ADR-065: A reading's promote_when asks only for what its title asserts

Context

An Active THEORY asserts its title: the THEORY statuses say whether the record believes the account, and ADR-059 makes every status credence in what the entry itself asserts. So a promote_when should ask for whatever would warrant that assertion, and no more.

ADR-047’s section “What one study counts for” (2026-10-19) said so for one case. An empirical THEORY may be promoted on one study read in full, when its title keeps that study’s hedges and nothing contradicts it. A demand in its promote_when for replication moves to a QUESTION. ADR-047 is Superseded and ADR-057 carries its rulings forward, so the rule now lives in a superseded document, worded as a step of a promotion audit. Intake did not apply it. Readings were filed Proposed with conditions copied from the general case. On 2026-10-30, 462 of the 597 Proposed THEORYs were still at version 1, and the worklist surfaces only entries something relies on.

A survey on 2026-10-30 classified every Proposed THEORY’s promote_when. Its scripts and data are in the session’s scratchpad (pw/), and its counts are estimates from keyword flags, with the cases below read by hand.

• Most conditions are right. Roughly 85–90% of titles assert a general claim. For those, a study nobody has run, a replication, or a specific further work is what would warrant it, and a reading cannot supply it. That is the difference between reading a source and believing it.
• About 58 ask for more than their titles assert.
  • 21 one-study titles keep the study’s own hedges and still ask for replication, or for the very thing the title concedes. THEORY-329’s title says “that anger causes the spread is not shown”. Its condition asks for a randomized study.
  • 19 attribution titles (“X proposes”, “On X’s analysis”) ask that X be right. One example is THEORY-1084. The endosymbiosis batch filed on 2026-10-27 is another. THEORY-1110 already shows the right shape: “the authors propose” is Active once read, and whether they are right stays with THEORY-070.
  • About 10 proofs say the condition is a checked re-derivation already recorded in the file. THEORY-924 is one. Three of them also ask for the version of record, with no discrepancy named (THEORY-943).
  • 8 need their titles narrowed first.
• Demands for the version of record split. Where a discrepancy between versions is named, the demand is justified (THEORY-931’s inconsistent table). Where none is, and the copy read is the author’s version of the same work, it asks for nothing the title needs.

Decision

1. A promote_when asks for exactly what the title asserts. This holds at filing and at every review.
2. A reported title goes Active on its reading. That is a title that says what a source proposes, argues or finds, or what one study measured with the study’s own hedges kept. It goes Active once a NOTE reads the source in full and nothing in the record contradicts the report. Its promote_when says so.
3. What a condition asked beyond the title is kept, not deleted. It moves by ADR-047’s routing:
   • A request to replicate or test the same claim independently moves to a QUESTION.
   • A wider thesis the source asserts moves to a Proposed THEORY that extends the reported entry. That THEORY keeps the old condition word for word (ADR-047 kind 1).
   • The record’s own extrapolation moves to an interpretive CLAIM (kind 2).
4. The version of record is asked for only where a discrepancy is named. That means a table, a sign or an omitted proof that the copy read may get wrong. Otherwise the NOTE records which copy was read, and that is enough.
5. A proof condition is met by a re-derivation recorded in the file, once a second checker has gone through it. The second checker follows each step and re-runs any script. A summary saying it was checked is not enough (ADR-047, “A summary that says it was checked is not enough”).
6. A general title keeps its strict condition. A title that asserts an empirical generalization, a causal claim or a theory’s truth keeps its demand for evidence beyond the source. Nothing here lets one unreplicated study promote a general claim.
7. A condition that disputes a study’s internal validity stays. It is not a demand for generality. Such an entry is narrowed to “X report…” only when that frees something (the status-review README, §2).

The status review applies this to the entries the survey found, in the same pull request that files this decision. Intake applies it from now on.

Alternatives considered

• Loosen promote_when across the board, so that a reading promotes its entry. Rejected. For a general title that would make Active mean “read”, which is what Proposed already means for a reading. The record would lose the difference between holding a source and believing it, and it would promote empirical generalizations on one study, which ADR-047’s one-study rule forbids.
• Rewrite only the conditions, without filing the wider theses beside the reported entries. Rejected. ADR-047 v4 was written because narrowing had been losing the wider claims. An attribution promoted without its “P” beside it would read as if the record held P.
• Status quo: leave the rule in ADR-047 and apply it at review. Rejected. Intake did not apply it, and review reaches only entries something relies on, so most such entries would stay misstated indefinitely.
• Put each case to the owner. Rejected, as ADR-047 rejected it. The owner delegated these rulings to the status review (2026-10-29), and the rule makes the cases mechanical.

Consequences

• About 50 readings become promotable under this rule. Each still needs a reviewer to confirm that its hedges are kept and that nothing contradicts it. Each attribution promoted gets its wider thesis filed beside it, Proposed. Almost nothing relies on these entries; the worklist’s largest reach is 1. So the gain is statuses that say what the record believes, not freed dependents.
• The risk is reading “On X’s analysis, P” as P. The paired Proposed entry carries P. Where a reported title also carries the record’s own critique, an Active status asserts that critique too, and the reviewer checks it.
• Filing gets stricter about wording. A reading whose title reports a source must say so in the title, and its promote_when must match. A reading whose title is general keeps a condition the reading cannot meet.