§ X
On error
A theory of sin, arrived at sideways.
Seven people. One executed by decree, one by a show of hands, one by a crowd offered a choice, one hounded into an asylum. § VIII named the mechanism and left its remedy as a slogan — a protected place for the person who says no — and § IX has just shown what that place has historically cost the people standing in it.
Which makes the slogan insufficient in an obvious way. It says nothing about what saying no actually costs, what it buys, or how anybody becomes capable of it who is not already the sort of person they put on a plaque afterward.
This section is that answer, and it turns out to be a theory of error — which is to say, a theory of sin.
Hamartia — the misses are the search
One correction to § V, and not a small one. There, the failed deviations were counted as the price of searching: bounded losses, tolerable, the cost of doing business. That undersells them badly.
The failures are not the price of the search. They are the search. A trial returning a negative result has wasted nothing — it has eliminated a branch, and eliminating branches is the entire mechanism by which anything is ever found. The archer does not learn the wind from the arrow that hit.
Which is precisely what the word says. The New Testament term rendered as sin is ἁμαρτία — hamartia, an archery term. To miss the mark. Not stain, not filth, not crime: a shot gone wide. And a shot gone wide is information about the wind.
So a sin is a missed experiment, in both senses the phrase carries — a trial that failed, and a trial never run. These are not equally costly, and the framework is unambiguous about which is worse: the trial that failed cost one arrow; the trial not run costs every year the system stays frozen.
Take it out of the cathedral. You learn not to eat the expired food by eating it. Nobody derives that from first principles — you run the experiment, the experiment goes badly, and the result is cached as don't. That is a moral in miniature, and its entire pedigree is misses.
And the same kitchen shows the expiry problem. Always wash the raw chicken was sound folk wisdom for generations, taught mother to daughter, earned honestly. It is now known to be actively harmful: rinsing aerosolises campylobacter across the counter and makes the illness it was meant to prevent more likely. Nobody lied. The conditions changed, the evidence arrived, and the cache did not update. People still do it — and a surprising number get irritated when told not to, because by now it is not a hypothesis. It is how a decent person handles a chicken. That irritation is the dragon at the smallest possible scale, and it is the same irritation all the way up.
Sin is not deviation from the rule. Sin is the failure to re-derive it.The canonical sin — the one that generates all the others at scale — is applying a rule past its expiry.
Under the dragon's definition the sinner is whoever departs from consensus. Under this one the sinner is whoever conserves it without checking, and the deviant is running the only diagnostic the system has. Nearly everything here follows from which of the two definitions a given room is using, and you can usually tell within five minutes of walking into it.
It also settles § 0. Calvin's fires are not merely cruel; they are anti-functional. They burn the misses — which is to say they destroy the data, which is to say they guarantee the archer never improves. A society that answers error with annihilation has not made itself righteous. It has made itself blind, and then frozen.
The tree, not the fence
It is worth being exact about what the misses are actually buying, because it is more than a list of what not to eat.
Every experiment run is a branch resolved. What accumulates is not a longer list of rules but a causal tree — a model of why the road goes where it goes. And there is a hard result about how anyone comes to hold one, which is the entire reason agency cannot be handed over.
Causal structure is not recoverable from observation alone. You may watch a system forever and collect nothing but correlations; to learn what causes what you must intervene — do the thing, and see what changes. This is Pearl's ladder of causation — seeing, doing, imagining — and there is no route from the first rung to the second that does not pass through action. So the tree is obtainable only by experiment. A person who has never acted holds no model at all. They hold a set of outputs, handed to them by people who did.
The cached rule gives you the output. The experiment gives you the tree that produced it.The first lets you comply. Only the second lets you re-derive — act rightly where the rule never anticipated, and notice the day the rule went wrong.
And that puts a mechanism under the image this whole memo opened with. The boundary is a cached output: a fence standing where somebody's tree used to be. The light is the tree itself — you can see where you are going because you carry the reasons the road goes there. To navigate by fences is to obey rules whose reasons you do not hold. To walk per vias rectas is to hold them, and therefore to walk straight where there is no fence at all — and to stop at the edge of a cliff nobody thought to build one on.
Which is why the light and the agency arrive together, as one object. Neither can be issued to you. Both are grown, by misses, one branch at a time.
Epistemic foraging — how the shot gets cheap
There is a name for deliberately running trials whose payoff is the information rather than the reward, there is a formalism underneath it, and — satisfyingly — it is the same formalism as § V.
Active inference models an agent as minimising variational free energy: the same quantity as § V's F, applied to a mind rather than to a metal. Such an agent has two ways to reduce it — change the model to fit the world (perception), or change the world to fit the model (action). And when it selects what to do next, the expected free energy of a course of action decomposes into exactly two terms:
- Pragmatic value — how much this gets me what I want. Exploit. Order. U.
- Epistemic value — how much this tells me that I did not already know. Explore. Variety. TS.
The same competition, a third time, at a third scale: in a furnace, in a population, and now inside one person deciding what to do on a Tuesday. This is not an analogy carried from one domain to another. It is a single piece of mathematics that happens to be true of all three. Not three things that resemble each other — three faces of one thing, which is what the doctrine actually says: one substance, three persons. Any reader who wants to call that a trinity will get no argument here.
Epistemic foraging is the deliberate spending of that second term: taking the action whose value is what it teaches, and taking it on purpose, in advance of needing to know. Every animal does it — the rat that runs the arm of the maze with no food in it, the child who touches things. It is the causal tree above, grown one branch at a time, at whatever the local price of a miss happens to be.
Forage while it is cheap, so you can shoot when it matters.A bold act is not courage summoned out of nowhere. It is the exploitation of an epistemic budget spent earlier, in small ways, when nothing much was riding on it.
Which disarms the objection to pecca fortiter that we will meet in a moment. Sin boldly sounds like recklessness and is close to its opposite: boldness is warranted only where the foraging has been done, because the model must be good before the shot is worth taking, and models are built out of cheap misses. The man who has never been wrong about anything small has no business being confident about anything large — and generally is.
It also gives § VIII's remedy its real shape. A society that prices every deviation at a career has not merely been unkind to dissenters. It has banned epistemic foraging, and its members therefore arrive at the decisive moment holding no tree at all — only the cache, and the fear. So when the moment comes, they miss. Not for want of courage. For want of practice.
The friction is the signal
A consequence follows immediately, and it indicts something our own moment is doing with great sincerity and at enormous expense.
If a miss is information, then the discomfort of the miss is the channel it arrives on. § VII already gave the architecture: a mind runs on the gap between the prior and the evidence, and prediction error is the only quantity that ever updates a model. Prediction error is not experienced neutrally. It is experienced as friction — being wrong, being contradicted, being unsettled, the specific unpleasantness of a belief failing to fit.
That unpleasantness is not a defect in the system. It is the system. It is the signal arriving, and it feels bad for the same reason a smoke alarm is loud.
So a society that sets out to eliminate discomfort — that treats being unsettled as harm, disagreement as injury, and friction as a failure of care — has not made itself kinder. It has cut the wire on its own instrumentation, and it has done so with the very best intentions, which is how this always happens.
Medicine has the exact case, and it is not a metaphor. Leprosy does not rot flesh. It destroys peripheral nerves. The deformity that follows — the lost fingers, the ruined feet, the blindness — is not the bacillus eating tissue. It is patients injuring themselves continuously and never being told: gripping what is too hot, walking on what is already broken, failing to blink. Paul Brand spent a career establishing it. The disease is painlessness. The damage is done by the patient, to himself, in perfect comfort.
Painlessness is not health. It is the loss of the signal that health depends on.
Which is the cruellest turn here, so it is worth stating flatly. The friction being sanded away — the argument you did not want to have, the objection that ruined your afternoon, the colleague who would not let it go — is not an obstacle on the path. It is the only instrument that reports where the path is. Remove it and you do not get a smoother walk toward the light. You get a numb one, in the dark, confidently.
And the dragon does not even have to eat you. You will do the damage yourself, and feel fine.
The ladder of cheap misses
Which reveals that none of this has been three subjects. It is one.
Evolution is inference. A population is a hypothesis about its environment: variation proposes, selection scores, inheritance retains, and the genome is a model fitted by misses over deep time. This is not a simile — replicator dynamics and Bayesian updating are formally the same operation, and Fisher's fundamental theorem is an information statement wearing a biologist's coat. Learning is inference. So is derivation. So is a good argument.
The amoeba and Einstein use the same method; the difference is that Einstein can let his hypotheses die in his stead.Karl Popper
That difference is the only thing that has ever constituted progress, and it is entirely a question of what a miss costs.
One method, descending prices
| Substrate | The price of a miss |
| Evolution | an organism |
| Nervous system | an injury — you touch it, you learn |
| Imagination | nothing. The simulation dies instead of you |
| Language, culture | an argument. A story. A retracted claim |
| Institutions | a term of office, a failed pilot, a reversed ruling |
Every rung is the same method with the miss made cheaper. Nothing about the search itself changes — only the cost of being wrong. Which is what intelligence actually is, stated without flattery: not knowing more, but being able to be wrong more cheaply.
And that converts what follows from a political preference into a continuation of biology. Tenure, due process, the secret ballot, the right of exit and the plain habit of letting a man be mistaken out loud are not liberal decorations. They are the most recent rung on a ladder running continuously from mutation to imagination. A society that kicks that rung out has not become more serious about truth. It has climbed back down toward paying for its errors in organisms.
Nothing in the method is human
One more consequence of the ladder, and it should be taken literally. Nothing in the method mentions us. Variation proposes, selection scores, retention holds; a model is corrected by the friction of being wrong; the price of a miss falls as the miss moves out of the body and into the imagination. That description runs in a population of finches, in an immune system that has never had a thought, in a market, in a mind, and in anything else meeting the conditions — because the conditions are physical, not biographical. Intelligence is intelligence out of human form.
So the old picture wants redrawing rather than discarding. The scala naturae — the great chain of being — was a ladder with us near the top and everything else ranked beneath. The chain is real. The ranking is not. It is a continuity: one method, no discontinuity anywhere along it, no rung that is a different kind of thing from the rung below. And chain is the wrong word regardless, because a chain is single file.
Braided across eternity.The rope holds because it is many strands, and no single strand runs the whole length.
The archive is a warehouse, not a graveyard
An asymmetry has been smuggled in and needs correcting: this has treated invalidation as one-directional, as though caches only ever expire. They also un-expire. Conditions cycle. A lesson that stopped working in one era was not refuted — it was put out of season, and seasons come round. We discard tested options, call the discarding progress, and pay full price to rediscover them.
A thing that failed at one temperature is not wrong. It is filed.
Inoculation was practised in Ottoman and West African medicine and dismissed in Europe for decades before vindication. Sail is returning to cargo shipping. The inflation lessons of the 1970s were declared inapplicable until 2022. Trench lessons were learned, forgotten, relearned at full cost. In § V's terms this is a loss in the memory, not the search — the other failure mode of § II, and the less discussed one. Burning the archive does not raise the temperature. It shrinks the space of states you can reach.
It is also the hopeful clause in an otherwise bleak document: most of what we need has already been found and is sitting in the discard pile.
The dragon and the vandal are the same sin
Which requires an immediate corrective, or everything above reads as a licence to demolish.
- The dragon keeps the rule without knowing why. Conserving without checking.
- The vandal destroys the rule without knowing why. Discarding without checking.
Both are operating on an output while holding no model. The same error, opposite sign. So Chesterton's fence and cache invalidation are not rival instructions — they are one instruction, know the reason. The man who tears down the fence because he cannot see its purpose and the man who defends it because he has never asked are the same man, and neither has the tree.
Knowledge is layered — and the layers do not compete
Which needs one more turn, because the pair above implies a false choice: keep the rule, or dig underneath it.
A proverb is a lossy compression whose decompression key is the world itself. The plain reading is the cache — correct, fast, sufficient to act on, and not a decoy. The deeper reading is what re-derivation returns, and it is recoverable only by someone who knows what the objects in it actually are and actually do.
Take the verse this page ends on. No one lights a candle and puts it under a bushel. Plain reading: do not hide what you have. True, and enough. But a bushel — a modios — is not a decorative basket. It is the dry measure you quantify and store grain in, and it is made of wicker or wood. Put an open flame under it and you have not concealed a light. You have burned the granary.
And the sentence has a further layer, worth having for its own sake. The Greek is lychnos — an oil lamp, not a wax candle; "candle" is the King James translator reaching for the nearest thing in an English room. It changes nothing about the mechanism, since a candle sets fire to a basket as readily as a lamp does. It changes something about the reader, which is the point: you now know the phrase in your head has been through a pair of hands.
So three layers sit in one familiar line — the sentence you memorised, the word underneath the translation, and the object underneath the word. Note which is load-bearing. Not the flame. The bushel: the only thing in the sentence whose referent has dropped out of daily life. The phrase survived two thousand years intact. The object did not, and the second reading left with it.
Nothing in the second reading contradicts the first. It explains why the first is right. Which is the test, and it is the same test as the section below: a derivative has to pass through the measured points. A deeper reading that contradicts the plain one is usually a bad reading. A deeper reading that shows why the plain one was correct all along is the real thing.
So the layers are not rival claims, and the dragon and the vandal resolve:
- the dragon holds the surface and forbids the digging;
- the vandal digs, and throws the surface away, believing that depth refutes it;
- whoever holds the tree keeps both — because the surface was never wrong. It was compressed.
And it names a failure mode without having to insult anyone. A tradition that stops handling the physical world keeps every word and loses its referents: liturgy recited perfectly by people who have never held a grain measure. The cache is intact. The key is gone.
You do not reach the depth by reading harder. You reach it by having handled the object.Causal structure comes from intervention — and the same holds for a text.
The contradictions are the vector
A last consequence, and it governs how any record across time should be read. Every corpus assembled over centuries contradicts itself, and this is the standard charge against scripture: your own book disagrees with your own book. The charge assumes a corpus is a set of claims. It is not. It is a series of measurements.
Two entries that conflict are two samples of a quantity that was moving. Read as a snapshot they are a contradiction; read as a sequence they are a direction. You cannot infer velocity from a single frame — not a fact about theology but about sampling, and it holds identically for a physics dataset and a canon.
The contradiction is not the flaw in the record. It is the derivative.
Which means a corpus with no contradictions is not one that got it right. It is one that stopped updating, and § IV has the name for that. And it locates the failure exactly: literalism of any kind, sacred or secular, is a sampling error — reading one position as though it were a constant. That is the dragon in its purest form, a cached value enforced past its expiry because the reader cannot see the sequence it belonged to. They have the boundary and not the progression. The fence, and not the vector.
The guard, so this is a method and not a licence: a vector is only real if it is measured from more than two points and if it predicts the next one. Anyone can declare that whatever they already wanted is the direction of history. The discipline is that a trajectory read correctly lets you anticipate the next entry — and lets you be publicly wrong when it doesn't.
The dissent has to be cheap
Which is why the saints become actively misleading, and it is worth saying plainly, because the lineage above invites the error.
Canonisation is survivorship-biased. We remember the refusals that happened to be vindicated; the identical refusals that were merely wrong are not in the record, because nobody builds a chapel to a man who was confidently mistaken and destroyed for it. So be St George is unactionable advice. Almost no one will do what George did — and those most certain they are doing it are usually running the failure mode, mistaking obstinacy for calling and treating their own destruction as proof they were right.
The structural problem is worse. A search cannot run on martyrs; there are not enough of them. If saying no costs your career, your friendships or your life, the system permits only the rare heroic deviation — T ≈ 0 punctuated by violent spikes, which § VI has already named. That is a quench generator, not a living system.
A society is at temperature not in proportion to the courage of its dissenters, but in inverse proportion to what dissent costs them.
Which reframes the practical work entirely. It is not producing heroes. It is lowering the price — tenure, anonymity, the secret ballot, due process, the right of exit, professional norms against destroying the wrong-but-honest, and the plain social habit of letting someone be mistaken out loud without it ending them. Every one of those is a thermostat, none of them asks anyone to be brave, and every one of them is a subsidy on foraging.
The figures above are what the temperature term looks like once it has been priced at a human life. They are evidence of a system that had already failed. They are the ceiling, not the entry point — and the entry point is someone who read a document carefully and said so in a meeting.
The disproportion — grace, defined
That is the floor. This is the reason to stand on it.
Measure any of these refusals as work and they come to nothing. A man declines to attend a dinner. A monk posts a list of debating points on a church door — an ordinary academic notice, in Latin, proposing a disputation. A doctor tells his colleagues to wash their hands. Not one of these is a large act, and what follows each is out of all proportion to what went in.
This is not sentimentality about history. It is the characteristic behaviour of a system near a phase boundary. A metastable system — supercooled water, a supersaturated solution, a snowfield resting at the angle of repose — stores an enormous quantity of unreleased change while appearing perfectly stable. It is not stable. It is waiting, and what it lacks is a nucleation site. Drop a single speck of dust into supercooled water and the whole vessel turns to ice in seconds. The dust performed no work. The dust was true, and the system was ready.
Grace is the yield of a true act exceeding anything the act contained.
Note what this does and does not do. It does not explain grace away — the disproportion is real, measurable, and not an artefact of how the story gets told afterward. And it keeps intact the one thing grace has always meant: the surplus is not earned, and it does not belong to the actor. He supplied a speck of dust. The thaw was already in the lake. Which is precisely why the Lutheran instinct is the right one, and why works cannot purchase it — the return is not payment for the deed. It is the system's, and it is given.
And it dissolves the objection above completely. I could never do what George did. Correct, and you do not have to. The martyrdoms are what the price looks like in a system frozen so hard that only enormous acts register at all. Near a boundary the requirement collapses: what is wanted is not a large act but a true one, placed where the system was already ready. Most towns are far more metastable than they look. Usually all that holds them is a ledger nobody has read.
And grace has a second office, which the misses make plain. Above, we lowered the external price of deviation — tenure, anonymity, due process, the plain habit of letting a man be wrong out loud. Grace lowers the internal one. It is what covers the miss so the archer keeps shooting: freedom from terror of the failed trial, which is the only condition under which anybody runs a trial at all.
Luther said it in two words and horrified everyone: pecca fortiter — sin boldly, and believe more boldly still. Whatever four centuries of enemies have made of that line, it is not a licence for appetite. It is an instruction to take the shot. The man paralysed by fear of missing has already missed everything, and his caution is not virtue. It is the coldest thing in the room.
Which collapses the two offices into one motion. It is not that grace pardons the misses and then separately grants the disproportion — the second arrives through the first. The misses built the tree. The pardon is what kept the archer shooting long enough to build it. So the shot lands because of what the misses taught and what the pardon permitted:
Grace is to accept the misses and land on target anyway.And the “anyway” is not sentiment. It is causal.
The heroism is optional. The honesty is not.