The rule, exactly. Before an election each citizen chooses: to vote — or to take a payment and not vote in that election. The payment is a percentage of the median income, set by referendum; there is no fixed sum. One ballot is one vote, with no multiplier in the count; the vote is amplified only in that each ballot's share grows as others step out. The budget pays under law; a candidate never pays. Only a referendum of all citizens — simple majority, no quorum — introduces, changes or repeals the rule.
The protocol has been introduced nowhere and no pilot has been run: shares, turnout and outcome figures in the chapters are estimates, and the protocol promises nobody an election victory. If a chapter says otherwise, Exact Answers and the Charter are correct. For a candidate: ten questions and ten steps. For a citizen, a mayor, a finance officer, a donor, a journalist, a scholar, a lawyer: answers by role. Everything in force in one file: llms-full.txt.
23b. Rules for handling evidence¶
Chapter: 05 — Empirical base File: 05_023b · v1 · 22 September 2026 Source: three corrections made by the architect in the course of work on country chapters, generalised here into transferable rules
This chapter is not about what we know but about how we handle what we do not know. All three rules arose the same way: the assistant made an error of a particular kind, the architect named it, and the error turned out to be systematic rather than isolated. Each rule is stated here in general form; the country case in which it was found appears as illustration, not as content.
The rules are ordered by what each protects against: the first against substituting a caveat for work, the second against adopting someone else's category, the third against an unfit source.
23b.1. The adjacency rule: what replaces the caveat "no data exist"¶
The error. The protocol describes a mechanism that has existed nowhere. From this the assistant derived "no data exist" twelve times in a single chapter and marked the text as unverified forty-three times. The architect's remark:
"We are reasoning about something that NEVER EXISTED, and yes, a pilot is best. BUT in forecasting we must minimise the error of our assumptions and look for confirmed indirect or adjacent entities. And you write me the same thing every time — that is inefficient."
What is wrong here beyond the repetition. The absence of direct data about something that has not happened is a constant, not a finding. It is known before the work begins, nothing follows from it, and reporting it in every section occupies the space where an argument should stand. Worse: the caveat substitutes for the search. In that very analysis the nearest confirmed analogue lay inside this same repository and was never opened — which came to light only after being pointed out.
The rule. Where direct verification is unavailable, no caveat is written until the nearest observable analogue has been found and how far it carries has been stated. The form of the record has three fields:
| Assumption | Nearest confirmed analogue | How far it carries |
|---|---|---|
The third field is obligatory and must state a boundary, not praise for the analogue: precisely how our case differs. An analogue without a boundary is worse than a caveat, because it passes partial similarity off as confirmation.
The architect's correction to the rule's first edition:
"'Does not carry' means worse by a percentage, not equal to zero."
The remark is right, and the error was in the wording. The first edition wrote the third field as a binary: "carries" or "does not carry". A boundary written that way works as a cut-off — and it reinstates precisely the disease the rule was invented against: if every analogue "does not carry", we are back at "no data exist", merely by a longer route.
In fact a difference in conditions is a discount on the transfer, not its cancellation. Alaska voted in 1976 in a democracy, and that does not mean its two-to-one margin says nothing about people and rent; it means it says less, and how much less must be stated.
Hence the requirement for the third field: it is written not as a negation but as three answers.
| What to state | Why |
|---|---|
| What transfers — which part of the claim survives the gap | usually the existence of the phenomenon survives while its magnitude suffers; the two must be separated |
| Which way the gap shifts it — does it make the analogue optimistic or pessimistic for us | the most useful field: the direction of the error is almost always knowable even when its size is not |
| How much remains — on an ordinal scale, without invented percentages | "almost entirely · largely · half · direction only · the existence of the phenomenon only · ≈0" |
The phrase "does not carry" is separately forbidden. One writes "carries this far". Zero transfer remains an admissible conclusion, but it must be argued for, not obtained free of charge by a choice of word.
And the ban on invented precision holds. The ordinal scale was chosen precisely because "transfers at 60 %" would be a figure picked at random — the very error that produced this rule. The direction of the shift can be known; percentages cannot, until something is measured.
Where a caveat is still appropriate. Where no analogue exists at all. It is then written once per chapter, names by name those assumptions that lack an analogue, and is not repeated at each of them. In the September 2026 analysis, after a full enumeration, two such assumptions remained out of six (056f.13).
The order of search, nearest to furthest: our own repository → a measured relationship in another jurisdiction → an observable historical case → a laboratory result → general theory. One stops at the first that carries; descending further is permitted only with an explanation of why the higher ones did not fit.
The by-product for which the rule is worth keeping. Searching for an analogue uncovers errors that a caveat never uncovers. In that same analysis it exposed a false citation that had stood in 014 for years: a work with non-existent authors and figures that do not agree with checkable ones. The caveat "data are scarce" stood beside that citation and noticed nothing.
23b.2. The foreman test: telling real consent from consent as labelled¶
The error. The assistant wrote about signature collection as though a signature were self-evidently an act of dissent. The architect objected with a scene:
"Where are the dissenters here? A fitter, Vasya, comes to the foreman for advice: I'd like to get money at the elections, he says, I don't vote anyway. And how is the foreman, man to man, to explain that voting is an expression of dissent from our successful government in the country with the largest subsoil wealth per citizen?"
What is wrong here. The assistant adopted someone else's category. "Dissenter" is a designation issued by the side whose behaviour we are analysing; it does not correspond to the content of the act. The man is not protesting, he is asking about money. Having accepted the label, the analysis inherits the whole frame in which the label makes sense, and then reasons inside it.
The rule. A measure is fit if the immediate superior cannot explain to his own subordinate why not to do it. The test is placed not at the level of the regime and not at the level of the law, but at the link where coercion is actually transmitted: the foreman, the shift leader, the HR officer, the commanding officer, the head teacher. That link has to answer personally and without a script.
Why that link. The drive belt of mobilisation rests on the superior voicing a requirement and the subordinate performing it, both understanding what is happening without words. The test works when voicing the requirement aloud and with a justification becomes costlier than staying silent. A superior who has only two answers, both losing, ceases to be a transmitter.
In the case analysed there are exactly two answers: "the state does not want to give you your share" — that is, repeating the protocol's argument in one's own words — or "asking for money is disloyal" — that is, admitting aloud what loyalty consists of. There is no third. This is a fork of the same construction as those collected in 001c: a move exists, but every move worsens the position.
The strongest part of the test is where the argument is taken from. It is taken from inside the official picture of the world, not against it. "The country with the largest subsoil wealth per citizen" is a thesis of state pride, not an oppositional one. The question "where is my share" takes the boast literally. The boast cannot be withdrawn, and the question following from it has no answer. The general form: a measure is the more robust the more fully it is derived from what the authorities themselves boast of.
What the test does not protect. It removes the frame, not the danger, and that distinction must be held firmly. Prosecution attaches not to the content of the question but to whoever asks it in an organised way: laws that label organisations make vulnerable anyone who touched a campaign, however innocuous its subject. The foreman test protects the participant and does not protect the organiser.
23b.3. The fear rule: when a survey is not data¶
The error. The assistant relied on two survey figures from a country in which the protocol is described as illegal. The architect's objection:
"Why do you give me statistics from a place where surveys are taken under fear? They are irrelevant."
What is wrong here, and it is an internal contradiction rather than carelessness. The whole construction rests on Kuran: in regimes of this type people systematically conceal their preferences — which is exactly why the protocol needs a way of reporting the truth anonymously. To rely thereafter on a survey of those same people is to apply an instrument whose unfitness one has just demonstrated oneself. The error is not in the figure but in its contradicting the text's own foundation.
The rule. Where an answer can cost the person answering, survey data are not a measurement. They are admissible only as a bound, and the side must be stated.
How to determine the side. The distortion is neither even nor random: it points towards the safe answer. Hence directly:
| What is asked | Safe answer | What the figure gives |
|---|---|---|
| Do you do something permitted but disapproved of | "no" | a lower bound on the share who do |
| Do you support the authorities | "yes" | an upper bound on support |
| Do you watch what is banned | "no" | a lower bound, very crude |
| Name whom you read | name someone safe | almost nothing: a name's absence says nothing about audience |
The last row is the easiest trap: somebody's absence from a list of names given to a stranger on the telephone reads as absence of audience, when it means only that the name is dangerous to say.
What to use instead. Behavioural data: traffic measurements, metered panels, view counts, download statistics, the fact of a block. These do not ask a person, they observe one, and so do not depend on what he fears.
A specimen of reasoning the rule permits. A platform is throttled for two years and its daily reach grows over that period — so circumvention is widespread; that is an observation, not a claim. But it yields no percentage, and a handsome figure must not be slipped into its place: replacing an unverified forty per cent with an unverified eighty is the same error with the sign reversed. The honest record runs: the share exceeds such-and-such, the exact value is unknown, the question would be closed by such-and-such sources, and they were not found in open access.
The rule's own boundary. It does not mean a survey is always useless. In a jurisdiction where an answer costs nothing, a survey is an ordinary measurement with its own known biases. The rule switches on where the respondent has reason to be afraid and switches off where he has none; applying it to all surveys indiscriminately is the same substitution as using their results without correction.
23b.4. How the three rules work together¶
The order in which they apply to any claim:
| Step | Question | Rule |
|---|---|---|
| 1 | Whose category am I using? | 23b.2: do not accept the label of the side whose behaviour you are analysing |
| 2 | Where do the figures come from, and could the source have been afraid? | 23b.3: a survey under fear is a bound, not a quantity |
| 3 | Direct verification is unavailable — what is the nearest thing that exists? | 23b.1: an analogue with a boundary, and only then a caveat |
The order is not arbitrary. Step 1 comes first because someone else's category spoils the question itself, after which one may check figures as carefully as one likes inside a wrong frame. Step 2 precedes step 3 because an unfit source creates the illusion of available data and thereby cancels the search for an analogue: while a figure looks measured, nobody looks for an analogue.
What the three have in common is what they protect against. Not ignorance: ignorance is irreducible and honest. They protect against three ways of disguising it — a repeated caveat, someone else's frame, and a figure that looks like a measurement and is not.
23b.5. Weak point¶
The rules are derived from three cases in work on two country chapters, which is few. That they transfer is an assumption resting on the character of the errors rather than on any check of their frequency in the rest of the repository; no continuous revision of older chapters against these three rules has been carried out, and how much text fails them is unknown.
What the revision of the tables produced. Both earlier tables — 056f.13 and 057e.5 — have been rewritten into the new form, and rewriting them changed two readings rather than merely their wording: the Alaskan turnout measurement and East German common knowledge turned out to shift in both directions at once, which the binary form had no way of expressing. That is an argument for the rule rather than an ornament: the shape of the field changes the conclusion.
Second, and this concerns 23b.1: the rule demands a search for an analogue but supplies no criterion for when the search is finished. "Did not find" and "does not exist" differ only by the searcher's diligence, and diligence cannot be checked — precisely that difference produced the false citation which the rule later exposed. The only available defence is to state where one looked, so the reader has something to contradict; it is a weak one.
Third, concerning 23b.2: the foreman test reasons about what a superior can say rather than observing what he does say. We have no recorded conversation of that kind, and the confidence that no third answer exists rests on an enumeration made by the assistant. The enumeration may have been incomplete. 🟡
Related: 056f.13 (origin of the first rule and the table of analogues) · 056f.7 (the foreman scene) · 056f.10 (the analysis of sources) · 001c (forks with no good move) · 014 (Alaska as nearest analogue and the false citation inside it)