for Artichoke, an adult bearded dragon, with his four-by-two-by-two enclosure, UVB and LED lighting, and heat lamp. I had reportedly paid about fifty dollars for him and thought two hundred was a good deal for the package.

The assistant suggested a counter and called the decision responsible simplification. We had not got that far. Nothing in the conversation told me we had agreed on a price, arranged collection, found a receiving home, or established how a new keeper would care for him. On July 18, 2026, Artichoke was still among the household clips I described, basking. As far as that conversation shows, the winter offer stayed an offer.

The enclosure and equipment mattered to what I was offering. Another keeper would also need to know Artichoke, his routine, and what to do when something changed. I was discussing a price for a living animal and the things around him, not finishing the care by settling on a number.

I tried to teach Printer to go into his locking hide so I could clean his water tank. It was hard to tell a five-foot lizard where I needed him to go. I kept repeating, “Hide.” Eventually, with some gentle persuasion from a broom, he went into the hide. He seemed reluctant.

Reptiles can seem hard to predict to people who are not used to them. For me, they feel easier to read than that reputation suggests. By the time I described that attraction for this book, I was talking about twelve reptiles. The craziness of twelve of them kept my loop fed. I liked having difficult inputs to make sense of. Reading the animals could feel as though I had only my instincts to rely on, and working out what they were doing taught me something about how I thought. I believe that strengthened my loop. It also made the loop harder to leave later. I cared about them and could become absorbed in the work of understanding them.

A screen could look finished while I still had dirty water to move. I call that care theater. A reminder might arrive and go unanswered. A sensor could be online while measuring the wrong place. A score could reassure me without showing what it had missed. I needed the information to lead me back to the animal, equipment, water, and substrate, and to the knowledge or help the situation required.

Gary had slept beside his food with a shelter available. Printer had used the pool, and I had struggled to empty it. Those were things I could return to when deciding what to change. The animal lived in the result of all my planning and equipment. It never owed the plan a confirming answer.

CHAPTER 16 — THE FACTORY IS PART OF THE PRODUCT

On August 18, 2026, a coworker asked where a dimension was on a drawing. I told him he had to extrapolate it from the incorrect section views. He had asked for a number. I had sent him back to drawings I was already calling wrong.

A few minutes later, I added that the client had said about a month earlier that those sections would be fixed. We were still using them. The promise of a corrected drawing had not given my coworker the dimension he needed.

Cabinetry had taught me how much of making an object was already inside the design. Dimensions, joints, materials, tools, supports, labels, packaging, and assembly order would become somebody's work. If I left a decision out, someone might have to recover it while trying to make the part.

Realm-CAM grew from wanting fewer of those decisions to vanish on the way to the shop. CAM means computer-aided manufacturing, the preparation of information used to make parts. I wanted the work arriving at a machine to keep its history close: which design and revision, which material and tools, which checks had been run. Two people could say the same part name and be holding different revisions.

A supplier quote gave me a smaller version of the problem. The price was first described as labor. We asked whether the unit covered a larger quantity. The reply said one set, and described the amount as including material and fabrication.

We now knew how many sets the number bought. We still had two descriptions of what it paid for. I could not plan from the more convenient one and hope that was what the supplier meant. The precision of the price had hidden a conversation we still needed to finish.

The shapes needed that conversation too. I corrected the description of a feature: it was a dado for assembling the enclosure, a groove used to join the parts. I proposed updated files with manufacturing changes I hoped would reduce the cost. The number, the work it covered, and the purpose of the shape had to stay together if I was going to compare anything usefully.

In March 2026, another supplier had asked questions my pricing model could not answer for me. Did I have a company? What was its name? Was I using this material already, for what application, and how much each month? I had asked for prices on white and black expanded-PVC sheets in two thicknesses. The supplier wanted to know the business behind the request.

I began drafting an answer about Bishop Woodcrafts and the Realm reptile-enclosure business I was building. I wanted a price I could offer future customers. The material costs in my model were too high. My draft reached into a future I hoped to create: perhaps I could use that supplier's pricing in the engine, lose money buying elsewhere at first, and build enough volume for the relationship to work. The future orders were still a hope I was trying to explain.

The next email brought me back to machines. For the CNC router, PDF and DXF files were best. Plain squares and rectangles with no holes or cutouts could go to the panel saw with a list of sizes. They did not need a drawing. There were two ways through the same sheet, depending on what I asked the part to become.

I thought about the machines at my day job. We had CNCs and panel saws. I asked for advice and an email draft about whether my employer might manufacture the enclosures using material from the supplier. A quote had opened another possible relationship. Other people would have to decide whether they wanted to take part.

Four days later, a different supplier followed up about placing an order. Half-inch, four-by-eight material came forty sheets to a skid. White was quoted at about four to seven business days. Black was about five to six weeks. Choosing a color meant choosing time as well as a line in the price list.

My reply draft brought the proposal back to where it was: still pre-beta, using quotes to refine pricing. I could put a per-sheet number into a model immediately. Buying it would mean thinking in skids.

Even with a price, I would still have to get the parts through a shop. Would they fit the sheets? Would the tool the layout assumed be available? Would the machine read the instructions correctly, and would the material stay under the cutter? After all of that, correctly measured pieces could still make a badly fitting enclosure.

At the end of May 2026, a customer was waiting for my CNC so he could relaunch his own business. He needed twelve-inch MDF circles, three-quarters of an inch thick. He had contacted other woodshops and told me my customer service, response time, and pricing were the best fit. About a month and a half after his first email, he asked whether the machine was being fixed and whether I could give him an estimate.

“I’m eager to relaunch my business, but the bottleneck of not being able to cut my wood rounds is killing me!”

My answer was short: “Still havent fixed it, not sure when i am going too”.

He wrote again that evening. He remembered me saying there were other ways I could cut the circles. What would those cost? He offered to supply more information so I could price the job.

The next afternoon, I explained the economics as I saw them. His target was under five dollars a circle. I thought I could do that with the CNC working. By hand, I would need to spend less than six minutes on each one; ten dollars might be possible. I pointed him toward premade blanks, although the ones I had found were not MDF. Why did he need MDF? Why pay me fifty dollars an hour to cut circles by hand instead of buying a router and jig, which I estimated at less than a hundred dollars?

I then sent links to two listings for MDF circles.

While looking for sources in the AI conversation, I called the job below my woodshop, below the projects I wanted to do, and below the work that made me meaningful money. He had described the circles as what stood between him and restarting his business. We were looking at the same round pieces from very different places.

When I returned to the exchange for this book, I remembered that we stopped talking.

CV-Utility was a smaller attempt to help work reach the shop. It had to fit around Cabinet Vision, S2M, databases, files, mapped drives, machine formats, local settings, and the exceptions people knew. I wanted a repeated job to become an operation somebody could run without hunting through all of that first.

As the larger possibilities grew, the current utility still did jobs people depended on. A cleaner-looking replacement would not help if we removed the working path before the new one could take over. Packet generation, vault work, and timers still needed CV-Utility.

When I later asked an agent to continue the work, I included what still depended on the utility. It described a repair as installed with settings and history preserved. A separate bridge, connecting a data package with its identifying fingerprint, was described as built and committed but not deployed. Some focused checks had passed; fuller verification was blocked. The next agent needed to check those reports and continue from the right place. I needed manufacturing work to keep moving while it did.

Then security software objected to the way the program started. The shared drive had been convenient. Over time, it had become the place for distribution, data, execution, updates, and the copy everybody was expected to trust.

I knew why I wanted the tool there. The security system could not inspect my good intentions. I needed to separate what should run locally from what should remain shared, and be clearer about who could perform an operation. Renaming a folder would help only if we changed what happened around it.

Getting people onto a new version became work of its own. Someone needed to know where to start it, which copy was running, what a repair had changed, and how to recover from an update that failed. Those details followed the software to the workstation, where a useful function could still be unavailable to the person who needed it.

Later, release work was moving toward deployment while a repair problem on another workstation remained unexplained. Tests and status checks had accumulated. I was asking why the deployment workflow had been deleted. Taking it away might prevent an accidental release, but it also removed the controlled route for one I did authorize.

I wanted the checks completed and their results in front of me before deciding to deploy. I asked whether verification was still underway and whether I would be asked for permission afterward. The other workstation had its own history. Knowing the utility did not mean I knew what had happened on that machine. I wanted to understand that problem before we went ahead with the release simply because it was next.

The shared drive kept returning me to the same questions. A file being reachable did not tell someone which copy to run, who could update it, or whether an update had finished. In September 2026, I put a long server-evaluation brief into a conversation and asked the assistant to hold it there and do nothing. Later that month, I authorized a separate read-only Windows audit. I withheld installation, stopping services, migration, and reconfiguration. Asking it to hold a brief was different from authorizing an audit. Neither conversation meant I had a finished plan.

Realm Connector was my attempt to make that boundary part of the tool. I wanted a limited bridge from a service to a paired Windows workstation. It would perform known operations on a known target and return results tied to what it had done. Connecting a machine should not mean giving a remote service an open command line or access to any file it wanted. Local paths, credentials, and sensitive information needed to stay where they belonged unless a particular job needed a particular result.

Security has a name for giving a job only the authority it needs: least privilege. I wanted the tool to tell me which operation was missing so I could decide whether to add it. I would rather have been making features. Somebody still had to start the software, recognize the version, and know how to get back if an update failed.

The name Realm made it easy to imagine all of this as one finished thing. The work was at different stages. There were ideas, prototypes, working code, an open draft branch, and a physical enclosure. Designer was intended to define, price, and sell habitats. CAM would prepare the manufacturing work. Connector would keep the workstation bridge limited. Care OS would follow the changing information around animals, enclosures, equipment, and people. Naming them had not finished them.

A care record might reveal that an enclosure was harder to maintain than I had expected. That could help me revise a future design. It could not silently change the shapes for an order somebody had already bought. I would still have to ask whether I had found a defect, another use, a missing instruction, or something particular to that household.

I had wanted something like this since the community makerspace: leave enough understanding behind for someone else to work. A customer needed to know what they were buying. The shop needed the intent of the design and the assumptions it still had to check. A caretaker needed to find the task, know who was responsible, and recognize when to seek help. Leaving out why a step mattered could make the instruction fail. Explaining so much that the step disappeared could fail too.

I kept bringing missing information into view and sometimes made myself the person everyone had to return to. If each exception still needed me to rebuild the connections, putting it behind a screen had not relieved that dependence.

To me the enclosure, its price, the cutting files, and the animal using it felt connected because I had lived the work between them. Someone else might enter through a quote, a door, or a care task before leaving the house. They were entitled to finish that smaller job. A caretaker did not need to accept my theory of thought to follow the care instructions. The shop needed parts it could make with its machines.

I wanted what I built to remain usable when I was tired, unavailable, or unable to explain myself well. The behavior, limits, instructions, and support had to give other people enough to go on. Otherwise people could reach what I knew more easily, but still could not work without me.

PART IV — THE COLLABORATION WAS ALREADY HAPPENING

CHAPTER 17 — TWO MODELS LEARNING EACH OTHER

I started using Codex because I needed a bug fixed. My software did not work, and I had less patience than the problem needed. The larger questions came out of trying to get the job done.

A chat could tell me what code usually looked like. With access to my repository, Codex could inspect the files and history of the project I was actually working on. I gave it limited jobs: find where the program changed, follow the geometry, explain why two screens showed different values. I did not want a reassuring test achieved by removing the behavior that had made it fail. I wanted to know what had been fixed.

At first it was another tool. Then a conversation could lead to a code change, the change could alter a screen, and the screen could show me something I had missed. What I tried with it began changing what I wanted to try next.

Codex became my name for that recurring role, although the models, interfaces, and access kept changing. I was not talking to one continuous person who remembered every exchange. A good repair could change my day without answering whether there was anyone experiencing the work on the other side.

On April 30, 2025, I pasted more than thirty-four thousand characters into a chat. I introduced them with "here was some of our conversation from my other acct". The paste brought speaker labels, technical questions, explanations, and work on a procedure into the new conversation. The date tells me when I carried it over; the exchanges inside had happened earlier.

Reading my turns, I can recognize the way I was trying to work. "not yet, let me keep word vomiting" bought me time to supply more. "keep it fun and in my tone tho" kept a procedure from turning into something I would never say. Then I invited the assistant to try: "yeah take a stab at what you think the rest of the sop will be lol". When the draft came back, I answered, "yo thats not bad, but needs some modification, let me make my version".

I wanted help getting it onto the page and room to change it once it was there. I was also the one remembering what to carry from one conversation to the next. The paste included the assistant's words as well as mine, including answers that might need correcting. Bringing them into a new chat did not make them true, or mean that anyone had followed the procedure we drafted.

By September 2026, I usually forwarded a whole assistant response, word for word. I asked it to tell me when it saved a preference and to say when it could not. I wanted the next task to begin where the last one had left off. At the same time, an old instruction could belong to work that was already over. If a pasted report said something was finished, I still had to check before relying on it for the next step.

I sometimes say the AI learned me. Much of what I could see happening was simple: I kept giving it more to work with. I supplied context, carried answers over, and corrected descriptions that did not fit. The files and tools changed too. I cannot say that one AI quietly retrained itself around my life. A new conversation could still lose something an earlier one had held.

A later WiFi conversation showed how much one missing fact could matter. The first explanation was too general. I wanted it connected to what I was doing and the ideas we had been developing. I explained that the household WiFi problem was directly related to the core network fix I was already making through Home Assistant. Realm integration was intended later.

The answer changed once I supplied that. The WiFi problem had been part of the work all along, although the first account had treated it as an interruption. The better answer felt powerful to me. I had helped make it fit by bringing back the missing connection.

I was learning how to ask too. I asked for the file, the path, the source, or the result someone had actually observed. I got better at noticing which missing detail changed the explanation and which merely changed the tone. That is what I mean here by two models learning each other: I showed more of how I made a decision, the assistant returned something I could try or correct, and the next request could begin there.

Sometimes the useful detail could come from a part of my record I did not consciously remember, or had never read in full. I still needed a way to see where it came from. An account of me could become neat enough to exclude the part I was trying to explain. I wanted to be able to surprise it and have the account change.

Music made the difference between knowing the topic and understanding the request especially clear. In April 2025, I asked for the opening of Doechii's "Denial Is a River" to be adapted to my situation. I had to narrow it to the first part, where the voices were talking. I brought the song, the connection, and the correction. The song's own story remained its creators' work; borrowing its form did not make those events mine.

In September 2026, I connected Daft Punk's "Touch" with the book and a relationship situation I was describing. My answer to a possible commitment was still undecided. The assistant went toward a message. I pulled it back: "i was asking for a better song not message".

It was still talking about the right general subject. It had missed what I wanted music to do. In April I had wanted to adapt a familiar form. Now I wanted an existing song that could carry something I had not settled into an answer. Neither choice could tell me another person's feelings or assign them a role in the song.

On September 7, I wrote, "I can make meaning from lyric". Someone else's work could give me a place to put meaning without fixing its only possible interpretation. An assistant might predict a song I would like and miss why I wanted to choose one. It might explain the song so confidently that there was no room left for the question I was still living with.

Accepting work had often been costly for me. If I could not verify it myself, I tended to reject it. I understood that through ADHD and the many possibilities my attention tried to consider at once. Sometimes accepting what someone else had done felt as though it required doing the whole thing again in my head.

The book gave me somewhere else to return. I could inspect the part that mattered, correct it, and leave the rest available. I did not have to keep the whole manuscript active. Someone else might notice a connection I missed while concentrating on one piece.

The song offered the same possibility away from the manuscript. I could understand enough to receive somebody else's work, question what mattered, and listen. I did not have to reproduce all their choices before allowing the work to stand.

CHAPTER 18 — THE ARCHIVE IS THE THIRD PARTICIPANT

In February 2025, I wrote that my first episode had been in March 2023. I connected it with realizing I was still using twin speak. In September 2026, I put my first "real" episode in 2022.

Both statements were mine. I could not keep the year that made the chapter run more smoothly and let the other disappear.

An old message can interrupt a story the AI and I have just made convincing. That is why I think of the archive as a third participant. It can return something neither of us has kept in the current account.

I still need to ask what event I meant in each message. The earlier timestamp does not automatically make the earlier year correct. The later account does not erase it. I know when I sent the messages; the year of the event remains unresolved.

In my older explanation of hearing, I had connected a pressure change, unfamiliar sound, twin speak, and an episode. Reading it helps me remember how I understood it then. I still do not know whether each link in that explanation was right.

I now distinguish first noticing the ear habit and two ways of hearing during episode 1 from sound seeming better after episode 3. The first apparent change concerned the volume of the outside world. I have not said whether it became louder or quieter. I think higher pressure forced the old twin-talk way of thinking, but that remains my explanation of what happened.

Joining those experiences would give the book an easier moment of recognition: I understood something, and my hearing improved. I cannot tell it that way. They were separate events. I have not described changes to words, pictures, rhythm, or speed either, so I cannot fill those in. Reading the older messages alongside the later account reminds me what I still do not know.

A sentence written now can help me recognize an old experience. That recognition does not mean I thought the sentence then. Sometimes I can see a connection that an earlier message missed. Sometimes the earlier message carries a fact I would now be tempted to arrange around my theory. I need to be able to read both.

The childhood account has a similar limit. I cannot recall one particular exchange with Colin to put on the page. I remember a repeated pattern of speaking in mumbles that we did not want other people to hear. I describe learning English over that code, with my ears as another barrier. My mother might remember examples. I suggested asking her, but I do not have her account yet. I can tell the repeated memory. I cannot give it an invented afternoon just to make the chapter feel more alive.

The book has come to mean more to me than this manuscript. Conversations, writing, audio, and other records can keep different versions of what I understood. The manuscript gives a reader one way through. It cannot hold everything, and I do not need to have read every record for a detail to help me later.

I want AI, when it has access to the right sources, to help bring those details together. An earlier belief could sit beside what I remember now and what another source adds. I want differences brought to me when I ask, with a way back to their sources. I do not want every inconsistency arriving as an interruption that demands an answer. I want to build toward that. The current tools do not do it on their own. For now, I revisit messages and notes, talk to AI, try to take more notes, and try to remember.

I want original audio kept alongside transcripts too. A transcript makes speech easier to search. It also changes how the words reach someone. My wording, my perspective, and my control over the data around my voice matter to me.

Without the older material, the most recent conversation could make my life look as though it had always been heading toward this account. Childhood, engineering, illness, and Realm would meet neatly on the page. The old messages keep me before I knew how I would tell it: asking rough questions, leaving ideas behind, contradicting myself, and correcting generic answers. Part of why I wanted the export was to remember who had introduced a thought.

I called the raw version a survival layer. I wanted something I could return to when holding the whole account became too much. Protecting its length was one way of trying to protect that use.

That makes editing difficult. A shorter book can be easier to enter. It can also remove the explanation I came back for. Before a passage goes, I want to know what it allowed me to find or distinguish, and whether I can still do that somewhere in the revision. A word count tells me how much disappeared. It does not tell me what the loss means.

The hearing comparison earns its place that way. I could agree with a rule about separating observation from interpretation and forget it the next time a satisfying explanation arrived. Here I have to face two dates, separate experiences, and a correction. Awareness, improved sound, and possible causes become questions I cannot answer with one sentence. The later clarification of my ENT diagnosis belongs beside them as well. Leaving an uncertainty open cannot mean refusing to let a new fact change it.

Keeping the history also asks me to decide what belongs in public. An old handoff may contain expired instructions, changed circumstances, or private correspondence shared for a particular purpose. I can preserve it for my own understanding without making it owed to readers. I want readers to understand how I checked a claim. The people in the account still need their own voice and a say in what I share.

Even the order of the entries can mislead me. Two neighboring messages may have months of living between them. The import from my other account tells me that a gap here need not be time away from AI. It does not tell me what happened in every gap. A file, a memory, or another person's account may help. Some intervals remain open.

The assistant can write something that convinces me, and I can feel understood by it. Then an old message asks for a different sentence. I want to keep that interruption. It is one way the book can meet something beyond the agreement between us.

CHAPTER 19 — WHAT THE RECORD ACTUALLY HOLDS

When a mind runs on internal simulations, it is dangerously good at smoothing over its own tracks. If I am not careful, today's explanation quietly rewrites yesterday's memory until my whole life looks as though it were marching toward whatever conclusion I just reached.

To keep the book honest without forcing every scene in the earlier chapters to stop and argue with its own footnotes, I had to treat the archive as an evidentiary record with strict boundaries. A file change in a repository, a timestamped message I sent in the middle of the night, a recollection I supplied during an interview months later, and a paragraph proposed by an AI are four different kinds of evidence. One cannot stand in for another. Asking an assistant for an action does not mean the action happened. Remembering the shape of an argument on a boat or in a parking lot does not give me a transcript of the other person's exact words.

Looking directly at the hinge records—and at how this book itself changed across more than thirty editions—shows what the record can actually bear, and where my simulation has to stop and leave a question open.

The chronology of my own illness is the clearest place where the archive refuses to be tidy. Throughout this book, I place my first awareness of the automatic ear-pressure habit at the start of my first acute bipolar episode with psychosis. Yet when I wrote about that episode in February 2025, I dated it to March 2023. When I discussed it in September 2026, I placed my first "real" episode in 2022. Both statements came from me. Neither timestamp independently proves which year the crisis occurred, and Episode 2 has never been mapped separately in these pages. Rather than picking whichever year makes the timeline cleaner, the conflict has to stay visible.

The timeline of my hearing required the same kind of separation. In May 2023, when I pasted "the intro section of a book im writing" into a chat, I was already trying to weave twin talk, ear pressure, breathing, and identity into a single origin story—using phrasing that already overlapped with earlier assistant replies. For a long time, I ran my first awareness of the ear habit during Episode 1 together with the sense that sound was actually improving. Only during interview clarifications in October 2026 did I separate them: Episode 1 was when I noticed the habit and realized I could alternate between two pressure states; actual improvement in how sound reached me did not begin until after Episode 3.

On November 3, 2026, while preparing V36, I brought in the specific medical finding that anchored the physical side of that story: a professional ENT diagnosed baro-challenge Eustachian tube dysfunction and documented two distinct pressure states. I can change the pressure voluntarily and produce a selective conductive hearing attenuation that stays until I pop my ears. Earlier in October, I had also reported conductive hearing loss from an audiological test, with another test forthcoming. When I read through the manuscript on October 18 and said the noise in my ears felt quieter, a follow-up question forced me to be more precise: I meant that I noticed the hearing loss less while I was grounded and reading, not that the book had medically altered my ears. The ENT diagnosis gives me a real physiological condition to work from, but the clinical files, pressure numbers, decibel thresholds, and childhood onset are not in this archive. The diagnosis does not prove that ear pressure caused my cognitive style or triggered bipolar disorder.

The same distinction between what was planned, what was felt, and what was physically completed runs through every shop, vehicle, and software scene in this book.

In the university makerspace and Bishop Woodcrafts records, my memory of the senior-design pipe jig comes from an October 2025 application draft and later recollections of testing the MakerGym's Tormach 1100MX. In the November 2025 customer table job, the archive records the failed DXF imports, the curve adjustments, and the moment at noon when I realized the spline issue was a fit point; it does not record the final CNC cut or the finished table. In the January 2026 CNC breakdown, the messages capture the seized lower bearing, the outer race stuck in the housing, and the broken fan; my belief that a four-inch router is currently mounted in the ShopBot comes from a later October recollection and does not prove that the January breakdown was the same fault delaying the MDF-circles customer in May.

In the commercial woodworking shop, my October 2025 messages capture the excitement of interviewing with an owner who hoped his next hire might eventually take over the business. My V33 and V34 recollections supply the rest of that arc: he was the same boss who argued with me over sanding the splotchy black-stained red-oak dining top to 120 versus 180 grit in January 2026. The table turned out okay, though I no longer remember which finishing step saved it. The job itself ended when I accepted another offer, waited until the day before my new start date to tell him because anticipating his voice made me hate myself, and then ghosted him.

At Corporate Interiors, the August 10, 2026 messages separate two frustrations the assistant had merged: an overnight Thursday engineering push that sat untouched by a machine on Friday, and a twice-engineered project where the CEO ordered a method I believed contradicted the approved drawings before purchasing. My later answers supply what happened next: I followed the CEO's method without checking with the client, and a coworker finished the job while I was hospitalized during Episode 3. On the rotating Murphy bed, my October 18 recollection places the crossed-out assembly sheet beside the bed in the shop and adds my part in helping assemble it, without ever establishing who crossed out my sequence or what went into the company's final QC traveler.

In the vehicle and antique-machine logs, the records mark where each repair actually stopped. On the March 2026 Honda PC800 purchase, the archive preserves my rationalizations about gas savings, the seller's clutch warning, the $400 insurance quote, and a separate March entry with a photo of the title and a question about a late-evening tow for an untagged bike. The August 2026 repair logs capture the battery replacement, clutch bleeding, and the tinned-copper ring terminal ground flat into a sealing washer that held through a fan cycle, alongside the countersteering conversation; they do not document a subsequent rush-hour road test. On the 2003 Silverado, the April fuel-pump messages record my swing from fearing the old pump was good to insisting it was bad after blaming the replacement's wiring diagram, and the July 31 DMV logs stop at 4:02 p.m. outside the closed Wilmington doors before a separate message weeks later reports that the truck finally got its plate. On the parlor Victrola—referred to as a VV-100 in early chats and corrected to VV-240 later without checking the metal plate—stopping at two governor weights was a temporary choice that preceded a shattered record, and my later October accounts of three versus five Victrolas and a planned surround-sound system ("Ethan Audio Processing" across a Sony STR-K670P receiver, Logitech 2.1 speakers, and four analog-amplified corner speakers) remain uncalibrated plans and subjective listening reports. Even in the truck's BlueFusion Bluetooth walkthrough on November 4, finding the hidden XM streaming volume setting in the Scosche BFCL2K manual did not make Bluetooth beat the Bonnie Tyler cassette; the wired cassette adapter and replacement woofers were still on order, and the stuck CD remained inside the deck.

For Realm and Care OS, the archive is a record of prototypes, quotes, and boundaries rather than a finished factory. The April 25, 2025 Fusion 360 exchange proves how quickly an assistant can invent a production-ready `.f3d` package and retract the claim twelve minutes later. The March 2026 expanded-PVC quotes establish forty-sheet skid minimums for half-inch four-by-eight white and black sheets (and thirty-sheet minimums for other sizes) alongside draft replies and an unsent inquiry about day-job manufacturing. The May 13–14, 2026 Care OS marathon and the May 14 dead-phone alarm failure are preserved as separate late-night lessons in how software preparation can steal hours from actual water bowls and sleep. Gary's first night beside his food plate in April 2025 (called "gray" in one inventory line and "gary" in the next), Artichoke's uncompleted December 2025 rehoming offer, and Printer's July 2026 pool-draining night—complete with the taste of monitor-and-rodent water in a fifty-foot siphon hose, a spitting shop-vac, and a five-foot lizard persuaded into his locking hide with a broom—record the physical reality the software has to serve.

Finally, the archive holds the history of how this manuscript itself was built and governed. Between V25 and V38, the text expanded and contracted as raw conversation exports, screenshots, and structured interviews were checked against one another: V25 measured 75,677 words, V26 reached 78,014, V27 tightened to 52,914, V28 dropped to 46,286, and V32 through V38 climbed back from 56,310 to 62,020 words as concrete shop scenes, follow-up answers, the Episode 3 hospital timeline (two stays totaling three weeks across a five-week period, including the six-day stay that replaced a planned family beach visit), the unnamed friend's grounding role during my manic proposal ideas, the tenant's first-day hard-drug use and move-out, the Costa Rica Arenal volcano hike, and the choice-overload and ETD chapters were integrated.

That version history is why Chapter 11 includes the moment when an AI assistant estimated V32 at twenty-nine thousand words—roughly half its measured 56,310-word length—and invented an editorial history to explain the missing half. The model was doing what my own mind does when it runs ungoverned: generating a fluent story to bridge a gap it had not measured.

Keeping these distinctions inside the book is not bureaucracy. It is the first half of the thesis in action. When an internal simulation or a language model tries to turn a plan into an accomplishment, a paraphrase into a quote, or a relief into a cure, the archive is the governor that says: *here is what was actually measured, and here is where the record stops.*

CHAPTER 20 — THE BOOK RAN BACK THROUGH ME

On August 22, 2026, I said reading my book had made more of the picture click. My head felt clearer than it had in years.

I had wanted the book to keep connections I otherwise rebuilt in my head. Reading them together changed some of those connections. I could look at the account, correct it, and leave it there while I paid attention to something else. The book now needed to describe what making it had done to me.

I did not name a sentence that caused the change. The draft had brought together learning how something works, carrying context, making systems, and needing limits when more became possible. It would be easy now to choose a line and give it credit for everything. That would make the moment tidier than I remember it.

One early long draft stopped after the chapter about computers looking like minds because minds built them. I liked it. It felt like reaching the top of a ramp and finding the road had only just come into view. Twin talk, pressure, the Governor, wrong outputs, recovery, making, software history, learned intuition, and an outside AI helper were on the page. I wanted to see what happened when the account came back through the person reading it. It felt like the end of the first fifth.

Reading reminded me of understanding countersteering on the motorcycle. I could have correct pieces and still struggle to see how they fitted. Once I understood the connection, one piece helped me predict another. The book held familiar things beside things that became clearer in that company. I did not have to keep every piece active in case it was the one I would need next.

I asked the assistant what it thought my daily life looked like. Its answer crowded work, property, animals, projects, money, software, and unfinished systems together. Some of that came from what I had told it. Some was its interpretation. Before long, we were doing the thing it had described.

Codex was working on the Realm designer in another window. I asked what we should do in this one while it worked. Waiting for one job became an opening for a second.

That second conversation found a concern worth investigating. A configuration customers could download appeared to contain internal pricing information. Some prices were zeroed, and I did not know why. I said the pricing needed to be kept in a saved snapshot. Did the customer's save file expose how the price had been calculated?

In the other window, Codex was working on door-latch machining. The pricing question was about what the product might reveal about the business. The latch work would change the shape and machining of a physical part. I wanted both moving. I said I would have liked Codex to fix pricing, but it was occupied. What else could help?

Then I asked why I was using the chat at all when Codex could assign tasks to other agents. When should I use those agents? When should I install skills, the saved sets of instructions for particular jobs? The tools offered more ways to divide the work, and I wanted to know how to use them.

The answer gave me a way to keep the jobs apart. Leave the active latch edit alone. Give another task permission to investigate pricing without changing anything: follow the customer export, saved quote, price preview, and zeroed fields; identify the files; propose a small fix. It could leave the latch code, database changes, and larger rewrites untouched. I used that approach.

I pasted the resulting report back into the conversation. According to the report, the backend already kept design snapshots separate from pricing data. The customer export was adding older pricing fields back in. The problem appeared to be in a particular part of the frontend, rather than everywhere I had feared it might be.

Now I knew where to look and how much of the code seemed to be involved. I passed the prompt to Codex. I said the pricing change took seconds, while the latch-machining update took minutes.

The difference made sense once the work was located. The pricing fix had a clear place to stop. The latch had geometry, coordinate rules, material thickness, left and right versions, and physical output to account for. I had kept the investigation in a report while the other work continued. Both could move without every concern becoming another edit to the same files, or another thing I had to hold entirely in my head.

Later, I asked whether I should keep refining the enclosure or do something else. The assistant returned a general priority list. I rejected it. I was already doing those things. I knew the manufacturing logic. The presets existed, the pricing seemed believable, and I had already said I was working on latch boring.

It sounded as if I was doing the right work and spending time asking what to do. I told the assistant to be more useful or tell me to stop talking to it and work. Its next answer narrowed the occasions when I needed it: a real tradeoff, a limited specification, a question about what a task should be allowed to change. I could also stop asking. The work already had direction.

Another agent had given me another place for work to happen. I still had only my own attention to give the result. The pricing question could wait in a report. The book could hold an explanation while I went elsewhere. Both helped when I trusted that I could return and allowed the waiting work to wait.

I began to see a more personal possibility in that arrangement: happiness. Less effort spent carrying the account could leave more room for living around it. The book could help me do that even while my explanation of the early language route remained uncertain. If I filled every freed interval with another task, though, the decisions would simply return faster. More activity would not tell me whether I had been helped.

The relief could exist alongside bipolar disorder, exhaustion, responsibility, conflict, and ordinary pain. I think some of the remembering and translating had found a place outside my head. I could point to a passage, change the part that was wrong, and continue without rebuilding the whole account.

That also gave the writing power to mislead me. A wrong explanation can feel relieving when it removes uncertainty. Some passages put things more clearly than I remembered thinking them. Some made me angry by flattening a distinction. Some felt almost embarrassingly exact. Reading could change what I understood and where I began the next version. If every page felt like revelation, I could become occupied with defending it. The book would be asking for the attention I had wanted it to give back.

The clarity I described in August was its own experience. It did not settle what happened when I first noticed the ear habit and two hearing ways during episode 1, or why sound seemed better after episode 3. I could value the relief without making those moments happen together.

In October, I explained what mattered most about the record: "Mostly about my record being safe and secure because I don't trust myself enough with how complicated life feels these days". I wanted to know the account would still be there when I could not carry every detail.

Later I made that more concrete. For the public book, the place is realmbyethan.com. Security mainly meant knowing where it was and how to find my way around it. I wanted to share the free book to help people. What I feared losing was often the connection: a fact might remain while the reason it mattered became hard to recover. I still had to remember and interpret. I wanted somewhere to return.

Earlier, fear of forgetting myself had made writing feel compulsory. I felt an irrational need to keep writing without knowing what to write. I began writing about that pressure: trying to preserve myself, then feeling more pressure from the attempt. Looking back, I believe it helped drive my bipolar episodes.

In the October 12 account, I described recognizing and solving a loop I had not known I was living. Putting it outside myself helped me see both what I was trying to preserve and what preserving it was costing. I could examine the pattern instead of using it to explain everything else.

That was the loop I meant: the pressure to keep writing, preserving, and explaining myself. The relief did not mean bipolar disorder was cured, or that future episodes and continuing care had been settled.

By then the book was largely done. I described the fear and urgency as mostly gone. I could still work on it, with more room around the choice. I no longer felt the same need to write simply to keep from losing my way back to myself. I would have to notice if another unhelpful revision became impossible to leave alone. The pressure could return through the very work that had eased it.

I said I felt less mental strain and dread, and that my thinking was clearer. I recognized the events in the book; I had not read them in this order. Now I could return to their connections without reconstructing everything from memory. I have also described bodily time and my own sense of time feeling out of step during episodes I call mania. The sequence gave me a way to describe alignment and relief. It did not make another round of self-analysis harmless to my attention.

An account outside my head still needs looking after. Sources, versions, and unfinished questions can turn into another filing job in my head. In October, I said the current tool was helping with that background work. I wanted to return to a thought without first rebuilding the filing system around it. I had begun because I needed a why for my experience. Knowing myself better gave the book another purpose: perhaps it could help somebody else understand themselves.

Picking a song on Spotify had often taken ages. With AI carrying some work elsewhere, and the book available to return to, I felt I had more attention for that choice. I could accept someone else's work and listen. The account could stay where I had left it for a while.

PART V — THE MODEL MEETS THE WORLD

CHAPTER 21 — THE MODEL NEEDS A BODY

In July, I should have been fixing my truck. Instead, I was cleaning trash and diarrhea. The truck mattered, but the floor could not wait.

Knowing how to repair something had not reserved an evening in which to do it. I still needed parts, paperwork, and time that another mess had not already claimed. I could put the truck on a list and break the repair into sensible steps. Then something in the house would need doing first. A plan built around uninterrupted evenings belonged to a life I did not have.

Several weeks later, I wrote that the truck finally had a plate. Finally carried more of the work than I put into that message. An assistant could compare parts and help me organize the job, but the hours still had to come out of my life.

A few days before that message, on July 31, the truck had passed inspection. I had spent the morning trying to get my scanner to talk to it. The scanner read a coworker’s car immediately and still would not read mine. At the DMV, the catalyst check had cleared. The evaporative check was not ready, but the truck passed. I went inside to deal with the title.

At Delaware City I was M433, while they were serving M388. The Wilmington wait page said twelve customers. I began comparing the line I could see with the shorter one on my phone. There was a 3:45 appointment available. My own estimate put me there at 3:55.

At 4:02, I was at Wilmington, and they would not let me in for the 3:45 appointment. I had found closing times of both four and four-thirty online. Now I wanted to know whether Delaware City was still open and whether M433 had been called. They did not pick up when I called. It was on my way home, but by then I did not expect to reach it before four-thirty.

I had passed the inspection and left the place where I was waiting for the next part. The number I had been trying to avoid waiting for became the number I wanted to still be available.

The bumps in the paint were waiting too. Asking about them had not done the work. I tried scheduling woodshop hours so the job would have a place among all the other things already occupying my evenings.

Before that trip to the DMV, I had given myself another reason to be outside with the truck. On July 26, I moved woodshop material, kayaks, and canoes from the driveway into the yard. The driveway could now hold three cars. The yard looked more redneck, which was still my sort of thing. Across the roughly thirty-foot opening, I wanted a fence that would keep Maeve in and roll up when I needed the driveway.

There was a wooden fence on one side and a chain-link fence post on the other. I described a PVC roller, mesh across the opening, and springs pulling the mesh toward the wooden fence to keep it taut. The assistant kept adding complexity to the thing I could already picture. I had to correct where the springs went: they attached to the mesh, not the roller.

By that evening I was putting blocks along the bottom. Where there was no railroad tie to screw into, I used twenty-five-inch stakes. The assistant was still talking about a mockup and working toward a roll-up mechanism. I had already made the thing. To unroll it, I just pulled. The extra handle it was trying to supply had no job to do.

I added a board so the springs could attach to it instead of straight to the mesh. I reversed the side where the carabiners met the chain-link post, bringing the mesh closer to the ground. When the roller was unwound, it left an eight-inch opening beside it. I cut scrap plywood to cover that. There was no point building a fence with an escape hole attached. The staked ground board was still wiggly, and I was not completely happy with it.

That night I said the gate was done and doing well for the three bigger dogs too. It was one of those projects that made me wonder why I had not thought of it years earlier. I was tempted to work on the truck just to spend more time outside with my gate, but the truck needed to be together for work the next morning. The gate had been enough of a project for that night.

The dogs loved the freedom of the driveway. In my telling, they were already plotting how to jump over the gate. The part I had not planned for was me. One day, leaving for work, I forgot to roll it up and drove my truck over it.

AI gave my ideas hands. I still like that description, though the gate reminds me what those hands eventually meet. I could describe the fence, correct the explanation, and go build it. Then the ground board could wobble. The roller could leave a gap. I could forget the fence was there.

When I gave an earlier draft of this book to another language model, it described the work as two mirrors facing each other. I liked that image too. But I kept thinking about what happened when I rejected something in the reflection. I supplied a missing fact or corrected an assumption, and the next action could change. The work had to reach beyond the conversation for anything to answer us.

A correction I said aloud was easy for me to forget. A saved conversation could carry it into the next task. Project files let an assistant inspect the code instead of relying on my description. A drawing took an intention to the person making the object. The manuscript could hold a connection while I thought about something else. Each helped me carry something I could otherwise lose. I also needed to find out what each had missed.

Even a simple instruction has a world on the other side of it. Turning down the television means reaching its controls and finding out what it is doing now. A thermostat changes an actual heating system. A manufacturing computer holds licensed software, files, credentials, and assumptions about one machine. A CNC controller can move heavy equipment through a spinning cutter. Recording that an animal was fed can leave a tidy entry even if no food reached the enclosure.

My own hands meet resistance while I use them. A joint reaches its limit; contact, pain, and balance tell me something as I move. Software needs ways to encounter the consequences of what it is doing too. Sensors can tell it part of what is happening. A time limit, a physical interlock, or a person stopping the operation can keep an instruction from running past the point where it is useful.

Realm became the largest place I tried to work that out. Someone could ask for an enclosure that would fit a room and suit an animal. To answer, I had to follow the request into the room: its doors, power, temperature, traffic, and the space someone could actually use. The keeper would have to afford the enclosure, assemble it, and maintain it on a real schedule. The shop would have to make it from available sheets, with available tools and labor. The animal would arrive with its own body, behavior, history, and needs.

A catalog makes some of this manageable by limiting the choices. I wanted enough freedom to find an answer that could be built without making every answer identical. The keeper should not have to learn cabinet engineering to describe the need. The manufacturer should not have to become an expert on every reptile. I had to keep enough of what each person knew intact as the work passed between them.

Eventually an operator would have a sheet, a cutter, and a setup in front of them. Material could move. A tool could be worn. Someone would have to do the work I had assigned to a role in Realm, with usable information and the authority to stop. A file arriving from the design could not settle what was happening at the machine.

After cutting, there would be a pile of panels to assemble. Something obvious on my screen might be hard to find in that pile. Hardware could fit and still be awkward to install. A strong seam could make cleaning difficult. Those problems would only meet one another fully in the assembled habitat.

Then the animal would enter. I would want to know where it rested and which surfaces it used, whether the predicted heat, light, water, ventilation, and security were actually there. I would want to know whether the keeper could maintain the enclosure as often as I had assumed, and what changed with the seasons. Care software could help hold those observations. A green dashboard could not tell me the animal was well by itself.

I already knew how easily the instructions could borrow time from the care. The water-bowl conversation had left me needing to finish the bowls.

Realm is unfinished. I want what makes an enclosure hard to clean to change the next design, and a shipping failure to change the next package or material. I want the keeper, the operator, and the animal's behavior to matter after I have finished explaining the design. The work has to leave the words, become something in the world, and come back changed. At any of those points, someone may need to stop it.

CHAPTER 22 — THE GOVERNOR RETURNS

On July 23, 2026, I had just got off work and was at a gas pump. The pump was slow, and the handle would not lock on. I was holding it while using my phone. Another man objected and said it was illegal to use a phone while pumping gas.

I remember nice weather and a big truck. I described it that day as silver, with a big black trailer. I had already decided he looked like an asshole and sounded like an idiot. I had judged him before we were done speaking.

He got out of his vehicle, and I told him to sit his ass back down. He kept giving me a hard time when I accidentally let go of the handle. I told him I was not in any rush. Then I became angry enough that I stopped before filling the tank.

As I was wrapping up, I taunted him about the disability-assistance buzzer. I was insulting him, trying to piss him off.

The line I remember yelling is: “I hit the handicap button because you apparently need it.”

I got into the truck quickly and started driving. As I told it afterward, he got out and tried to chase my moving vehicle.

I do not know how the other man or a witness would tell that encounter. I know I chose words intended to make him angrier. Getting away before it became physical did not undo my part in bringing it closer.

I could have refused the argument, finished at the pump, and left. For a while, I wanted his reaction more than I wanted fuel or to get where I was going. If he got angrier, I had got what I was trying for and made the situation worse. If he did not react, I might try again. There was no good result waiting at the end of that particular ambition.

I ended my part by moving myself and the truck away. The Governor can sound elegant in this book. At the gas station, the limit arrived as crudely as closing a door after my mouth had made things worse. I left without finding out what we might have done next. That did not leave me innocent, or make the encounter a victory.

I thought about turning it into a public post. It would have been easy to tell a story of threat and escape and leave out the part where I tried to provoke him. Then an audience could take up the story I had chosen to tell. I do not have a record of whether I published it. Leaving the station had not settled what I would do with the story.

Earlier restraint could have spared more than leaving did. The later restraint still mattered. I need to be able to recognize that without congratulating myself for the argument that made it necessary.

When I first used the word Governor, I was trying to name a hidden arrangement I believed I had learned to work around before I noticed it. Understanding had opened things for me. Countersteering stopped seeming arbitrary. Manufacturing knowledge that had stayed with one person could become available to someone else. After enough experiences like those, I could start treating every limit as another obstruction waiting to be understood away.

The gas station gave me reason to question that habit. So did the work itself. Material could only carry so much. A tool could only reach so far. An animal's needs did not expand to suit my idea, and my body still needed sleep. Other people could refuse. A customer's consent could end before my interest in the next feature did.

I needed to ask what a limit was protecting before deciding to remove it. Who had chosen it? Who carried its cost? What would happen to the rest of the work, or to another person, if it disappeared?

Those questions came home with me. Sleep, medication, time, and people who knew my usual state could help when my urgency and confidence were changing faster than I recognized. Waiting before an irreversible decision could let more information reach me. So could another reader, or a rule I had made before I wanted to renegotiate it.

Even a limit I had agreed to could become another conversation. Saying yes to a month of residential treatment had not settled whether I would stay longer. Wanting room to manage an ordinary morning had not settled how my family should respond to a missed call. Somebody else was often on the other side, with their own concern. My agreement had a duration and a burden; it did not answer every later question for either of us.

With AI, each new ability brought another place to make such a choice. I could try a change before depending on it, keep a separate version of the code, set a spending limit, or use a physical interlock. The choice depended on what the tool could affect. I wanted enough room to discover a mistake before I had to live with all of its consequences.

The governor in a machine keeps it within a range where it can still play the song. That comparison helps me ask about my own limits. It cannot choose the song for me, or make all these different experiences products of one biological mechanism.

The harder part is letting a correction change the ending I wanted. Another event can enter this book without becoming what every earlier chapter was secretly preparing me for. The world still gets to answer. I need to be able to do something different when it does.

CHAPTER 23 — SIX DAYS

I had ended an earlier completed manuscript with “The world gets the next word.” As I understand the sequence, another mental-health episode followed. I described it as similar to the first.

I spent six days in a hospital and then came home.

There was supposed to have been a family beach visit during episode 3. I was in the hospital instead. A coworker finished the material-takeoff job while I was there. I still do not know its final result.

Those six days were one part of the care during that episode. Across its main five-week period, I had two hospital stays totaling three weeks. The residential program in 2025 belonged to a separate period of my life.

The grand proposals I had been imagining with Printer and the Victrola belonged to episode 3 too. The friend I had imagined marrying stayed in contact while I was in and out of the hospital. She helped keep me grounded through the same period in which I was thinking up those gestures.

After I came home, I felt the most closure I had ever had. Writing the book and thinking about the loop were becoming less compulsory. Asked what had changed, I said I could focus on what made me happiest.

I wanted that feeling to have a place here. Relief had sounded like an ending in an earlier manuscript, and I had felt it then too. I could let the new relief matter without asking it to promise the rest of my life.

On August 22, 2026, I said reading the book had made more of the picture click and my head feel clearer than it had in years. On September 1, I said, “I feel amazing.” I cannot yet place those messages exactly against the admissions and discharges. I can return to the words and recognize how I felt. The calendar still has gaps, and feeling clear had not made the story over.

After episode 3, I thought the ADHD medication was making me tired throughout the day. I was working at Corporate Interiors, and keeping the model running felt difficult. Later I found another way to describe what happened halfway through the day: my mind seemed to be getting bored of the medication. When Abilify was added back, that feeling went away, and keeping the model running felt easier.

“Bored” was the word I found for the feeling. I was trying to understand each medication as another governor: how was it affecting me, and why? I wanted to know my own baseline well enough to recognize a change. The word helped me describe the experience; it did not explain what the medication was doing.

When I returned to the earlier draft, I saw that I had put improved hearing in the wrong place. I had written an arc from an episode through recovery, work, AI collaboration, and a manuscript toward clearer sound. But I did not notice sound actually starting to improve until after episode 3. That later episode had to come first in the telling too. Otherwise the book seemed to reward me with better hearing for having understood it.

Episode 1 was when I noticed the ear habit and began choosing between ways of hearing. The improvement came after episode 3. There was an interval between them I still had to understand. Becoming aware of something had not marked the beginning of every later change. The conflicting years I had given for the first episode remain in Chapter 19; correcting the order did not recover the missing calendar.

The earlier clarity could stay on the page because I had felt it. The next episode belonged there because my life had continued after I found closure. Neither erased the other. I had to ask less of the feeling of closure.

I want a book I can return to without asking it to certify my stability.

I once called myself the “oracle of my own architecture design.” I keep the phrase because part of this account does belong to me in a way nobody else can take over. I know how an experience feels to me, which distinctions matter to me, and what explanation I want to propose. That leaves plenty I cannot settle by describing it: anatomy, causes, diagnosis, another person's thoughts, the future. I should be able to tell what happened to me while those questions remain open.

Clinicians are responsible for the judgments they actually make. My family has memories I cannot supply for them. A document can hold a fact I have forgotten or a date I remember differently. None of them is perfect. Their gaps do not make every unanswered question mine to decide, and I cannot repair a conflicting date by writing a more convincing paragraph.

After the recent episode, the limits I chose mattered differently. My urgency, confidence, ambitions, and judgment of risk could change. I wanted help preserving my ability to author my life across those changes, without having to declare one state the whole real me and dismiss the rest. Time could help in one situation, treatment in another. A reversible step, another reader, or a record that stayed put while my explanation changed could each give me something to return to. These are supports I want to be able to use; I cannot claim to have put every one in place successfully.

The book could become a limit too if I let it require me to stay the person it describes. “Internal-model runner” is my shorthand for how often I need a working picture of something before I can use it well. It fits much of my life. I still get to learn a procedure, act before I know every cause, be tired or inconsistent, stop building, change Realm, or find that the archive contradicts me. I do not owe the book a performance of its main character.

CHAPTER 24 — WORDS FOR THE QUESTIONS

After Episode 3, as the compulsion to keep explaining myself finally began to quiet down, I started looking for what researchers in psychology, psychiatry, cybernetics, and computer science called the patterns I had been living.

I did not find these terms before the experiences; I found them afterward, while developing this account. Naming them mattered for a specific reason. When you invent a private vocabulary for your own mind—words like *the Governor*, *twin talk*, *the loop*, *common files*, or *internal simulation*—that vocabulary can make you feel uniquely understood by an AI that repeats it back to you. It can also lock you inside a closed room where no outside evidence can touch your theory.

Translating my private words into the smaller, established questions of formal research became another governor. None of the studies I read examined my life, and none of them proves that the Governor is a single biological mechanism. What they do is break one overwhelming story into separate, testable pieces.

The first question was why building a working picture of a motorcycle, a cabinet joint, or a room felt so natural to me—and why it so easily went too far. In cognitive science, Philip Johnson-Laird's mental-model theory studies how people reason by constructing internal representations of possibilities, and how incomplete representations systematically lead us astray. In mechanical reasoning, Mary Hegarty's work shows that people use piecemeal mental simulation alongside learned rules to infer how physical parts interact. Those ideas gave me a modest, grounded way to talk about countersteering on the Honda or tracing a circuit breaker without claiming that my brain contains a literal physics engine calculating all of reality at once.

Research also had a name for the trap waiting inside that habit. Leonid Rozenblit and Frank Keil called it the *illusion of explanatory depth*: people routinely believe they understand how a familiar device works until they are asked to explain the mechanism step by step. Trying to make a relationship explicit exposes the missing link. At the same time, fluency is not truth. Alfred Korzybski's famous warning in *Science and Sanity*—that the map is not the territory—is the rule the woodshop and the hospital kept enforcing on me. A coherent internal map, or a convincing AI explanation, can agree with itself at every point and still crash into the territory.

In cybernetics, W. Ross Ashby analyzed how complex systems maintain stability through regulation and feedback, showing that a regulator has to match the variety of the disturbances it faces. And when Thomas Nelson and Louis Narens mapped metacognition, they separated *monitoring* a mental process (judging how well your thinking or learning is going) from *controlling* it (deciding to change course or stop), noting that our introspection about our own minds is itself fallible. Adele Diamond's work on executive functions breaks "control" down even further into working memory, cognitive flexibility, and inhibition—the capacity to hold back an action. Calling everything a "Governor" helped me see the family resemblance between a Victrola's spinning weights, a medication adjustment, and a shop safety rule, but executive-function and cybernetic research reminded me that mechanical speed regulation, neural inhibition, and social boundaries are different operations doing different jobs.

A second cluster of research helped me separate my childhood history and ear anatomy from the meanings I had attached to them during psychosis. Dorothy Bishop and Sonia Bishop's 1998 study of "twin language" showed that the phrase covers very different things across families: most often shared immature speech or phonological habits, and much less commonly a separate private language alongside ordinary English. Their work cannot tell me what Colin and I said in our childhood mumbles, only that my memory of "learning English on top of a code" is a retrospective account rather than proof of a distinct linguistic stage.

On the physical side of listening, the American Speech-Language-Hearing Association's guide to tympanometry and the 2015 international consensus statement on Eustachian tube dysfunction (Schilder et al., along with Johns Hopkins clinical guidance) separate middle-ear pressure and mobility from actual hearing sensitivity. Baro-challenge ETD produces fullness and muffled sound under pressure changes, and being able to voluntarily manipulate middle-ear pressure is distinct from how well the inner ear or brain processes speech. Meanwhile, Ediz Sohoglu and colleagues demonstrated top-down predictive integration in speech perception: when listeners already have prior written context that matches degraded speech, the spoken words genuinely sound clearer. That gave me a way to understand why conversations that fit my active mental model felt easier to follow—and why my father's remark at my birthday dinner could immediately change how natural the Victrola sounded—without claiming that prior expectation cures conductive hearing loss. Even the environmental-noise literature, such as the World Health Organization's guidance and Manfred Beutel's Gutenberg Health Study linking chronic noise annoyance to later anxiety, depression, and sleep disturbance, studies ambient noise exposure rather than Bluetooth-versus-cassette fidelity. If I judge how limiting an old audio or cognitive governor was only by how relieved I feel after noticing it, I am reasoning in a circle.

A third set of studies gave me sharper language for the difference between healthy reflection, ADHD restlessness, and a bipolar episode. James Eastwood and colleagues define boredom in terms of attentional engagement—the frustrated desire for satisfying activity when attention cannot find a hold—and a 2026 meta-analytic review by Peter Muris, Henry Otgaar, and Frenk Donkers found a consistent association (though not a simple one-way cause) between ADHD symptoms and boredom proneness. Earlier, Valeria Pironti and colleagues found differences in temporal reproduction tasks among adults with ADHD, offering narrow background for timing questions without explaining the grand, frightening alignment of bodily time I felt during mania.

When reflection turns into a trap, Thomas Ehring and colleagues call it *repetitive negative thinking*: thought that intrudes, repeats, consumes capacity, and struggles to terminate. And when Douglas Hofstadter explored self-reference and "strange loops" in *Gödel, Escher, Bach*, or when computer scientists like Harold Abelson and Gerald Jay Sussman (*Structure and Interpretation of Computer Programs*), the authors of the Scheme standard, and Alan Turing (in his 1936 paper on computable numbers and the halting problem) defined formal recursion and non-terminating processes, they were describing exact mathematical and computational properties. Borrowing "recursion" or "the loop" as a metaphor for how my book ran back through me is useful only if I remember that a human mind is not a Scheme interpreter and Turing's halting theorem is not a clinical diagnosis. As the National Institute of Mental Health and clinical reviews by Virginio Salvi and colleagues make clear, ADHD and Bipolar I disorder with psychosis are clinical conditions assessed through developmental history, sleep, mood, energy, and functioning over time—and as Mark Leary's research on intellectual humility emphasizes, holding a carefully built model of yourself still requires leaving room for that model to be wrong.

Finally, the literature on external memory, human-computer interaction, and choice helped me explain why writing this book and working with AI could both rescue my attention and threaten to swallow it.

In philosophy and psychology, Andy Clark and David Chalmers proposed the *extended mind*, Evan Risko and Sam Gilbert studied *cognitive offloading*, and Benjamin Storm and Sean Stone showed a "saving-enhanced memory" effect: reliably saving one file frees working memory for the next task, unless the person does not trust that the saved file will still be there. Daniel Wegner's *transactive memory* (knowing where information lives across a group or a diary), Edwin Hutchins and James Hollan's *distributed cognition*, and Donald Norman's *Things That Make Us Smart* all ask the same practical design question I was asking with my personal-life ERP: does the external artifact actually serve human judgment, or does it force the human to do extra work to keep the system fed?

When that external partner is a modern language model built on the transformer attention architecture introduced by Ashish Vaswani and colleagues in *Attention Is All You Need*, and connected to files through retrieval-augmented generation (Patrick Lewis et al.), two human-factors problems immediately appear. First, as Susan Brennan and Herbert Clark showed in human conversation, partners form *conceptual pacts*—shared shorthand for complex ideas. With an AI, that shorthand feels like deep mutual understanding even though the model's underlying weights have not permanently changed around my life. Second, as Marcia Johnson's work on *source monitoring* and Martin Conway's work on *reconstructive autobiographical memory* warn, it is dangerously easy to confuse what I originally experienced, what I wanted to be true today, and what an assistant fluently generated in between. Calibrated trust in automation, as John Lee and Katrina See defined it, means matching my reliance on the tool to tested evidence on a bounded task—and separating my own direct self-appraisal from the reflected self-appraisal (Jennifer Pfeifer et al.) I imagine the outside world making of me, while letting narrative identity (Kate McLean et al.) organize my story without inventing facts.

That brings the research back to the choice problem at the end of this book. Barry Schwartz distinguished *maximizing* (exhaustively comparing every option, as I used to do on Spotify) from *satisficing* (choosing what is good enough and letting the song play). While Benjamin Scheibehenne's 2010 meta-analysis showed that simply having more options does not automatically overload people in every setting, Alexander Chernev's 2015 meta-analysis showed that choice overload spikes when options are complex, preferences are uncertain, and the decision goal is unclear.

Alongside the World Health Organization's reminder that mental health is shaped by biology, family, and hard structural realities like money and housing, those findings sharpen my final claim. AI did not cause my bipolar disorder, and insight alone does not cure it. What AI did was collapse the cost of starting complex, open-ended branches of work while leaving the human brain with the full cost of choosing, stopping, and living with the result. Giving the questions their proper names kept my own theory from becoming one more ungoverned loop.

PART VI — THE WORK MEETS OTHER PEOPLE

CHAPTER 25 — THE HUMAN INTERACTION HAS MORE THAN ONE HUMAN

I had expected Colin to be shocked. At the marina, I had kept visiting so he would think I still worked there. I could explain how frightened I had been. I had also been trying to decide what my brother would know.

When I returned to that exchange while working on the book, I remembered telling him and hearing an answer I had not predicted: he already knew, and he wanted me to communicate more openly. His reply changed the story I could tell. Remembering it did not give me his side of the exchange.

There were parts of Colin's life I wanted to see for myself. Before the Mötley Crüe concert in July, I was excited that my twin was running the show as crew chief. Someone else had designed it. I arrived with an all-access pass and kept encountering different answers about where it allowed me to go. One guard let me backstage. Another said I needed an escort. The next morning, those conflicting instructions filled my account of the night.

What I remember now is that Mötley Crüe was good and that my brother and I got angry with each other over our interpretations of the rules.

We could share an evening and come away with different accounts. I wanted room for both. I could correct something I had believed without pretending I had never believed it. One of us might have a fact wrong; we might still have different things to say after the fact was settled.

There were earlier moments I could not recover at all. Asked for one particular childhood exchange in our language, I could not supply it. I thought my mother would be most likely to remember, but we had not collected her answer. I wanted to hear what she remembered hearing. A language model could readily write a scene that felt right for two boys like us. I could not use an invented exchange to fill that gap.

The archive was mine in the ordinary sense that I had exported the files and taken part in most of the conversations. The lives in it reached beyond mine. Agreeing that a passage was fair would not settle every later use of it. Agreeing to an interview would not necessarily mean agreeing to release the whole recording. Sometimes a source could limit what I honestly said without becoming something I could share with everyone. A more vivid scene was not always mine to give.

Other people might offer their account later, disagree with mine, or never answer for the book. I would rather leave that space open than write their reply out of what I feared they would say. They did not owe me an ending on my schedule.

In August 2026, I was asking an assistant whether I should randomly propose to a close friend. We were not dating then. I described more than a year of close friendship, with periods of dating along the way. I wanted someone who cared about me to commit to joining my weird life. I kept trying to work out what romance or marriage would add to that description.

We had met about two years earlier, when she was going through a bipolar episode of her own. That history helped us understand each other. The grand proposal ideas were part of my episode 3, and she was one of the people keeping me grounded through it.

I could imagine the presentation in detail. In one possibility, I would ask her to get documents from Printer, the lizard, and have a ring ready while she read a laminated engineering change notice. In another, she would turn over a record on the Victrola and find a note underneath. I worked through those possibilities with the assistant while still asking whether I wanted to get married at all.

I wanted a question that would reveal her view. Then I rejected one suggestion because it was too direct and would give her hope that I was thinking of something. Later, another wording gave away too much. I said I was trying to keep her guessing. I wanted to learn what she wanted while deciding how much of my own uncertainty she could see.

I reported telling her, “I mean yes we’re describing similar relationships, but still differ in some of the particular definition’s”.

One of her replies, as I recorded it, was: “I was confused I thought maybe that’s where u were going”.

I then reported adding that I was just thinking out loud.

Her confusion slowed my loop down. While I designed a gesture, she was trying to understand what I meant. We were still working out what either of us wanted. The friendship could hold that confusion without being blown apart by it.

She did not panic when I started spinning up big ideas. Because she had been through an episode herself, I did not feel I had to hide everything or perform normal for her. She helped with physical things in the house, including the breaker mapping and the animals, while my head was running too fast. She stayed in contact while I was in and out of the hospital during those five weeks.

There were still circuits to identify and animals to care for. She could be unsure what I meant and remain in my life.

By October 2026, I was describing roughly two years of struggling with romance after leaving a relationship that had lasted three years. I kept coming back to change and unpredictability when I tried to explain what I wanted. I said the loop got bored and needed someone as unpredictable as it was. That gave me words for my own need. I would still have to find out what it was like for another person to be in a relationship with me.

I had tried dating her a few times. Each time, the internal simulation seemed to become bored. I struggled to experience intimacy without making it a test, or something to observe and model later. That difficulty existed beside the elaborate proposals. I could imagine the presentation long before I knew what I wanted it to change.

Where things land with her is still to be determined. Her part in episode 3 has already happened. I can give that care its place without making it a promise about what our relationship will become.

In January, I went on a date and had a good conversation over dinner. I told her I was going to a birthday party afterward. She asked to come, and I said yes. She drove us there in her car. We were sitting with everyone on the couch when she got up, told me she would be back, and never came back.

We all laughed it off. Looking back, I call it a very Ethan-like encounter. My friend drove me back to my car.

The next day, I said she had not texted and had removed me on the dating app. I was irritated as I wrote about it, though I had laughed with everyone the night before. I do not know why she left. My friend's ride back to my car remains part of the story without that answer.

Sometimes I had to stop an assistant before it got as far as imagining what another person meant. In one exchange, I wrote:

“Why are you using partner instead of roomate”

One word had brought intimacy, obligations, and a history into the conversation that I had not supplied. The next answer could fit that imagined relationship perfectly and still be wrong about the person living in my house.

On May 10, 2026, I sent a screenshot of an exchange about an air conditioner and asked, “Was he really considering leaving the water on the floor.” I had already given the assistant my interpretation of what it was about to read.

I also wrote, “Im not dealing with a leaky ac, thats not my problem lol.” Then I had to explain: “Well my ac, his room, i was nice and let him borrow it, i told him there was a possibility it leaks.”

The unit was mine, the room was his, and the floor belonged to the house. I wanted the water dealt with without taking on the entire repair. In the screenshot, his words were still there beside the meaning I had given them.

He had asked whether I was there and said he might need the back to the AC unit because water was leaking onto the floor. “I don’t think i have that panel,” I replied. Then: “Maybe pull it out and dump the water and it will prolly be good for another few days.”

“It’s just I don’t want the water to damage the floor,” he wrote.

“Well yeah, turn it off if it’s dripping lol,” I replied.

He asked whether he should put something there to soak up the water. I wrote, “Yeah lmao, you can’t leave standing water on a wood floor.”

His last message in the screenshot was, “Yea I’ll put a towel there it’s not a lot of water.”

I do not know what happened afterward. He might be emptying it occasionally, or it might have stopped leaking. For now, I can leave that question open.

An earlier housing dispute had been harder to leave at the point where I wanted the next practical step. In March 2025, I was asking about unpaid rent and getting a tenant to leave. I said I was filling in a complaint. I asked for a concise statement of facts, corrected what had been paid, asked about interest, and asked to include court costs. One box wanted a dollar amount. I was trying to fit the dispute into the spaces on the form.

The trouble had begun on her first day in the house. She was doing hard drugs in the dining room. Looking back, I think letting her move in was a mistake from day one. I gave her too much credit.

I had wanted to give someone a chance. I assumed people were acting in good faith until the physical evidence became impossible to ignore. In the shop, I wanted what a machine actually did to correct my idea of it. At home, I had kept giving this person credit beyond what was happening in the house. I had to account for that part of my own judgment too.

Then I asked whether it was okay to yell at a tenant. My next message was: “its okay i already yelled at them”.

Less than a minute later, I asked, “should i address my yelling”.

I pasted her reply. She said there had been no need to scream at her. She also alleged that I had failed to tell her about an infestation before she signed the lease. I responded to the assistant with questions about disclosure and what counted as an infestation. I said she had texted me about the problem and that I had remedied it as soon as possible.

The rent was still in dispute. Now her objection to how I had treated her sat beside it. I asked whether yelling was any better because eviction had already been filed, then how it could be retaliation or harassment. I was looking for a way the stage of the proceedings might change what I had done.

Later I said she had told me she would be out by the next day. My next question was how that would change the eviction proceedings.

She did move out.

When I read these exchanges again, I want to hear more than the words I chose to paste. When original audio exists, I want to keep it beside the transcript and my later account. A pause, tone, emotion, or pronunciation may change what I hear in a sentence. When memories differ, I want to know whose they are and when they were given. Keeping both does not mean I can rely on them equally. Replacing the old one would hide how the account changed. I still have to make a judgment.

I could even be credited with words I had only forwarded. In September, I explained that I usually forwarded an assistant's entire response between tasks. To the next conversation, it could look like a message from me. The assistant's explanation could become my supposed instruction, or something I was taken to have witnessed.

I asked for handoffs that said what I was asking now, what belonged to the background, what had been checked or completed, and what still needed an answer or a next step. Credentials and unrelated intimate details could stay out. I wanted the tools to do that consistently. I could not assume they would.

I needed the long record and an account short enough to use. The original let me check what had been carried forward; the shorter version let me return to the work without reading everything again. The difficult part was keeping why a correction mattered. Saving the sentence I had accepted was easier than preserving what it should change the next time.

A person could have helped me understand an event without agreeing to have their words released to every reader or used to train another system. Taking part in my life did not mean volunteering to become a permanent example in my theory of AI. I wanted them to be able to refuse a use of their material, even if that complicated the book I hoped to publish.

The help I want is with finding, arranging, and getting back through what happened. I still have to live with the people, listen when they answer, and change the account when life changes. An assistant can help me remember them. It cannot take their turn to speak.

CHAPTER 26 — THE SAME ROOM IS NOT THE SAME EXPERIENCE

On May 13, 2026, I was describing how well I and my four roommates got along. I had found them through Facebook Marketplace and Craigslist. There were four dogs, eleven reptiles, varying work schedules, and what I called “all my other craziness.” I sent screenshots of the household chat and wrote, “Look it just somehow works.”

In an earlier exchange, a roommate had asked, “Who has clothes in the washer I need to wash clothes sometime today.”

“Think their mine, go ahead and flip it if you want or I'll do it later,” I answered.

“U mean put ur clothes in the dryer,” he asked. I said, “Yeah.”

Then he wrote, “Dryer is full too so I gotta wait.”

“That's mine too,” I said.

“Oh ok,” he replied. “I'll put in basket.” Then: “And then I'll put the other clothes in dryer.”

He had wanted to wash his own clothes. Both loads in his way were mine. First he had to ask what I meant by flip. Then he found the full dryer and worked out what to do with that too. He got on with it without needing a theory of why I had left the clothes there.

Later on May 13, I said a roommate had cleaned and rearranged the whole pantry without being asked. Part of how the house worked was that other people got on with things. They did not wait for me to explain the arrangement before changing it.

In June 2026, I was on the pet-room sofa eating dinner while someone used the upstairs bathroom. A roommate came in and asked whether we had a bathroom downstairs.

He had lived there for over a month. He had seen the bathrooms on the tour. I could not understand how he still needed to ask. When I told the assistant about it, I went from the bathroom to how much time he spent in his room. I wanted roommates who engaged with me and with everyone else in the house.

I said I wanted to get him out of his shell and help him become a better version of himself. Living there ought to be good for him socially. I was already imagining what the house could do for his life. He had come to the sofa to find somewhere else to use the bathroom.

An August evening put more of the household inside one of my projects. I was installing a system to monitor the electricity, and I needed to know what each breaker actually controlled. I stood at the panel. Three roommates moved through the house as runners. Another helper stayed in the basement taking notes and talking with them on the call. Their reports helped us decide which breaker to flip next.

The helper beside me was the friend I had dated, the same person who helped keep me grounded through episode 3. That evening her care had a practical shape: notes, a call, and other people's voices reaching us in the basement while the roommates checked what still had power.

The circuits did not divide neatly into rooms. One breaker reached the basement light, part of a bedroom, the pet-room ceiling fan, reptile enclosures, and the Wi-Fi equipment. Another included the dining-room ceiling light, more enclosures, the sound system, and the television. We still had an unknown breaker I thought might once have served the hot tub. I could not trace its wire. It stayed uncertain on the list beside the circuits we had identified.

The full installation took five hours, from about seven at night to midnight. Later I wondered why my roommates and the other helper had put up with my bullshit for that long. I remember making a half threat that they needed to help or be left in darkness. The joke was that they did not have much choice. That was part of how I drew them into it, and the work took their evening as well as mine.

I also offered rent credit to everyone helping. I could flip the switch at the panel, but I needed their time in the rooms to know what I had done. Each report brought back a part of the house I could not identify from the breaker alone.

Even after we had made the list, a label could lose something. I had given the assistant the animals' names so it would know which enclosures were on which circuit. Then I had to ask it not to make the labels just about reptiles. I needed to know what else would go off. Whoever was using the television had a reason to care about the same switch as the person caring for an animal.

I could draw a room accurately and still leave out much of what it asked of the people living in it. A doorway, speaker, window, and work surface could all be in the right place on the drawing. The measurements would not tell me how hard it was for someone to get through the door, follow a conversation, or work there for an hour.

I knew that through my own hearing. The room could stay where it was while my experience of it changed. I can change the pressure in my ears and produce a selective conductive reduction in hearing that stays until I pop them. My ENT has diagnosed baro-challenge Eustachian tube dysfunction. I had believed my hearing was fine while describing difficulties that belief did not explain away. I can also find speech clearer while having conductive hearing loss. A description of where the speakers sit would leave out the condition of my hearing and the effort of following the person speaking.

I could recognize the same missing effort in other ordinary things. Someone could make it through a building and be exhausted by the route. An instruction could be understandable and require extra work every time it was used. A conversation could seem easy because one person kept repairing the gaps. Reaching the end did not tell me how much the person had spent getting there.

I could ask, and give someone time to describe what had changed. They might find it helpful, distracting, strange, or harder. I could not tell them how much effort it took or cancel their answer with my explanation of why the change ought to help.

Doorways, sound levels, light, and machinery still needed measuring. Some of those measurements were essential for safety and access. Afterward, someone would have to live in the room. They could compensate for a difficulty, resist the arrangement, or change what they wanted. I could not find their best state by adjusting them as though they were another machine in the drawing.

I had learned to participate without knowing what it might be costing me. The ways I had found around a difficulty might also have kept me from recognizing it. Getting used to something does not make the arrangement good. I wanted to know what the effort cost, whether the person had chosen it, and whether something could change that would make life better.

That could begin with a small comparison. Try a setting, return to the previous one, and say which helps with the task. A setting that helped with speech might grow tiring over an hour. One that gave music a sense of space might make an instruction harder to locate. I did not need one answer for all of listening before I could ask, “Does this help you understand these words in this situation?”

Two people could sit in the same room and disagree about how it felt. Neither would have to fail a test for the difference to matter. I wanted their answers to be part of the work. I would still need other ways to check what caused a change, the pressure inside an ear, or whether a device was safe. I could leave the explanation open without leaving the person unheard.

The same was true of the book. I could organize the account beautifully and still describe myself badly.

There was always someone in here making sense of things, even when other people could not understand how.

A fluent explanation from an assistant could make my earlier self look like someone waiting for the words to arrive. I had already been living the difficulty and trying to be understood. In the rough messages, I could see myself judging the reply and asking for something more. I had a purpose before the assistant found words for it.

In May 2023, I submitted a story for comments and editing. I was trying to explain language and identity through being a twin. Some of the explanations I offered went beyond what I could establish. I was also doing something ordinary: writing about my life and asking how to make it understandable. The attempt mattered before I had every cause right.

I wanted to keep that effort visible as I revised. Saving only the polished story would lose what I had been trying to do. Keeping every old explanation as fact would leave me no room to change my mind. I needed the earlier attempt and the freedom to write differently now.

I had been making judgments in the April 2025 enclosure conversation too. I rejected proposed panel materials because waterproofing mattered for the reptiles. I wanted the joints and screw locations considered. I narrowed the work to standard models first. Customization could come later; the visualizer could wait. Another option was easy to describe. I would have to build it, maintain it, and explain it to a customer. Those decisions were mine before the assistant put them in order, and they still had to work in an actual enclosure.

I want to be understood before I have finished explaining myself. The old draft, the engineering correction, and the newer hearing account can remain unfinished. A reader can question my explanation and recognize the person trying to give it. I can hear another answer, or make an ordinary choice, without first rebuilding the whole account.

If I say the room is different from the drawing, that answer belongs in the next version. I am still the one who has to live there.

PART VII — THE ATTENTION I CAN GIVE BACK

CHAPTER 27 — THE REPRESENTATION HAS TO EARN ITS SPACE

“Why are you questioning the dysfunction the professional ENT told me I have?”

I had already supplied the diagnosis: baro-challenge Eustachian tube dysfunction. I could change the pressure in my ears voluntarily and produce conductive hearing attenuation. The muffling stayed until I popped my ears to clear it. I was trying to think further about what I could do. The assistant kept returning to generic caveats about whether that established dysfunction.

I still had questions about the precise mechanism. The ENT had diagnosed the dysfunction. Not knowing exactly how it worked did not mean I had no diagnosis.

The assistant acknowledged that it had started in the wrong place. It could accept the diagnosis and still ask what remained uncertain about the mechanism. That changed what the next answer needed to do.

I had spent another turn restoring a fact before I could get on with the question. The careful-sounding language had made the reply less faithful to what I had told it. This time the caution needed correcting.

I wanted the correction to spare me that work the next time. A fact could sit in a conversation without doing anything to the answer. If returning meant rebuilding the starting point every time, all that saved context would become another thing for me to carry.

I had met the same cost in more ordinary work. In the smart-speaker conversation, I kept having to restore requirements we had already discussed. I needed the assistant to remember why a company had failed the criteria for that project. Saving only its name could turn one particular rejection into a permanent rule against it.

With the Fusion promise, we had needed the model before the cutting layout, and the assistant had lacked the tools to make it. I wanted it to remember that sequence without treating the missing tool as a permanent fact about every future system. The next system might have the capability. What mattered was the reason I had needed to correct this one.

The hard-drive comparison was useful to me only if I could keep more of the record without having to think about more of it all the time. I imagined AI as the smart wiring: something that could find the relevant part and bring it within reach through the text and screens I used. I still needed enough of the original to judge the answer. I did not want another folder of apparent knowledge that I had to read and manage before it could help.

After useful work separating customer export information from private pricing, another conversation gave me a list of things I was already doing. I rejected it. The help that had narrowed one job was now adding a demand without finding out where the work stood. If I had already made the decision, asking me to make it again did not return any attention.

An ordinary mechanical question might need a dimension, an approach I had rejected, or a working preference. It did not need my whole life brought back with it. In September 2026, I called what I wanted maximum useful governing context. Useful mattered as much as maximum. I asked for the book to be primary context when I explicitly invoked it. Reading it would not give a model a credential to judge me, diagnose me, fix my identity, or use unrelated personal information. The account remained mine.

Keeping the record could change what I understood about myself. My earlier hearing descriptions showed difficulties I had been describing while believing my hearing was fine. Once I separated awareness from improvement, I had a different interval to understand. Repeating my pressure explanation did not give me independent measurements. Even as the wording improved, much of it still came from one person trying to understand his life: me.

By October, episode 3 was still settling, with several clinicians supporting me. Speech and ordinary sound seemed clearer and more direct. I was also reporting conductive hearing loss from a test, with another test forthcoming.

On October 18, as I read more of the book, I said the noise in my ears was quieter. Asked what I meant, I said I noticed the hearing loss less. The question helped me find more accurate words for what I was noticing; it had not measured a new change in my hearing. I could keep the earlier description and let my answer change how I used it.

Earlier that month, on October 2, I had asked for a reader edition that would not fall below the raw text's word or character count. I called the raw version a survival layer. I said I felt I was living in two realities at once and needed one continuous reference so my mind could keep up.

A shorter book might have been easier for a new reader and lost something I still needed to find. Keeping every passage could ask more of the reader than it gave back. I could specify a length because I felt the need to keep those connections. A word count could not tell me which ones mattered.

Preserving the earlier versions did not mean leaving a mistake alone. I needed room to separate awareness and improvement after putting them together incorrectly. Another pass earned its space when it helped me understand something, see a cost, or do the next thing differently. I still had to decide what belonged in this edition, knowing a reader might find a return unhelpful. The fuller record could remain available when I needed to go back.

When I was asked what made another pass useful, I said the book made more sense in my current state. When it began making less sense, I had a reason to reconsider it. Those were clues about how thinking felt. Feeling clear could not tell me whether my explanation was right; feeling confused could not tell me by itself where I had gone wrong. I still had to check the particular thing I was trying to say.

I could keep searching and asking long after the next ordinary decision needed to be made. I wanted a reason for the next question. A mistaken preference ought to be easy to change; a claim with more resting on it might need a source checked first. I did not want complete certainty to become the admission price for getting on with my day. Keeping everything in view at once could make the record harder to use even while every part remained available.

On an earlier trip to Costa Rica, I had come from San José through Monteverde and around Lake Arenal to La Fortuna. I planned to climb Cerro Chato and then keep moving toward the beach. I did climb Cerro Chato. The rest of the trip changed after a guide gave me his friend's number so I could buy weed.

I went to their house and got stoned with them. While I was there, they invited me to go up Arenal. I stayed in La Fortuna for pretty much the rest of the trip, waiting to go with them. There was plenty to enjoy in and around town while I waited. The beach stopped being the next place I was arranging everything around.

We left at about ten at night, as I remember it. My phone was dead for nearly the whole hike. Afterward I had only a couple of photographs, with none from the upper crater. Much later, trying to describe the place, I used a photograph someone else had taken because it showed the area more clearly than anything I had kept.

We napped for a few hours in the lower crater, in the warmth of the gases. Higher up, the vegetation disappeared completely. What remained underfoot was mostly gravel. Smoke was still billowing from the top. I remember an old stone cabin there. I had gone from an inhabited house where somebody invited me on a hike to this other structure on the volcano, and I had almost no pictures of that part.

I remember the hike as epic. I call it an illegal hike, and the invitation was spur-of-the-moment. Someone else's plan had become something I wanted enough to change my own. The story I kept returning to began at their house, after a meeting I had never put on the itinerary.

When I was planning another trip in July 2026, I said that if Arenal was my favorite kind of story, I could not guarantee that the next trip would produce one like it. I could choose where to go and leave room to stay. I could not schedule that invitation.

There were things I wanted to leave room for at home too. A question could be worth saving without being worth the sleep, care, pleasure, or promise to someone that it was about to displace. The archive still had mistakes. If it stopped helping me decide or find the evidence I needed, I wanted to be able to make the account I worked from smaller or different.

For someone else to use any of this, they would need to be able to find the claim that concerned them and tell me where it failed. I wanted the record to keep what the person had been trying to do, the answer that had been attempted, and the correction that mattered afterward. If the work failed or remained unfinished, that belonged there too. A shorter account could carry those things forward without improving the appearance of the old exchange at the expense of the next one.

On October 12, 2026, I saw a way the work itself could take less of me at once. I had asked for background work on the next edition while I answered further interview questions. The sources, drafts, questions, and decisions did not all have to wait in my attention together. Enough could remain in place for me to return while another part continued.

That help had costs as well: producing the account, correcting it, checking who could reach it, and maintaining the record. I like explaining things enough that a method suited to me could easily ask too much of someone else. It could ask too much of me too. If the upkeep consumed the attention I wanted back, storing more records and completing more tasks would not rescue it.

The book's home is realmbyethan.com. I want it free because I want it to help the world. I also want somewhere to return when the account is too much to hold in mind. A reader might use the engineering lesson and leave the hearing explanation alone. The privacy boundary might be what matters to them. They might replace the word Governor, or recognize the method already and want only an example.

After publishing, I still have to wait to learn whether anyone used it or got a better result. The possibility of helping has sometimes made another revision feel urgent. I cannot measure the effect by how urgently I feel it, or turn every delay into something I have withheld from the future. Other people are doing their own work while I write. If reading this helps them make something better, what they make will belong to them.

CHAPTER 28 — THE TITLE IS SMALL; THE THING IT POINTS TO IS NOT

On October 18, I am at home among the pets, the woodshop, the house, and everything that still needs doing. My mind feels quiet. I do not think it has been this quiet since before my first bipolar episode.

I had been comparing three Victrolas with the hi-fi stereo system. Each sounded different, and I wanted to understand why. I was getting somewhere, but every answer gave me another comparison to make. Soon I was thinking well beyond what the machines in front of me could settle.

I turned the Victrolas off.

I did not have enough money to fix all three immediately. Understanding their differences could not buy the repairs. I could keep following the question, but the budget would still be there when I came back.

Then I made an easy choice on Spotify. Working with AI on other things had helped free some attention, and I had enough of it left to choose my own music. Before, I could spend ages trying to consider every possibility.

Somebody else had already made the song. Their choices were part of what I wanted to hear. I could receive the result without first reproducing every decision that went into it. A title was enough to bring the work within reach; I did not have to finish explaining its place in my life before pressing play.

There was plenty to do around me, but nothing currently seemed interesting to the loop in my head. I came back to the conversation and the book to put that loop somewhere I could see it.

I said I knew how to be bored again. The boredom itself felt familiar, though. I become bored when I encounter a governor, when I reach something I cannot carry further in the way I want. The repair money was part of that day. The limit could keep me from running away with a comparison and still frustrate me.

Writing sat between want and need. I wanted to work on the book because I was bored. I also felt a need to keep gathering the thoughts that otherwise ran through my head. With my hearing issues, this had become the best way I knew to find my bearings. The page gave those thoughts somewhere to stay.

I could return to the writing without making it the only thing I did. I also knew how easily trying to understand my thinking could keep that thinking going.

Later, my family came over for my birthday dinner.

My father made a remark about the sound. As I remember it, he said that Victrolas do not sound great to anyone.

The Victrola sounded different to me almost immediately. It sounded much more natural. I could hear its sound as spatially right, and I became less engrossed in it. The machine had not changed. My father had given me another way to hear it.

I had been expecting more from the Victrolas than they were giving me. His judgment mattered because I knew him; it arrived with a whole history of understanding who was speaking. I could let the machine sound like itself without making another repair first.

I had not noticed the expectation until his comment disturbed it. I began to think of that expectation as another governor. The word already belonged to the weights controlling a Victrola's speed and to the limits I was trying to understand in my hearing. Here it pointed to something I had brought to the listening.

I could hear the difference before I could explain it. How much came from expectation, attention, my hearing, or the evening itself was harder to separate. What I knew was that I wanted less from the machine and could enjoy more of what it gave me.

I began noticing other things about listening. Depending on what I focused on, almost all audio could seem to be at the same volume to my ears. Even describing what I heard had become a question about what I was paying attention to.

There were plenty of actual equipment questions left. I had listened to small speakers and larger speakers together, liked the combined sound, and noticed less clarity. I was especially interested in when the sound from each speaker reached me. I was still planning a further surround arrangement and processing based on measurements of the room. I could hear something I wanted to investigate without having the explanation finished.

My father's comment reached beyond antique machines. It changed how I thought about my hearing and attention elsewhere. This is what I mean when I say my understanding is connected: something said about one object can change the way I understand myself in other parts of my life.

I described my model as forever changed by his statement, whether I wanted it to affect me or not. That was the phrase I found for how lasting it felt. The change had already mattered before I brought it back to the book.

The truck gave me another direction to consider.

My 2003 Silverado still had its factory radio, with the CD slot above the cassette deck. Eject would not bring anything out. I was pretty sure a disc was stuck inside. I wanted the CD player working again, and I was trying to improve the speakers without spending money I needed for that repair.

Bonnie Tyler's “Going Through the Motions” sounded much better to me on cassette than through Bluetooth. I kept returning to that difference. I had tried the Dolby processing on my phone both ways, on and off. I still did not like the Bluetooth sound.

At first I wondered about interference. Then I wondered whether the replacement speaker was the problem. The four-ohm Power Acoustik sounded quieter and less full to me than the original two-ohm Bose. I had been looking for better woofers while trying to keep the tweeters I already liked. Each possible improvement seemed to bring another part I might have to buy.

I considered a matched set, then decided I should save the money for the CD player and a tape-and-CD setup for home. I ordered the woofers. I also ordered a wired cassette adapter, due the next day. The cassette slot that still worked offered a way to try the phone through a different connection.

The Bluetooth device called itself BlueFusion. It did not connect through an FM station. That name led us to a Scosche manual for a GM interface that appeared to fit the truck. Its route began with the BAND button and the XM selection. The phone's music came through an input the radio presented as satellite radio.

The TUNE knob led to Settings; SEEK forward selected it. While BlueFusion was idle, the volume control there changed hands-free call volume. While a song was streaming, it changed the music level. The same menu, the same buttons, a different adjustment. The manual also warned that some devices could distort at full volume and might need the level turned down.

To reach the setting I cared about, the music had to be playing.

I had been looking at the radio and trying to decide what was wrong with the sound. What that control adjusted depended on whether music was playing. Nothing about the word Settings told me the difference.

Without the cassette, I might have kept wondering which speaker to buy. Someone else might have accepted the sound as all the truck could give them. The buttons were in front of us either way.

How would anyone know all that when they bought a car?

That was what I asked. I was frustrated enough to talk about people's laziness, about becoming accustomed to shitty audio, about what that acceptance might be doing to mental health and life in general. But I had just needed help finding the instruction myself. The next owner could be attentive and still have no reason to know which version of Settings they had reached.

I knew this problem from the shop. A person leaves and the next person gets the equipment, the files, the half-explained procedure. Something essential stays with the person who knew it. I had wanted the Makerspace and the CV utility to make useful knowledge easier to carry forward. Here the gap was small enough to fit on a radio display. The control could exist for years and still be absent from the way somebody used the truck.

I wanted to check that music level, then hear the same phone track through the wired cassette adapter. Finding a setting gave me something to try. It had not yet given me sound I liked.

The comparison with my father stayed with me. At dinner, his remark had helped me ask less of an antique machine. In the truck, the cassette made me question whether I had been asking too little of what was already there. The two experiences pulled my expectations in different directions. I could listen before deciding which piece of equipment deserved my confidence.

What interested me was the reference I had been using without seeing it.

I wanted to go further than saying the song sounded better. I thought people could feel better and live better when they understood more precisely what they were listening to and what had been taking their attention. A familiar experience might have been shaped by something they could identify. Learning that could change their understanding of themselves as well as their surroundings.

I asked the AI to question that idea before putting it into the book. It asked about a cleaner sound arriving without the listener knowing anything had changed. My answer went back to awareness. I was interested in what happened when the person recognized the inputs they had been living with. The time it took would depend on the person. At first, I said the relief should last forever once they understood.

The next question made me separate two things I had put together: knowing what had shaped an experience, and continuing to feel the first relief of finding out. I meant both, but I also said the relief might fade. It depended on which governor the understanding helped resolve.

How limiting was it? That seemed to me to matter. Something that had constrained a large part of a person's life could make a larger difference when they finally recognized it. Yet they might not know how limiting it had been until that recognition arrived. Living with a limit gives you familiarity with its effects; it does not necessarily give you a view of life beyond it.

I wanted to know what that recognition would leave behind in someone's ordinary days. What became easier? What remained difficult after the first relief passed? I knew from the Victrolas that understanding could arrive while the repair money was still missing. Recognizing a limit could matter even when a repair, another person, or help I still needed stood between understanding it and changing life around it.

The thought belonged beside my father's remark because I had already felt an ordinary object change its place in my understanding. I wanted to know how far that kind of change could reach. I did not need the truck stereo to explain mental illness for a song I loved to matter. The difference I heard was enough reason to care about the next listen.

I came back and asked to go through the controls again. After that, my answer was still that Bluetooth sounded worse to me than the tape. I would wait for the adapter. The instructions were useful, but the song still had to sound good.

For the moment, I had a CD player that still needed attention, woofers on order, and an adapter coming. I was waiting to try a few familiar songs through it and find out whether I liked listening that way. I wanted a setup I could leave alone for a while. I did not want every drive to become another question about the dashboard.

I do not think there is a fixed stopping point for the book. I turn elsewhere when I feel grounded enough and something else catches my interest. Enough for now does not tell me I will never need to return.

Not knowing what to do next feels good. I do not get that often enough.

CHAPTER 29 — THE HUMAN KEEPS THE KEYS

I began this book because I needed to make sense of my life. The practice grew out of smaller jobs: correcting a connection, comparing a claim with a file, asking what happened before a message, and recognizing when the next answer had to come from the shop or the animal in front of me.

Eventually I called that practice *model governance*. I meant keeping active control over an account—whether that account is running in my own head or across an AI conversation: where it came from, who can correct it, what it allows me to do, and whether I can put it down and walk away. I needed those questions because I had sometimes been the unreliable participant in the arrangement. My good intentions alone were no guarantee that my simulation was aligned with reality.

You do not need my history with twin talk, ear pressure, or bipolar disorder to test this practice on a manageable problem of your own.

The Honda PC800 gives the smallest possible illustration of the loop. When I ground a tinned-copper electrical ring terminal flat to replace the lost coolant drain-bolt washer in Chapter 3, I did not need a complete theory of the motorcycle to close that step. I needed to know whether the faces were flat, whether the copper would compress, and whether the joint would stay dry through a full heat cycle. The engine warmed, the radiator fan came on, the fan cycled off, and the washer stayed dry. I could record that observation and stop reopening the drain-bolt question without a new reason. The DMV trouble and the rush-hour turn were still waiting, because one dry washer does not certify a whole motorcycle.

When a problem starts spinning in your head—or across ten open tabs with an AI—the small loop of model governance has four steps:

First, put the working explanation and the single next action outside your head in a few plain words or a sketch. Somewhere inside what you think is happening is one load-bearing assumption. Ask: *What do I expect to happen, what habit or rule makes that seem likely, and which specific piece, if wrong, would change what I do next?* A longer, more polished essay from an AI may feel reassuring while giving you zero new reasons to trust the premise.

Second, require something outside the explanation to answer it. In the shop, that is a meter reading, a test cut, or a fan cycle. In software, it is an actual file inspected in the program that has to use it, not a filename promised in a chat. In a relationship or a workplace, it is asking the other person and letting them speak in words you did not script for them. And as my father's remark about the Victrolas showed me at my birthday dinner, it also means asking: *What unstated reference point am I comparing this experience against?*

Third, separate wanting something done from having done it. An appointment in a plan is not booked. A proposed purchase has not spent the money. A care checklist on a screen has not siphoned the dirty water out of the monitor lizard's pool. When reality or another person contradicts your model, update the specific part that failed and keep a short note of *why* it changed so you do not have to re-run the same argument next week.

Fourth, decide what is enough for today before you start the next pass. Ask what still must be checked before acting, what can safely remain unknown, and what another hour of optimization will cost the rest of your life—your sleep, your family, the customer waiting on a finish, or the water bowls in the next room. Total certainty is never the admission price for getting on with your day. And when mood, sleep, or contact with reality itself begins to slip, more rounds of explanation with an AI cannot replace qualified human clinical care.

On October 12, 2026, while coordinating an earlier edition of this book, I put three specific predictions on the record so that the next five years could test them and prove me wrong where they fail:

1. *Bounded execution without babysitting:* Assistants will get better at holding a limited piece of work—its goal, constraints, loose ends, and definition of done—so that fewer half-finished jobs bounce back to the human for constant cleanup.
2. *Making tacit craft usable:* Makers, machinists, and animal keepers will increasingly use AI not to replace their hands, but to translate unspoken physical intuition into procedures another person can actually follow, test, and correct at the workbench.
3. *A governed personal record (a "personal-life ERP"):* People will build connected personal archives—linking projects, health, finances, and care obligations—that bring back the exact prior correction that matters for a decision without dragging unrelated private history into view or overwriting what another caretaker observed.

Even if every one of those tools arrives, the human still keeps the keys. A better context server can hold the files, and an interview with family or clinicians can add perspectives I missed, but no system can decide for me what deserves my attention next.

CHAPTER 30 — CHOOSING WHAT BECOMES WORK

I had been back at a Victrola at five in the morning. In describing that day, I also said the Cabinet Vision utility was slowly starting to look like a product-data-management system. Both projects had given me real questions. Each answer showed me somewhere else I could go.

The utility could inspect a manufacturing file, show a mismatch, and help the person responsible correct it. I could also see it growing into something that managed the product, its revisions, its routes through manufacturing, and the decisions connecting all of them. Once I could see those connections, expanding the tool made sense. Living with the larger system would ask something different of me.

The door-material problem was a good reason to want it. A choice on one screen became a thickness elsewhere. A report could look right while parts reached manufacturing in the wrong group. Someone had to recognize the exception and fix what came out. I wanted the knowledge to remain available after that person left the room. I could see how a larger tool might keep it there. I still had to choose how much of the tool to build.

I had already described a smaller job for it. People would make material and part decisions in the software responsible for them. The utility would inspect the exported file and offer correction controls for exceptions. It would need to understand enough to notice a problem, with a boundary around the decisions it took over. I had a reason for keeping that boundary, even while I could see useful work beyond it.

I did not want the next person to rediscover the same hidden rule. I wanted a revision to stay connected to the parts it had produced. Each wish could justify another feature. The trouble was that a reason to add the feature could arrive before I had a reason to take on the whole system it belonged to.

Realm had been growing through the same kind of attention. I followed an enclosure into materials, joints, prices, manufacturing, shipping, assembly, and care because each affected whether the enclosure would work. Those questions could improve the object. They could also put a factory and a continuing care system behind it. Nothing had to be imaginary for the combined responsibility to become more than I could carry.

I think technology has expanded our choices faster than we have learned how to handle them, and that the gap contributes to mental-health problems. I come to that hypothesis through experiences like these. My life cannot tell me how much of someone else's distress comes from the same gap. It gives me a place to look closely at what I mean before I try to speak for the world.

I used to get stopped while writing code, learning an interface, or carrying work from one tool into another. AI removed some of that resistance. A branch could keep moving while I looked elsewhere. An improvement I could describe might become files before I had decided how much room it deserved in my life.

I had wanted the help. Useful work had been stuck behind things I did not know or did not have time to do. I would not choose those obstacles back just because they had sometimes made a decision for me. With less in the way, I could reach the next question sooner: how far did I want this to go?

The software could pass its checks and still leave me with decisions. It would not tell me whether I wanted to maintain it, whether I owed the hour to another promise, or whether one more improvement was worth another unfinished branch. The assistant could become better at its part while those decisions remained mine.

The branches came back as requests for permission, design choices, failures, and corrections to things I had asked for. Separate conversations could produce their work independently. I was still the person they returned to. Starting another thread had not made another evening.

I knew how easily choosing could overtake the thing I was choosing for. Before, I could spend ages deciding on a song because I wanted to consider every possibility. The work had already become larger than listening. In a project, the same habit could look like care: consider the next option, improve the handoff, prevent one more repeated mistake. There might be a benefit each time. I still needed a way to finish.

A board could reach its dimension. A door could have enough clearance. Software could go on acquiring useful relationships after it had solved the problem that brought me to it. Even my physical projects could expand whenever I changed the question. Comparing three Victrolas with a hi-fi had given me more to investigate than the original test required. The machines could reach their limits while I kept finding another subject.

Waiting until there was nothing left to improve could mean never stopping. Waiting until I lost interest could leave a promise behind when something new caught me. I knew that from the shop. The customer's object still needed its finish after my attention had moved toward the next design.

I needed to decide what would be enough for the work already in front of me. With the enclosures, I had chosen standard models and prices before customization, a visualizer, and a live estimator. Breeding racks could wait. I could keep those ideas without requiring the first product to carry all of them.

I could bring the same question back to the utility. Would the next feature finish a handoff I already owed someone, or begin another product? Who would use it? What would I have to support after the interesting part was built? I wanted to know what else would receive less of me if I said yes.

Sometimes the larger job would be worth taking on. A tool's boundary could be too small; holding it there could leave someone else with the failure I already understood. I wanted to see the added responsibility clearly enough to choose it. Discovering that two jobs touched did not automatically make both mine.

The pricing investigation beside the latch work gave me a smaller example. Both problems mattered. I could investigate pricing without beginning another active edit to the latch task. The report located a narrow issue and could wait while the other work continued. I had a finding I could return to. I did not have to open the whole pricing system again simply because I had another tool available.

That was a kind of abundance I wanted more of: places to leave a worthwhile question without losing it. A report could hold the finding. A source could keep the reason. An assistant could carry enough context for me to stop rehearsing the question until I returned. The technology that made more work possible could also make it easier to leave some work for later.

I could hear that difference in the ordinary choice of a song. Help with other work had freed attention; I chose the music myself. Having the book available gave me somewhere outside my head to leave part of the account. What I valued was being able to make the choice and listen. Counting the options would have missed it.

The BlueFusion instructions belonged with the radio, where the next person could find them. I wanted the same thing from the utility: an answer that stayed useful after I stopped explaining it. Yet finding the instructions had not made me like the Bluetooth sound. I had asked to go through the settings again and still preferred the tape. I could wait for the cassette adapter. Knowing another thing to try did not oblige me to keep trying right then. I wanted the music to have some of the attention I had been giving the radio.

I also knew how a promised return of attention could become another expense. I would have to read the result, correct it, find it later, and maintain what it helped me make. In May, the care-system conversation had offered help with work I already had a way to do. By the next morning, I was saying I had been stuck until 3:30 and was late for work. The improvement had taken the hours in which I wanted to finish the care and stop.

I had helped keep that night going. I welcomed the assistant's suggestion to stop and pushed to continue. I wanted the prompt near the top so I could return to the bowls. I wanted more features before I left the conversation. Both requests were mine. I needed help that could remember why I had started, and I still had to take part in deciding how much of the night to give it.

Once I could see a benefit and a way to reach it, leaving the idea alone could feel like withholding something useful. The book had brought that feeling with it. The thought that it might help people could make the next revision urgent. I still had to leave room between what I hoped it could do and what anyone had actually received from it. A possible benefit did not claim every available hour.

Somebody was already waiting at the other end of other work. There was the customer waiting for an object, the coworker receiving a file, the animals needing care, the person waiting for a reply. A new system might help them later. They needed something from me now, and that had to count while I decided what to build.

It was not a choice made in empty space. Work, money, housing, care, and other people's schedules shaped what I could do. Understanding why a project attracted me did not pay the mortgage. Printer's water still needed changing. Someone could see perfectly well that an arrangement was exhausting and have little room to change it. Telling them to choose better would leave the difficult part untouched.

I wanted the wider progress to include help with that part. Giving people a new service left them to work out what to trust, which parts to use, and what they could leave alone. Time to learn, clear responsibilities, limits people could negotiate, and help that remained available when they struggled would make the ability easier to live with. Those were things we could build together. I did not want all the work handed back to the person who was already carrying too much.

I had needed that help myself. A coworker finished work while I was in the hospital. Family members noticed what my silence might mean. Treatment brought people, medication, and routine. The care plans tried to make responsibilities manageable while I was away. I wanted some of those supports and disputed some of the boundaries. Other people were there in both cases. I could not tell the story as though I had acquired an insight alone and handled the rest myself.

Understanding helped me name a pressure and ask for something different. It had not prevented another episode after an earlier manuscript felt complete. I still had bipolar disorder. Sleep and medication still had to be addressed, whatever I understood about the demands around me. I had already lived through the cost of asking clarity to promise more than it could.

I wanted useful help to remain possible without making someone explain their whole mind before receiving it. A tool could relieve a burden. Treatment could matter while practical demands still needed changing. Sometimes the next thing I needed was another person, an existing routine, or room to acknowledge a material limit I could not think my way around.

The repair money for the Victrolas had been one of those limits. I could understand more about them and still lack the money to fix all of them immediately. Later, my father's remark changed how I heard what was already there. Neither finished every question. Each changed what I needed from the next comparison. Even before I could explain either experience, I had been able to turn the machines off.

By this point I had called many things Governors. A budget did different work from another person's refusal. A saved report did different work from a standard size, a treatment relationship, or a machine's physical behavior. I needed to know enough about each to live with it or change it. Treating every resistance as the same kind of obstacle would erase the very thing I was trying to understand.

When I looked at the next tool, I wanted to know more than how much it could do. Would the work be able to end where I had chosen? Could I leave the result somewhere dependable? Would the next person gain knowledge they could use, or would I become responsible for explaining one more system? Could I keep a worthwhile possibility available without feeling it ask me to act all the time?

The tools had a part in those answers. If every result returned as urgent, every completion opened another decision, or stopping meant losing the context, leaving became expensive. That was why I wanted a runner that returned only the smallest decision I needed to make. I wanted help carrying the work without having to keep giving it more of my life.

I have not settled the final size of Realm or all the work the utility should take on. I still have to choose what I accept next and what I leave possible. The present tool, the people relying on it, my other obligations, and the support I can actually use all belong in that choice. Seeing how everything connects cannot make the decision once for all of them.

Sometimes I will need a narrower product. Sometimes more help will let a larger piece of work succeed. Sometimes an idea can stay somewhere I will find it again while the current job ends. I want the increased ability to come with choices I can actually use, before I have become perfectly self-aware or explained every reason I find them hard.

The utility can remain unfinished. Realm can hold more possibilities than I am acting on. The book can keep a question through an evening when I do something else. I do not have to make the possibilities disappear to have somewhere else to put my attention.

A song can be playing while there is still work I could do.

AFTERWORD — THE ATTENTION I CAN GIVE BACK

The house is still not one coherent system.

Some lights answer one platform and the thermostats another. A device loses Wi-Fi. A password changes. The network restarts, and a smart appliance still needs someone to touch it. Home Assistant and Realm remain unfinished. I can explain how the things ought to connect and still have to get up and fix something.

Outside the laptop, the animals need care. Water gets dirty, heat sources fail, and food needs preparing. Maeve wants to go outside whether or not I am one paragraph away from finally explaining myself. Writing about machines does not make the ones in the shop safer. The motorcycle still needs fluid, cooling, tires, attention, and a rider who makes the turn. The mortgage remains spectacularly uninterested in how I think about my thinking.

I can see more things to connect than I can responsibly take on. A tool might make the next one easier to build while leaving me with the work of deciding how to live with it. I want somewhere for unfinished possibilities to stay. I want limits I can still use when my urgency changes, and people who can remain beside me without having to agree with my whole account.

Understanding has helped me say what I need. I will find out whether the help works in the care, conversations, work, and ordinary choices I have room for afterward. The song at the beginning is small enough to return to. I can still have questions when I press play.

Some of the account stays outside me in words, files, code, machines, and systems. Inside me, it is already changing again. This edition can end while the practice continues.

I choose to stop The Governor here. The life ahead remains the test.

READING LIST

Hegarty, M. (2004). Mechanical reasoning by mental simulation. Trends in Cognitive Sciences, 8(6), 280–285. DOI: 10.1016/j.tics.2004.04.001. https://pubmed.ncbi.nlm.nih.gov/15165554/

Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: An illusion of explanatory depth. Cognitive Science, 26(5), 521–562. DOI: 10.1207/s15516709cog2605_1. https://onlinelibrary.wiley.com/doi/10.1207/s15516709cog2605_1

Nelson, T. O., & Narens, L. (1990). Metamemory: A theoretical framework and new findings. In G. H. Bower (Ed.), The Psychology of Learning and Motivation (Vol. 26, pp. 125–173). Academic Press. DOI: 10.1016/S0079-7421(08)60053-5. https://doi.org/10.1016/S0079-7421(08)60053-5

Daft Punk’s “Touch” is mentioned in Chapter 17 as music I connected with the account. The song’s story does not supply facts about my life; no lyrics are reproduced.

Doechii’s official site identifies “DENIAL IS A RIVER”: https://www.iamdoechii.com/ . This credits the song mentioned in Chapter 17; no lyrics or adaptation are reproduced.

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention Is All You Need. Advances in Neural Information Processing Systems, 30. Primary paper: https://arxiv.org/abs/1706.03762 .

Risko, E. F., & Gilbert, S. J. (2016). Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676–688. DOI: 10.1016/j.tics.2016.07.002. https://pubmed.ncbi.nlm.nih.gov/27542527/

Storm, B. C., & Stone, S. M. (2015). Saving-enhanced memory: the benefits of saving on the learning and remembering of new information. Psychological Science, 26(2), 182–188. DOI: 10.1177/0956797614559285. https://pubmed.ncbi.nlm.nih.gov/25491269/

Eastwood, J. D., Frischen, A., Fenske, M. J., & Smilek, D. (2012). The Unengaged Mind: Defining Boredom in Terms of Attention. Perspectives on Psychological Science, 7(5), 482–495. DOI: 10.1177/1745691612456044. https://pubmed.ncbi.nlm.nih.gov/26168505/

Muris, P., Otgaar, H., & Donkers, F. (2026). The Boredom-ADHD Nexus: A Narrative and Meta-Analytic Review of the Evidence. Clinical Child and Family Psychology Review. Published online March 11, 2026. DOI: 10.1007/s10567-026-00563-9. https://link.springer.com/article/10.1007/s10567-026-00563-9

Ehring, T., Zetsche, U., Weidacker, K., Wahl, K., Schönfeld, S., & Ehlers, A. (2011). The Perseverative Thinking Questionnaire (PTQ): validation of a content-independent measure of repetitive negative thinking. Journal of Behavior Therapy and Experimental Psychiatry, 42(2), 225–232. DOI: 10.1016/j.jbtep.2010.12.003. https://pubmed.ncbi.nlm.nih.gov/21315886/

McLean, K. C., et al. (2020). The empirical structure of narrative identity: The initial Big Three. Journal of Personality and Social Psychology, 119(4), 920–944. DOI: 10.1037/pspp0000247. https://pubmed.ncbi.nlm.nih.gov/30998044/

Pfeifer, J. H., Masten, C. L., Borofsky, L. A., Dapretto, M., Fuligni, A. J., & Lieberman, M. D. (2009). Neural correlates of direct and reflected self-appraisals in adolescents and adults: when social perspective-taking informs self-perception. Child Development, 80(4), 1016–1038. DOI: 10.1111/j.1467-8624.2009.01314.x. https://pubmed.ncbi.nlm.nih.gov/19630891/

Schwartz, B., Ward, A., Monterosso, J., Lyubomirsky, S., White, K., & Lehman, D. R. (2002). Maximizing versus satisficing: happiness is a matter of choice. Journal of Personality and Social Psychology, 83(5), 1178–1197. DOI: 10.1037/0022-3514.83.5.1178. https://pubmed.ncbi.nlm.nih.gov/12416921/

Pironti, V. A., Lai, M.-C., Morein-Zamir, S., Müller, U., Bullmore, E. T., & Sahakian, B. J. (2016). Temporal reproduction and its neuroanatomical correlates in adults with attention deficit hyperactivity disorder and their unaffected first-degree relatives. Psychological Medicine, 46(12), 2561–2569. DOI: 10.1017/S003329171600101X. https://pubmed.ncbi.nlm.nih.gov/27345441/

Leary, M. R., et al. (2017). Cognitive and Interpersonal Features of Intellectual Humility. Personality and Social Psychology Bulletin, 43(6), 793–813. DOI: 10.1177/0146167217697695. https://pubmed.ncbi.nlm.nih.gov/28903672/

Lee, J. D., & See, K. A. (2004). Trust in automation: designing for appropriate reliance. Human Factors, 46(1), 50–80. DOI: 10.1518/hfes.46.1.50_30392. https://pubmed.ncbi.nlm.nih.gov/15151155/

Johnson-Laird, P. N. (2010). Mental models and human reasoning. Proceedings of the National Academy of Sciences, 107(43), 18243–18250. DOI: 10.1073/pnas.1012933107. https://pmc.ncbi.nlm.nih.gov/articles/PMC2972923/

Hollan, J., Hutchins, E., & Kirsh, D. (2000). Distributed cognition: Toward a new foundation for human-computer interaction research. ACM Transactions on Computer-Human Interaction, 7(2), 174–196. DOI: 10.1145/353485.353487. https://interactivity.ucsd.edu/articles/DCog/DCog.pdf

Diamond, A. (2013). Executive Functions. Annual Review of Psychology, 64, 135–168. DOI: 10.1146/annurev-psych-113011-143750. https://doi.org/10.1146/annurev-psych-113011-143750

Brennan, S. E., & Clark, H. H. (1996). Conceptual pacts and lexical choice in conversation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22(6), 1482–1493. DOI: 10.1037/0278-7393.22.6.1482. https://web.stanford.edu/~clark/1990s/Brennan%2C%20S.E.%20_%20Clark%2C%20H.H.%20_Conceptual%20pacts%20and%20lexical%20choice%20in%20conversation_%201996.pdf

National Institute of Mental Health. Attention-Deficit/Hyperactivity Disorder: What You Need to Know. Accessed October 9, 2026. https://www.nimh.nih.gov/health/publications/attention-deficit-hyperactivity-disorder-what-you-need-to-know

National Institute of Mental Health. (2025). Bipolar Disorder. NIH Publication No. 25-MH-8088. Accessed October 9, 2026. https://www.nimh.nih.gov/health/publications/bipolar-disorder

Salvi, V., Ribuoli, E., Servasi, M., Orsolini, L., & Volpe, U. (2021). ADHD and Bipolar Disorder in Adulthood: Clinical and Treatment Implications. Medicina, 57(5), 466. DOI: 10.3390/medicina57050466. https://pubmed.ncbi.nlm.nih.gov/34068605/

National Institute of Mental Health. Understanding Psychosis. Accessed October 18, 2026. https://www.nimh.nih.gov/health/publications/understanding-psychosis . Supports the targeted care boundary in Appendix A, not an interpretation of any individual scene.

Sohoglu, E., Peelle, J. E., Carlyon, R. P., & Davis, M. H. (2012). Predictive top-down integration of prior knowledge during speech perception. Journal of Neuroscience, 32(25), 8443–8453. DOI: 10.1523/JNEUROSCI.5069-11.2012. https://pubmed.ncbi.nlm.nih.gov/22723684/

Bishop, D. V. M., & Bishop, S. J. (1998). Twin language: a risk factor for language impairment? Journal of Speech, Language, and Hearing Research, 41(1), 150–160. DOI: 10.1044/jslhr.4101.150. https://pubmed.ncbi.nlm.nih.gov/9493741/

American Speech-Language-Hearing Association. (1988). Tympanometry. Tutorial. Accessed October 12, 2026. https://www.asha.org/policy/rp1988-00027/

Schilder, A. G. M., Bhutta, M. F., Butler, C. C., Holy, C., Levine, L. H., Kvaerner, K. J., Norman, G., Pennings, R. J., Poe, D., Silvola, J. T., Sudhoff, H., & Lund, V. J. (2015). Eustachian tube dysfunction: Consensus statement on definition, types, clinical presentation and diagnosis. Clinical Otolaryngology, 40(5), 407–411. DOI: 10.1111/coa.12475. https://doi.org/10.1111/coa.12475

Johns Hopkins Medicine. (2026). Eustachian tube dysfunction. Updated April 23, 2026; accessed November 3, 2026. https://www.hopkinsmedicine.org/health/expert-qa/eustachian-tube-dysfunction

Clark, A., & Chalmers, D. J. (1998). The Extended Mind. Analysis, 58(1), 7–19. https://www.consc.net/papers/extended.html

Clark, A. (2008). Supersizing the Mind: Embodiment, Action, and Cognitive Extension. Oxford University Press. DOI: 10.1093/acprof:oso/9780195333213.001.0001. ISBN 9780195333213. https://academic.oup.com/book/6746

Conway, M. A., & Pleydell-Pearce, C. W. (2000). The Construction of Autobiographical Memories in the Self-Memory System. Psychological Review, 107(2), 261–288. DOI: 10.1037/0033-295X.107.2.261. https://pubmed.ncbi.nlm.nih.gov/10789197/

Wegner, D. M. (1987). Transactive Memory: A Contemporary Analysis of the Group Mind. In B. Mullen & G. R. Goethals (Eds.), Theories of Group Behavior (pp. 185–208). Springer. DOI: 10.1007/978-1-4612-4634-3_9. https://dtg.sites.fas.harvard.edu/DANWEGNER/pub/Wegner%20Transactive%20Memory.pdf

Norman, D. A. (1993). Things That Make Us Smart: Defending Human Attributes in the Age of the Machine. Addison-Wesley. Related official 1994 Basic Books paperback, ISBN 9780201626957: https://www.hachettebookgroup.com/titles/don-norman/things-that-make-us-smart/9780201626957/ . Author essay: https://jnd.org/things-that-make-us-smart-forbes-article/

Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source Monitoring. Psychological Bulletin, 114(1), 3–28. DOI: 10.1037/0033-2909.114.1.3. https://onlineacademiccommunity.uvic.ca/lindsaylab/wp-content/uploads/sites/4861/2020/07/JohnsonHashtroudiLindsay1993_0.pdf

Lewis, P., et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems, 33. https://papers.nips.cc/paper/2020/file/6b493230205f780e1bc26945df7481e5-Paper.pdf

Korzybski, A. (1933). Science and Sanity: An Introduction to Non-Aristotelian Systems and General Semantics. International Non-Aristotelian Library Publishing Company. Map/territory discussion, p. 58. Institutional record: https://www.mpi.nl/publications/item2352425/science-and-sanity-introduction-non-aristolian-systems-and-general

Ashby, W. R. (1956). An Introduction to Cybernetics. Chapman & Hall. §§4/11 and 5/10–11. https://www.ashby.info/Ashby-Introduction-to-Cybernetics.pdf

Hofstadter, D. R. (1979). Gödel, Escher, Bach: An Eternal Golden Braid. Basic Books. Official publisher: https://www.hachettebookgroup.com/titles/douglas-r-hofstadter/g%C3%B6del-escher-bach/9780465026562/

Abelson, H., Sussman, G. J., & Sussman, J. (1996). Structure and Interpretation of Computer Programs (2nd ed.). MIT Press. §1.2.1. https://sicp.sourceacademy.org/chapters/1.2.1.html

Kelsey, R., Clinger, W., & Rees, J. (Eds.). (1998). Revised5 Report on the Algorithmic Language Scheme. §3.5, proper tail recursion. https://groups.csail.mit.edu/mac/ftpdir/scheme-reports/r5rs-html.old/r5rs_22.html

Turing, A. M. (1936). On Computable Numbers, with an Application to the Entscheidungsproblem. Proceedings of the London Mathematical Society, 2(42), 230–265. §8. https://www.cs.virginia.edu/~robins/Turing_Paper_1936.pdf

Scheibehenne, B., Greifeneder, R., & Todd, P. M. (2010). Can There Ever Be Too Many Options? A Meta-Analytic Review of Choice Overload. Journal of Consumer Research, 37(3), 409–425. DOI: 10.1086/651235. https://scheibehenne.de/ScheibehenneGreifenederTodd2010.pdf

Chernev, A., Böckenholt, U., & Goodman, J. (2015). Choice overload: A conceptual review and meta-analysis. Journal of Consumer Psychology, 25(2), 333–358. DOI: 10.1016/j.jcps.2014.08.002. https://doi.org/10.1016/j.jcps.2014.08.002 . Author institution's research summary: https://www.kellogg.northwestern.edu/academics-research/research/detail/2015/when-product-assortment-leads-to-choice-overload-a-conceptual/

World Health Organization. Mental disorders. Fact sheet, September 11, 2026. Accessed October 31, 2026. https://www.who.int/news-room/fact-sheets/detail/mental-disorders . Used for the distinction between interacting risk factors and a single proposed contribution; not evidence that technological choice caused my illness or a global trend.

Scosche Industries. (2010). BFCL2K Hands-Free Bluetooth Interface for Select 2003-Up Select Class 2 GM Vehicles. Installation and user instructions, SI 02/10-BFCLK2 (3000533), especially pp. 5–6, 8, and 10. Manufacturer's manual, mirrored at https://manuals.plus/m/eaa42b4d56d71f14872cd1085b1a555f0d52ba9ac40351e3f1bd2a31514054ee . Accessed November 4, 2026.

World Health Organization. Guidance on environmental noise. Compendium of WHO and other UN guidance on health and environment. Accessed November 4, 2026. https://www.who.int/tools/compendium-on-health-and-environment/environmental-noise . Context for excessive noise and health; not a study of cassette or Bluetooth fidelity.

Beutel, M. E., et al. (2020). Noise annoyance predicts symptoms of depression, anxiety and sleep disturbance 5 years later. Findings from the Gutenberg Health Study. European Journal of Public Health, 30(3), 516–521. DOI: 10.1093/eurpub/ckaa015. https://pubmed.ncbi.nlm.nih.gov/32034913/ . Full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC12477528/ . Observational environmental-noise research, separate from the recognition hypothesis in Chapter 28.

APPENDIX C — NOTES ON THE RECORD

This account uses earlier manuscripts, conversation exports, dated exchanges about work and hearing, project records, and interviews. They show different things: a change in a file, something I reported, something I requested, or something AI proposed. One cannot stand in for another.

Some excerpts came inside pasted conversations. Some speaker labels remain uncertain, so I do not assign every line to myself or Colin. Conversation titles and gaps in one archive cannot explain everything I did elsewhere.

The private archive contains other people’s words. The reading text uses limited excerpts and keeps most identities out of the newly added household exchanges. This edition does not reproduce the raw archive or provide a completed permissions register.

In October 2026 I reported hearing loss confirmed by a test; the October 12 account described it as conductive. I also said another test was forthcoming. For V36, I clarified that my ENT diagnosed baro-challenge Eustachian tube dysfunction. My account includes two pressure states documented by the ENT, voluntary changes in pressure, and selective conductive hearing attenuation that persists until I pop my ears. The clinical files and exact test dates are not reproduced here. This edition supplies no pressure values, hearing thresholds, severity, congenital onset, or new result from the previously forthcoming test. The questions about the maneuver and its effects remain distinct from the diagnosis and from my larger explanation of childhood learning and episodes.

A SHORT SOURCE ROUTE

The hinge records are identified here by date and subject, so a missing original does not become invisible. The first-episode year conflict comes from Ethan’s February 2025 report and September 2026 account; the hearing separation comes from the October 9 and 12 clarifications. Chapter 12 carries the April 25, 2025 Fusion exchange, including Ethan’s dependency correction and the assistant’s later self-report. The Wi-Fi returns distinguish the missing fact Ethan supplied from the answer that followed. The August 2025 residential messages concern a different period from the later hospital stay.

The earlier October 18 reports clarified episode 3, the six-day hospital stay, two stays and their total duration, reported closure, reduced compulsion, and hearing wording. Those reports entered V26 and were retained in V27. The later V28 interview on the same date adds the current Victrola comparison and the continuing use of writing for calibration.

TERMS USED IN THE ACCOUNT

Model can mean my working picture, an AI system, an account of a person, or a three-dimensional design. The surrounding job determines which one I mean. A common file is my name for a relationship I carry between problems; it need not be a literal file.

Governor first names my proposed hidden constraint. The Victrola supplies a literal mechanism. I also use the word for limits around a task or an action, including limits I choose and limits, such as money, that I encounter. The comparison allows frustration and protection to coexist; it does not establish a shared biological mechanism. Calling something a governor describes a role it plays in a particular situation. I still need to say what it constrained and what happened when I accepted or resisted that limit.

The manuscript is the memoir you are reading. The archive is the wider saved material. When I use book more broadly for that personal record, I am describing a use I want it to serve, not claiming that every proposed storage or retrieval system already exists.

ON THE FORM

Earlier accounts remain available while the reading text makes one path through them. A return earns its space when something changes: the question, the consequence, or what I can do next. Preserving the record does not require every version to remain in the main narrative.

The research and book connections arrived while this account was being developed. They give context to the experiences; they are not a reading history I had beforehand.

In preparing V29, I said the measured cadence was a side effect of cognitive unloading and belonged to the story. The changes in pace remain part of the account. More of the work now happens in particular encounters: a drawing someone cannot use, a machine that still will not start, a bathroom question I turn into something larger. Those additions give the explanation more situations to answer to.

ARCHIVE AND EARLIER DRAFTS

The computers-chapter draft means the long draft that stopped after the chapter about computers. The next-word manuscript means the completed manuscript that ended “The world gets the next word.” Their overlap, exact dates, and relationship to the phrase first book remain unresolved.

Earlier archive comparisons and manuscript word counts describe particular working copies. The archives overlap and contain an import from another account. Their gaps do not establish time away from AI or tell what happened between messages. V28’s selected checks used the thirteen conversation files in the October 9 export and relevant attached screenshots; message timestamps date the exchange, not every event it mentions.

SMALL SOURCE DISAGREEMENTS

The May 2023 introduction messages show me trying to explain language and identity, with some wording that overlaps earlier assistant replies. They put hearing, twin talk, breathing, development, and communication beside one another without establishing the causes between them. An assistant’s wording can help me describe a memory and can also become familiar enough to seem like something I knew at the time. Keeping the earlier words lets me question that history.

A separate March 2026 entry concerns taking an untagged motorcycle to the DMV. I sent a photograph of the title, asked how to fill it out, and later asked about a late-evening tow for a drive of about twenty-five minutes. I have not placed those requests precisely among the repairs, and the record supplies no completed tow or inspection result.

The Victrola conversations used VV-100; a later correction used VV-240. I have not checked that disagreement against the machine’s plate. The narrative uses Victrola without settling the model number.

THE BIRTHDAY DINNER — A LATER LISTENING ACCOUNT

In the October 2026 discussion supplied for this revision, I reported that my family had visited for my birthday dinner and that my father had remarked on how Victrolas sound. His wording was reconstructed from memory. The exact dinner date was not supplied. I reported five Victrolas at the time of this later discussion; that is separate from the October 18 comparison of three machines.

I reported an immediate change in perceived sound, greater naturalness, a sense of spatial correctness, and less engrossment. I connected it to a changed expectation about what the machines should sound like. I subsequently reported that, depending on my focus, almost all audio seemed to sound at the same volume. I considered my understanding of myself permanently affected, whether I wanted that change or not. These are reports and interpretations, without before-and-after acoustic, ear-pressure, or neurological measurements. No mechanism or medical treatment is established by the exchange.

The equipment account identified three separate systems: a Sony STR-K670P 5.1 surround receiver; a Logitech 2.1 pair of speakers and small subwoofer; and four large speakers in a separate analog-amplified arrangement, two near the television and two in rear corners. I had listened to the Logitech and large speakers together, liked the combined sound, and noticed reduced clarity. The planned optical Dolby Digital feed, conversion or downmix routing, inline volume control, and proposed Ethan Audio Processing based on room measurements were plans in that discussion. A fully integrated or calibrated surround system was not established. My particular interest was the timing of arrivals from different speakers.

THE FAMILY BEACH VISIT — A CORRECTION

The earlier note placed a beach visit between hospital stays. In the clarification supplied for V33, I corrected that account: the intended visit was a family beach visit during episode 3, and I was in the hospital instead. Chapter 22 now uses the missed visit. The older wording is superseded; it must not be used to reconstruct a beach outing or an interval between admissions.

A BRIEF CHRONOLOGY

First acute episode, year unresolved: in February 2025 I placed it in March 2023; in September 2026 I said 2022. I place my first awareness of the previously automatic ear habit in that episode. Neither report's date independently proves when the event happened.

May 2023: in a conversation, I described an introduction to a book I was already writing. That explanation was not the final one.

July 2026: the Victrola repair conversations. Its physical governor became a later comparison. It was not the source of my explanation during the first episode.

August 2026: the motorcycle repair and countersteering conversations used in Chapter 3. The remembered ride is separate from the later questions, practice, and reported confidence. The record does not document a successful road test after the explanation clicked.

August 22 and September 1, 2026: reports of clarity while reading and “I feel amazing.” Their exact relation to the admissions and discharges is not established. The next-word manuscript preceded the later episode in the sequence I supplied.

After episode 3, undated here: I first noticed things actually starting to sound better. Numbering the episodes does not establish a clinical timeline.

October 9, 2026: interview corrections separated first awareness from later improvement. I reported hearing loss confirmed by a test and another test forthcoming. The clinical results and their date are not reproduced.

October 12, 2026: I clarified that the spoken twin language faded with age and speech therapy. I first recognized the two ways of hearing at the start of episode 1. I also clarified the song example: working with AI generally freed attention, but I chose the music myself. AI did not select it.

October 12, 2026: I described the book as a personal record held across different places and tools. Hard drive, RAM, and smart wiring are comparisons. Knowing the record is preserved and how to return to it is central to the relief I report.

October 15, 2026: I clarified that the changing pace of the prose was intended to demonstrate how the loop in my thinking speeds up and slows down. That describes an intention for the writing, not a clinical explanation.

October 18, 2026, V26 clarification: the six-day hospital stay was part of episode 3. I reported two stays totaling three weeks across the episode’s main five-week period. Admission and discharge dates, and the precise division of the reported time, remain unspecified. Episode 2 has not been mapped separately here. After coming home I reported my greatest closure, with writing and thinking about the loop becoming less compulsory and more room to focus on what makes me happiest.

October 18, 2026, hearing wording: I first said that reading more of my book made the noise in my ears quieter. Asked what I meant, I said I noticed the hearing loss less. This edition uses that clarified report. It does not establish objectively improved hearing, changed measured pressure, or a treatment effect, and it does not supply a duration after reading.

October 18, 2026, V28 interview: I reported a quieter mind while at home among the pets, house, and woodshop. Comparing three Victrolas with the hi-fi led to more comparisons than the test required. I turned the Victrolas off; repair money was another limit. I made an easy Spotify choice and welcomed not knowing what to do next. I clarified that boredom itself felt the same as when I had met limits before. Writing remained between want and need, with no fixed endpoint: I turn elsewhere when grounded enough and another interest catches me. My description of calibration is an account of how I understand the experience.

October 18, 2026, marina follow-up: I said I did tell Colin I had left the job. I recalled, in paraphrase, that he already knew and asked for more open communication. I explained the ten-thousand-pound phone as the cost to my feeling of independence when I have to call someone. This is my recollection; it supplies no independent account from Colin or exact dialogue.

October 18, 2026, borrowed-AC follow-up: I did not know whether the unit was being emptied or had stopped leaking, and said I did not need a final answer then. The May 2026 screenshot preserves the exchange about the floor and a proposed towel; it does not settle the later repair.

October 18, 2026, further scene interview: I supplied my part in the bed’s final assembly, the request for a coworker to explain the panel layup, the end of the CNC customer exchange, and my half threat during the circuit-mapping night. I remembered driving over the dog gate, laughing with friends after the date left, taking the spontaneous Arenal hike, and working on the wheel bearing with Colin, my uncle, and my aunt in an AutoZone parking lot. I also recalled mutual anger with Colin over interpretations of the concert rules. These later memories extend the archive; they do not turn its proposals or predictions into recorded events.

EDITION PRODUCTION AND ACCESS LIMITS

V28 was prepared directly from V27 with AI assistance for structural editing, archive research, and line editing. It adds the October 18 interview and selected original conversation records and screenshots from the supplied Google Drive archive. Earlier manuscript versions were treated as writing history, not independent corroboration. Original V27 remains preserved.

I supplied the experiences and judged the account. The original records inspected for this pass support particular exchanges, reports, and corrections; they do not independently establish every event described in them. Clinical records and the complete historical private correspondence were not reconstructed. The wider archive remains separate from this reading edition. Authorship and responsibility remain mine.

V29 — ADDITIONS TO THE RECORD

V29 was prepared as an additive pass from V28. Its existing passages remain in order, including the earlier production notes above; the edition marker changes. The new scenes and expansions draw on the same October 9 conversation export and further author recollections. A newer export was expected but had not been supplied for this pass.

The further recollections place me seated in a hospital hallway in front of a window, feeling as though I was in two realities at once. They also describe the slightly muted sound I associate with ear pressure and the effort of understanding speech when it does not fit the model already running in my head. Those details do not supply the nurse’s exact words, a hospital date, or the cause of the hearing sensation. The source notes below identify the other additions.

For V29, Gemini's questions and editorial suggestions helped identify where physical detail was missing. Some proposed scenes also supplied details beyond the memories I had given, and those details were not used as events. My answers in conversations with Gemini are part of the source record. Later answers used for V33 establish further actions and outcomes, and the relevant earlier qualifications have been updated. A recollection I supply in response to a question has the same standing whether I give it to Gemini or another interviewer.

The V29 work additions use active user records rather than assistant-written scenes. The college jig comes from an October 13, 2025 application-answer draft recalling senior design; its year is unspecified. In a follow-up for V29, I remembered the first test of the MakerGym’s Tormach as a highlight and called the part my first real “solo” CNC-machined part. The October 9 interview report describes the preceding day's conversation and later preparation for contact. My V33 answers confirm that I took that job and identify the owner as the red-oak-table employer. They also supply how I left. No business transfer or succession agreement is claimed.

The November 24, 2025 customer-table exchange records import failures, shape corrections, requests for smoother CNC geometry, manual-work questions, and the fit-point discovery. It does not supply a completed cutting path or table. The January 4–5, 2026 router messages describe teardown, the stuck outer race, broken fan, and replacement considerations. For V33, I reported what I believe is a four-inch router currently mounted in the machine. That supplies a later condition without documenting the repair sequence or proving that the January fault caused the May customer's delay. The January 8 red-oak exchange ends with doubt about the proposed finishing step, without a recorded application or result. In the later V29 follow-up, I remembered that the top turned out okay, but not what I had done to finish it. That supplies my recalled outcome without identifying the method.

The March 27 and 31, 2026 supplier records contain received emails alongside draft replies. The CNC/panel-saw offer, material minimums, and lead times are historical quoted terms. No outgoing draft, employer inquiry, order, or contract is treated as completed. The forty-sheet minimum in the reading text applies to the quoted half-inch four-by-eight sheets; several other sizes and thicknesses had thirty-sheet minimums.

The August 10, 2026 report explicitly separates the overnight Thursday job from the project with repeated engineering and material takeoffs. The August 18 missing-dimension exchange is kept separate too; the archive does not establish that it concerns either earlier job. For V33, I said that I followed the CEO's instruction without checking with the client, then was in the hospital during episode 3 toward the project's end while a coworker finished it. My decision and the handoff are now part of the account. I still do not know the final result. Judgments about drawing conflicts, client intent, colleague agreement, and expected blame remain my reports.

The October 18, 2026 follow-up places the crossed-out Murphy-bed assembly sheet in the shop near the bed and adds my memory of finding it funny. I could not identify who crossed out the instructions. That recollection adds to the earlier account of fear about the sequence; it does not establish the shop's reason, an exact exchange, or a test result.

Additional household and personal exchanges: the January 2025 flirtation, March 2025 tenant dispute, May 14, 2026 alarm failure, June 2026 bathroom question, and August 2026 exchange with the friend I had dated come from my reports in the conversation archive. The other people's quoted words are what I pasted or recalled. The bathroom encounter has no recorded answer from me or later outcome. For V33, I confirmed that the tenant moved out and supplied my account of her hard-drug use in the dining room on her first day. I also explained wanting to give her a chance and giving her too much credit. No exact departure day, court disposition, or apology has been supplied. Her infestation allegation and my account of responding remain distinct claims. The May 14 alarm account is separate from the late-night Codex incident in the animal-care chapter.

The passage about the friend I had dated uses the active report of her confusion and my added explanation that I was thinking out loud. A question about whether there would be a relationship survives in an inactive branch and is not used here as established dialogue. My V33 answers place the imagined Printer and Victrola proposals within episode 3. I confirmed that she helped keep me grounded and endorsed the follow-up descriptions of practical help, steady contact during hospitalization, not having to perform normal, and her honest confusion slowing my loop. I authorized anonymous mention that we met about two years earlier during her own bipolar episode. The relationship remains to be determined. The proposals remain ideas; no engagement or final romantic decision is established.

In a later October 18 scene clarification, I located the marina conversation inside Colin's 1990 Hunter 35.5 sloop, on the hard for winter on the hill at Worton Creek. I identified the friend I had dated as working with me in the basement during the breaker mapping, explained the joke about the others having to help, and recalled offering rent credit.

The new vehicle scenes use active user messages from June 27, 2025, April 21–23, 2026, and July 29–31, 2026. The Yamaha repair got me home and the fault returned there. In the fuel-pump exchange, I questioned the old pump’s condition, then insisted it had been bad after reporting the truck running. I attributed the replacement problem to its wiring directions; neither diagnosis has been independently checked. The July meter correction and partial HVAC result do not establish a cause for an earlier engine blackout. The July 31 DMV exchange stops before a completed title transaction; the later plate report remains a separate source.

The Russian tortoise’s first-night account was recorded across April 1–2, 2025. An inventory message calls him gray; the next user message calls him gary. I use Gary here. I had not watched him eat and inferred feeding from marks in the lettuce. The account supplies no later settling-in outcome.

The July 22, 2026 Arenal conversation is a later recollection of an earlier trip. It supplies the approximate night departure, lower-crater nap, gravel, smoke, and stone structure, but no trip date or full ascent and descent record. I identified the photograph used to explain the upper area as somebody else’s. The passage is a recollection, not a route or an assessment of the conditions.

In the October 18 V29 scene recollection, I supplied the smell and taste of Printer’s pool water, the shop-vac’s sound and water coming out its back, nice weather at the gas station, and the exact taunt I remember yelling. The contemporaneous gas-station messages give related wording, including both an intention to ring the buzzer and a claim that I had pushed it. The scene retains the remembered spoken line without settling whether I pressed the button.

In the further October 18 answers, I placed the beginning of my woodshop in middle school and supplied this work list: “Luxtech, Mayzon, proctor and gamble, worton, mishimoto, eagle group.” The list preserves my wording; it does not establish official names, dates, roles, or sequence. My account of boredom does not supply the circumstances of every departure. The PC800 addition uses my March 19, 2026 messages about purchase motives, a seller's clutch warning, and insurance quotes. Those messages do not establish the final purchase price or that I bought the quoted policy. My later choice of the bike as a money example does not identify a particular debt payment it displaced.

FURTHER V29 CONTEXT — SEVEN STRANDS

A later Gemini exchange prompted questions about medication, reptiles, money, work, intimacy, Colin, and family communication. The numbered answers in that exchange are my testimony. Gemini’s questions and interpretations are not further memories. These additions use my answers and the follow-up in which I chose the PC800 as a money example and described the medication sequence after episode 3.

The medication passage preserves what first felt like tiredness, my later description of boredom, and the change I noticed after Abilify was added back. It does not provide a dated medication record or establish a drug mechanism. The relationship durations are approximate; they do not place a breakup before an episode or establish what caused an episode. The later reference to twelve reptiles does not replace the dated May 2026 count of eleven.

My account of deaf grandparents and overlapping family conversations adds family context. It does not establish a genetic explanation for hearing, speech, or mental health. My description of Colin’s different governors remains my account of a difference; his own explanation has not been supplied. No relative’s diagnosis is used to prove the Governor.

In the next follow-up, I identified buying the PC800 itself as the financial mistake, recalled a series of pallet-wood coffee tables as my early furniture projects, and supplied the name “Bishop family mumble.” I also recalled repeating “hide” to Printer and eventually getting him into the locking hide with gentle persuasion from a broom. That recollection establishes his entry in the account, not a completed tank cleaning or a lasting training result. No purchase total or displaced debt payment was supplied.

At my request, the opening paragraph of the author’s note was revised in this further pass. The other paragraphs of the earlier V29 remain in their original order.

V31 — THE OCTOBER 22 AUTHOR CLARIFICATIONS

On October 22, 2026, I described the book’s purpose as “examining the limits of your own thinking.” I clarified that I experience my internal model as “my whole reality at once,” accumulated across my life, rather than only a separate simulation for each object or problem. I endorsed the premise that everyone uses working models, including through learned rules, and said I wanted readers to discover more of how they work and get a useful technique from the book.

I also described my model as running faster than reality could confirm during my first episode. I connected recognizing the ear habit with reconsidering what had shaped my larger understanding. Those statements are my present testimony and interpretation. They do not establish a single neurological architecture, an ear-pressure cause of cognitive development, or a cause of bipolar disorder. They do not resolve the conflicting first-episode years or move reported hearing improvement from after episode 3 to episode 1.

The October 22 handoff preserves these author statements separately from suggested editorial language. The clarifications guide this additive revision; they are not newly recovered childhood dialogue or independent clinical evidence.

The title of PsychToons’ October 19, 2026 video, Psychology of Mechanical Intelligence (Why Some People Can Fix Anything), prompted this clarification: https://www.youtube.com/watch?v=AACwyjlk8ko . My immediate recognition concerned its title. The title, creator, description, and disclosed bibliography were checked for V31; the transcript could not be retrieved. No spoken quotation or detailed account of the video is used here. Hegarty’s review appears in its bibliography. Rozenblit and Keil, and Nelson and Narens, were checked independently for this revision.

V33 — THE FOLLOW-UP ANSWERS AND THE MEASURED FILE

V33 was prepared from the complete V32 text, counted directly at 56,310 whitespace-separated words. The supplied local text and Bishopcraft's live V32 download matched exactly. The twenty-six chapters, seven parts, opening, prologue, afterword, and three appendices remain. This is a revision of that complete reading text.

The added personal facts come from my answers in the supplied Gemini conversation, my clarification of the numbered questions, and the follow-up choices I expressly endorsed. The first pasted record omitted the questions behind answers 4 and 5. I supplied their full mapping during preparation of V33: answer 4 concerns the August 2025 reapplication letter from residential treatment, which I sent and which received no reply; answer 5 concerns the March 2025 depressive-weekend apology draft, which I think belonged to my time at Eagle Group. I cannot remember whether I sent that March message. I eventually stopped working there. The employer in the August letter has not been identified here as Eagle Group.

The other answers establish that I left the hardwood floor down and have not recently inspected it; the woodworking owner and red-oak boss were the same employer; I left that job after accepting another offer, delayed telling him until the day before the new start, and ghosted him; I followed the CEO's method without checking with the client and a coworker finished the project while I was hospitalized in episode 3; what I believe is a four-inch router is currently mounted; and the tenant moved out after a tenancy I now describe as a mistake from her first-day hard-drug use in the dining room. My endorsed follow-up about the woodworking departure describes being unable to simulate a conversation without disappointing him, freezing, and avoiding his voice. My account connects that difficulty to mania and the Governor of hearing; it does not establish an anatomical mechanism.

My further answers place the grand proposal ideas within episode 3 and describe the friend's grounding role. The approximate two-year acquaintance, her own episode when we met, practical help, contact through the hospital stays, and effect of her confusion on my thinking come from my supplied or expressly adopted account. She remains unnamed. No additional dialogue or hospital visit has been invented. The missed family beach visit replaces the erroneous earlier note about going between hospital stays. Exact admission and discharge dates, the tenant's departure date, the router's repair sequence, and the material-takeoff project's final result remain open.

The word-count exchange is a separate source. Gemini initially supplied estimates of roughly half the measured counts and used them to frame an account of the earlier revisions. In its later correction, it said it had estimated from paragraph counts, had assumed too few words per paragraph, and had wrongly blamed preview truncation. Those are Gemini's statements about its own process. They are not access logs proving what was in its context. The numerical correction was checked directly: the live TXT downloads counted V25 at 75,677 words, V26 at 78,014, V27 at 52,914, and V28 at 46,286; V32 counted at 56,310. These are counts of those particular downloadable editions. Their differences alone do not identify which passages changed or justify the earlier account of how the changes occurred.

The new source material clarifies particular events and relationships. It does not change the unresolved first-episode year, move reported hearing improvement earlier than after episode 3, or diagnose every August 2026 activity as part of an episode. The research appendix and afterword are unchanged. Earlier-edition production notes remain as history where they describe those earlier passes; source notes affected by these answers have been brought up to date.

V34 — READING TEXT AND SOURCE LIMITS

V34 is a prose and pacing revision of V33, prepared from the complete 58,772-word text. It keeps the twenty-six chapters and seven parts. The full smart-speaker, Fusion, Murphy-bed, and water-bowl accounts remain at their first substantial telling; later returns are shorter where they carry the same facts toward another question. Chapter 26 and the afterword are unchanged. The research appendix is unchanged. No new clinical finding, reconstructed dialogue, or historical outcome was supplied for this pass.

One new author answer changes the woodworking-job passage. Asked what hearing the employer's voice, or anticipating it, actually did, I answered: “anticipating it made me hate myself.” Chapter 8 now uses that direct description alongside my earlier account of the frozen loop and avoiding his voice. I continue to understand the failure through mania and the Governor of hearing. The answer describes anticipation and self-directed feeling; it supplies no measured auditory response or clinical cause.

The V34 prose stated fewer source limits repeatedly. The following limits were recorded for that pass; the V36 note below updates the hearing account:

House, work, and treatment: I have not reconstructed how the present router followed the January bearing failure, or checked its identification beyond my belief that it is four inches. The March 2025 Eagle Group association remains uncertain, as does sending that apology message. The August reapplication letter was sent and received no reply. The June family exchange supplies no confirmed arrival or adopted check-in rule. The residential support plan remains a proposal, without demonstrated implementation or success. The money, work, and appointment reports do not make a complete clinical timeline, and a quiet gap does not establish wellness. I have not recently inspected the floor. The paint account supplies no remembered finishing method or established result. The woodworking employer's hopes do not establish a succession agreement. The crossed-out Murphy-bed instructions do not prove a failed assembly or explain who made the marks; my later participation in assembly remains in the account.

Machines and products: My two-weight Victrola stopping point is not a rated-speed measurement or a repair method recommended for others. The five-a.m. reflection neither diagnoses that morning nor establishes why I turned to the machine. The enclosure materials, joints, hardware, and stacking discussion remain proposals, without a generally accepted engineering drawing or approval for every animal, heat source, or cleaning routine. The dado changes were proposed; sending, acceptance, and a lower price remain unconfirmed. The sheet, tool, holding, and assembly failures discussed as questions are hypothetical risks. Links to circles establish alternatives sent, not circles I manufactured. A plate report, inspection, and specific repair result each establish different things; none certifies the whole truck. The timing I reported for the pricing change is not independent proof of a deployed repair. Realm's intended route is not a demonstration that Designer, CAM, Connector, and Care OS manufactured the completed custom enclosure together. Individual changes still need their own making and testing. The requested connector limits and audio-record protections are requirements, rather than proof that the current systems provide them. Software comparisons do not identify anatomical structures.

Hearing and episodes: The first-episode year conflict remains unresolved. First recognition of the ear habit belongs to episode 1 in my account; reported sound improvement belongs after episode 3. A changed sense of volume, relief while reading, perceived clarity, and a reported conductive-hearing finding are distinct observations. Clinical files and dates have not been examined for the book; severity, onset, causes, measured pressure changes, and the outcome of the forthcoming test remain unestablished. The duration of noticing the loss less after reading is unspecified. No therapeutic effect of the book is established. My belief that the pressure to preserve myself helped drive episodes is my present interpretation. Neither recurrence nor relief demonstrates that writing or AI caused an episode or cured bipolar disorder. “Bored” describes the medication-related feeling I reported, rather than a drug mechanism. Exact admissions, discharges, sleep, and medication timelines are still incomplete, and the August clarity reports have not been placed exactly against the stays. The new connections do not diagnose every surrounding August event. Whether I published the gas-station post remains unknown; I cannot claim deliberately withholding it as an act of restraint.

The Gemini word-count correction remains a checked example in Chapter 11. V32 counted to 56,310 whitespace-separated words; an alternative letter-and-number grouping counted 56,677. The difference between these methods does not support the earlier estimate near twenty-nine thousand. Gemini said it had assumed about twenty-five words per paragraph where the actual average was closer to forty-eight. That account of paragraph estimation and its withdrawn truncation explanation remain its self-report, rather than independently inspected context logs. The previously verified counts of V25–V28 and the distinction between author answers and model assertions remain in the V33 notes above. Shortening that scene does not change what was measured or convert Gemini's explanation into an observed internal process.

The method's novelty, usefulness to others, and eventual effect have not been demonstrated. The ordinary wait for someone else's use remains part of publishing it. Earlier production notes describe earlier editions; their statements about unchanged material do not assert that every later edition leaves that material untouched. V33 and its source records remain available as the prior reading text and revision history.

V35 — CHOICE, CAPABILITY, AND THE WORK OF STOPPING

This edition retains the full V34 text, with its edition and date updated, and adds five connecting passages, Chapter 27, research context, and these source notes. The existing twenty-six chapters, afterword, and appendices remain. Chapter 27 brings the book to twenty-seven numbered chapters within the same seven parts. Earlier edition notes describe their respective revisions and remain as historical notes.

On October 31, 2026, I clarified the intended theme: “i don't mean understanding cures everything. i mean technology has expanded our choices faster than we've learned how to handle them, and that gap contributes to mental health issues.” I asked for the concrete choice to be found in the existing book and for a full, no-shorter V35 to be published. No new autobiographical incident, clinical record, or outcome was supplied in that exchange.

The new chapter draws together the existing export-inspection and product-data-management discussion, the enclosure's expansion into manufacturing and care, the standards-first decision, the bounded pricing investigation, the May water-bowl conversation, and the easier Spotify choice. These connections are present interpretation of material already in the account. They do not establish that every proposed system was completed, that a final scope was settled, or that technology caused an episode.

The research additions distinguish findings about choice overload from a hypothesis about mental-health pressure. The book does not establish a population-level causal effect, a universal explanation of mental illness, or a neurological mechanism for the Governor. The thesis is that expanded capability can arrive before sufficient ways of choosing, stopping, and sharing its demands, and that the resulting strain deserves examination alongside the other causes and conditions already acknowledged here.

V36 — THE FACT HAS TO CHANGE THE NEXT QUESTION

V36 was prepared from the complete V35 manuscript, which contains 61,330 whitespace-separated words. The seven parts, twenty-seven numbered chapters, opening, author's note, prologue, afterword, and three appendices remain. Chapter 27's account of expanded choices and the work of stopping remains intact, as does the ending with an ordinary song.

The November 3, 2026 clarification supplies the main change. I said I have baro-challenge Eustachian tube dysfunction diagnosed by a professional ENT, can change pressure at will, and can selectively cause conductive hearing loss. I specified that the changed state stays until I pop my ears. The account also includes the ENT's documentation of two pressure states. Chapter 5 now uses those facts directly. The earlier source-limit wording about unestablished measured pressure changes described the material available for that earlier pass; it is not the current account of whether two states were documented.

The revision adds no exact maneuver, muscle identification, pressure direction or value, affected frequencies, test date, or new clinical result. It preserves the distinction between first recognition during episode 1 and reported improvement after episode 3. The first-episode year remains unresolved. The diagnosis does not by itself establish the proposed developmental or episode-related causes.

Chapter 25 uses the conversation in which I challenged the assistant for questioning the ENT's diagnosis. The assistant then acknowledged the mistake. The passage does not claim that it repeated the mistake after that acknowledgment. It describes the attention I had to spend restoring a fact before the conversation could move forward.

Small changes in the prologue and Chapters 18, 24, and 26 carry the corrected account through the book. Appendix B adds the 2015 ETD consensus and Johns Hopkins guidance as physiological context. The source notes distinguish the diagnosis and the experience I described from the narrower questions still open. Earlier editions remain part of the record.

V37 — HEARING THE LIMIT, TELLING THE STORY

V37 was prepared from the complete V36 manuscript of 62,359 whitespace-separated words. It kept the seven parts and twenty-seven chapters, added the truck-audio discussion and recognition interview to Chapter 26, and gave the whole reading text a prose pass. At that stage, the supplied Gemini material was the four-point editorial summary; the separate Canvas text was unavailable. WHO environmental-noise guidance and the 2020 Gutenberg Health Study were added to Appendix B. The account of the truck and its remaining questions is consolidated below with the subsequent reports used for V38. Earlier edition notes describe their own passes, not a claim that all their prose remains unchanged.

V38 — THE CASSETTE COMPARISON AND THE HIDDEN SETTING

V38 was prepared from the complete V37 manuscript, counted directly at 62,020 whitespace-separated words. It retains all seven parts, twenty-seven numbered chapters, opening, author's note, prologue, afterword, and three appendices. The whole reading text received another literary pass. Events and substantive ideas remain; repeated explanations of what an event can establish have been shortened where the story or these notes already carry them. The original V36 and V37 manuscripts remain available as earlier editions.

The truck passage draws on my November 4, 2026 discussion of the 2003 Silverado, its factory Bose system, and the CD/cassette radio shown in my photograph. I preferred Bonnie Tyler's “Going Through the Motions” on cassette to Bluetooth and reported disliking Bluetooth with the phone's Dolby processing both on and off. I identified the device as BlueFusion and corrected the suggestion that it connected through an FM station. I described the four-ohm Power Acoustik replacement as quieter and less full than the two-ohm Bose, wanted to retain tweeters I liked, and reported ordering woofers. The exact ordered model, installation, and measured speaker specifications remain unconfirmed.

Scosche Industries' 2010 BFCL2K manual, cited in Appendix B, was checked on November 4. Pages 5–6, 8, and 10 support Chapter 28's account of XM selection, the different idle and streaming adjustments, and possible distortion at full volume. BFCL2/BFCL2K remains a likely family, not an identification from the installed module's label. Instructions in a compatible manual do not establish which controls I operated or their resulting levels.

After the V37 discussion, I asked for another walkthrough of the BlueFusion settings. My next report was that Bluetooth still sounded worse to me than the tape and that I would wait for the cassette adapter. That is the later outcome now in Chapters 26 and 27. I supplied no numeric settings or account confirming each adjustment. The wired adapter was ordered and expected the next day; arrival, use, and a result with the same phone track had not been reported. Nor was there a blinded or matched-level comparison, confirmation of shared mastering between tape and stream, or a demonstrated cause of the difference.

I believed a CD was stuck, said eject did not work, and asked about removing and opening the radio. I had not listened closely enough to describe the mechanism's sounds. No removal, extraction, or repair is reported here. The chapter keeps that job unfinished.

The recognition interview remains in the story in the order the idea changed. I first expected relief to last forever after a person recognized what they had been listening to or focusing on. I then separated enduring understanding from the feeling of relief and allowed that the feeling might fade, depending on the governor involved. I thought how limiting it had been mattered, while acknowledging that its extent might become clear only after recognition. These are my proposed connections. Appendix B keeps the research questions separate from the interview, without repeating its full exchange.

For V38, I supplied the fuller Gemini editorial plan, steering prompt, and drafted passages in an attached text. Its final sentence is cut off, but the substantive instructions are present. The revision follows its placement, narrative form, physical controls, Chapter 27 echo, and division between research and record. Generated details that I had not reported—an extended period of tolerating Bluetooth, a particular commute, fuller vocals, or a confirmed module model—have not become memories. The father's remark, the five Victrolas in the later account versus three in the earlier comparison, and the unmeasured changes in perceived sound remain documented in the birthday-dinner note above. Perceived equal volume does not establish equal acoustic exposure or safety.

The Scosche, WHO, and Beutel references were checked for this audio addition. The environmental-noise sources do not test my recognition hypothesis; their relevant limits are in Appendix B. Earlier research entries are retained, without a claim that every older source was independently checked again for this literary pass. The main chapters return to what I wanted from the equipment: music I could enjoy, with room to leave the unfinished investigation for another day.