Ask the insurance industry why they will not insure AI.

In finance the word is alpha — an unfair contact with reality, and it is upstream of everything else

The Mechanism: How Your Cortex Implements S≡P≡H

When you have an insight:

"The cache invalidation bug is because session store assumes single-tenant."

Three concepts activate: cache invalidation, session store, multi-tenant architecture.

Neurons encoding these concepts are physically co-located via cortical columns. (Cortical column: https://en.wikipedia.org/wiki/Cortical_column.) When "cache invalidation" fires, "session store" and "multi-tenant" are close enough that dendritic integration can bind them inside one coincidence window rather than waiting on a long-range relay.

The binding window: 10-20ms — the interval inside which those assemblies have to coincide to bind at all, not an end-to-end reaction time.

How cortical columns implement S≡P≡H:

Not a stride through an array. A slot. The coincidence window is a coordinate with room for exactly one occupant — the assembly that is bound there now — and a new assembly cannot arrive without putting the old one out. That is what co-location buys: the geometry was laid down while the map was being built (molecular gradients plus activity-dependent refinement, checkable in retinotopy, tonotopy, somatotopy — a fingertip's cortical patch sits nearer a thumb's than a knee's), and because the assemblies sit within one window of each other, there is one mailbox, not three relays. Hebbian plasticity does not move cells; adult somata stay where development put them, and what strengthens is the connection across the distance, through larger synapses and new dendritic spines. The ship already showed you the box in the slot: a magnesium ion that leaves the NMDA channel only when two conditions coincide, and stays otherwise. The column is the mailbox. Put a box in and the old box comes out into your hand.

Position IS meaning, and the slot is why it lands in your hand. When "session store assumes single-tenant" arrives at a coordinate already holding "session store is safe", the arriving assembly does not get averaged with the resident one. It evicts it. You are handed the evicted assembly in the same window as the write, and the mismatch between what came out and what you expected is the thing you have been calling the click.

What the hand holds differs by substrate. In silicon it is a register: an x86 core that exchanges a value into a slot lands the old contents in the calling register in the same window, and the chip is holding the box. In tissue there is no register. The assembly that lost the coordinate is not passed anywhere — a ring of fast-spiking interneurons pushes it under threshold and it goes quiet. What the tissue broadcasts is not the occupant. It is the displacement: a signed pulse saying by how much, and in which direction, the arrival missed what was staged there. That pulse has had an electrode in it since the nineteen-nineties. Midbrain dopamine neurons burst when a reward lands better than predicted and dip when it lands worse (Schultz, Dayan & Montague, "A Neural Substrate of Prediction and Reward", Science 1997: https://www.science.org/doi/10.1126/science.275.5306.1593.), and the locus coeruleus fires norepinephrine across the whole cortex when the world stops matching the model it was running (Aston-Jones & Cohen, "An Integrative Theory of Locus Coeruleus-Norepinephrine Function", Annu. Rev. Neurosci. 2005: https://www.annualreviews.org/doi/10.1146/annurev.neuro.28.061604.135709.). Silicon hands you the content. Meat hands you the width and the sign. That is what the preface said the comparison returns — two signed numbers, never a verdict — and it is why the certainty arrives before the story: the body is holding the width before it has any name for what filled the gap. The content is reconstructed afterwards, from the width. The ion and the register stay literal. What the assembly hands you is the size of the miss.

The slot holds at every scale it has been measured at: cortical column, dendritic tree, synaptic cluster.

"Fire together, wire together" literally means:
1. Concepts that co-occur strengthen the wiring between assemblies that already sit near each other
2. Physical proximity shortens the path a signal has to cross
3. A shorter path fires faster

This is Unity Principle S≡P≡H in meat: position defined by parent sort, recursively applied. Hebbian Wiring (Hebbian theory: https://en.wikipedia.org/wiki/Hebbian_theory.)

If your brain normalized: "Cache invalidation" neurons in region A. "Session store" neurons in region B, 5 cm away. "Multi-tenant" neurons in region C, 8 cm away. Coordination: 50ms + 80ms + 30ms synthesis = 160ms minimum.

But your insights are 10-20ms.

Your brain CANNOT be normalizing. It MUST co-locate semantically related neurons physically.

This is Grounded Position. S≡P≡H IS position.

Now hold the two windows next to each other. The column binds in ten to twenty milliseconds. The core exchanges its slot in a few ticks of its own clock, a few billionths of a second — the same eviction, the same box coming out into the hand, run roughly a million times faster. That ratio is the reason a chapter about your cortex sits in a book about machines. When an agent writes, it writes at the clock of the core that runs it, and the record it evicts leaves at that clock too. A person reading the transcript afterwards is the cerebellum from two sections back — measuring, correcting, always one crossing behind — and care does not close a gap of a million to one; only a witness at the writer's clock does. There is exactly one thing in the room that already ran at that clock: the slot, holding what was there before the write arrived. That is the whole of what the rest of this book means by a receipt. Not the log the writer produced, which is the account from inside. The box that came out into the hand — the one record the arriving write did not author, because it was already sitting in the slot when the write got there.

Read this in the book, in context →

You already have a control environment, and it is not a bad one. The prompts are versioned. The models are pinned. The eval suite runs in CI and goes red when it should. Every action the agent takes is written down with a timestamp and retained for seven years. Ask the industry what more you could possibly do and you get a longer version of that same list.

There is one property that list does not have, and no amount of doing it better will add it. Every record on it was produced by the process it describes. The agent reports its own actions. The eval was written by the team that wrote the thing it grades. An account of a run, produced by that run, cannot contain what the run never noticed — and that is not a claim about anybody’s honesty. It is a claim about where the account came from.

Which is a solved problem, and it was solved in 1494. Double-entry bookkeeping never asked a merchant to be honest — nobody has ever claimed it did. It required the second entry to come from somebody who did not write the first, and that one requirement is why a stranger will lend you money. Your AI keeps one book. So does everyone’s.

The industry’s answer is to build better first entries: more evals, more guardrails, another model watching the first one — each of them written from inside the same boundary as the thing it watches. We went the other way and put the instrument on the far side of it, so the record is one the agent did not author. That is the entire difference, and it is why the claims that follow sound outlandish coming from a small shop. It is not cleverness. It is where the instrument is standing.

And if it is wrong, it is wrong in one specific place, so here is the sentence to swing at. Name one record in your stack that was not produced by the thing it is about. Reply with it and you have taken this apart in public — and we will publish that you did.

Insurance never prevented a single fire; it made fire a number. Your logs already say something happened — what they do not say is whether it was allowed.

Open this commit's attestation — verify it yourself →

Why we believe this matters: the difference between what a system says it is doing and what it is doing has weight — that gap is where every AI failure and every uninsurable liability lives. But the same measurement, read the other way, is the most personal thing in the book: it means you are not about to be averaged out by a generalist. That is what today's passage was doing, and it is why the receipt above exists: the gap is measurable, so it is priceable.

Rice’s theorem says nobody can prove your agent is good. We have never claimed to. We made where it went countable instead — and there are two doors on the next page, only one of them yours.

Yes. A decidable, still-semantic spec boundary for every autonomous deterministic AI agent.

If it’s debatable, it’s not insurable. If it’s not insurable the roads don’t get built, so the cars don’t get better either — for every semantic case that actually matters.

And because this is the business of AI: buy at iamfim.com. You will sleep like a baby. For the discerning sceptic — fork github.com/wiber/thetacog-mcp and redpill your CLI in far fewer than twenty-one questions.

Markets do not underwrite safety. They underwrite checkability. The count is live.