ADR-0039: ADR-0031 cites a deleted study objective; its ruling stands and the observation moves

Status: Accepted — amends ADR-0031, which stands Date: 2026-09-10

Decisions

D-93 ADR-0031's citation of study objective R-04.2 is void, and the observation behind it survives on R-24's surviving study objective instead

ADR-0031's Consequences say a higher Not Stated rate "is intended behaviour and is itself a finding for study objective R-04.2". That objective does not exist. The ruling is untouched and the observation is untouched; only the objective it was attached to changes.

What it is now a finding about: R-24's second objective, "Study the problems and needs of workers accessing chemical safety information." A Not Stated rate measures the gap between what a worker asks and what the Corpus holds, which is a finding about worker needs and about Corpus coverage.

What it is not evidence for. The deleted objective was about the benefits of using AI. The surviving one is not, so a Not Stated rate says nothing about whether extraction is a good approach. Quoting it that way would repeat the original error in a new costume.

What each amendment touches

• ADR-0031 — every decision stands, D-65 through D-71 unchanged. The router still routes, the two tiers are still two tiers, and nothing about Routing's behaviour moves. One clause in its Consequences section is void as written and re-landed here.

Context

R-24 replaced R-04's five objectives with three and, in doing so, deleted the objective formerly numbered R-04.2. R-24 says so in its own text: "The surviving study objective is about worker needs, not about AI efficacy."

ADR-0018's D-33 recorded this on 2026-09-02 and enumerated the five ADRs whose reasoning cited the dead objective. ADR-0017 had already corrected ADR-0009 for the same citation before that.

ADR-0031 then cited it again on 2026-09-07, five days after the correction was written into an Accepted ADR. This is the third time this repository has shipped a claim that was recorded as wrong somewhere nobody re-read — after the R-03 Coverage rows, which produced ADR-0035 and the IDSTATUS gate, and the assertion that jsqr was already a dependency, which produced an open map ticket.

The pattern is not carelessness about this particular id. It is that a correction living only in prose is read once, by the person who writes it. ADR-0035 closed the version of this problem a gate could see. This one a gate cannot see: no check can tell that a sentence cites an objective that a finding deleted, because the objective is not an id the checker tracks.

Decision

The clause is void as written. The observation it carried moves to R-24's surviving study objective and the spec states it there. ADR-0031's rulings are untouched.

Rejected options

• Void the clause and claim nothing further — this is ADR-0018's precedent applied literally, and it was rejected because the observation is true and load-bearing. It is the reason the spec can say a higher Not Stated rate is "intended behaviour, not a defect to tune away", which is a standing instruction to whoever evaluates this system. Deleting it without replacement removes that instruction, and someone later tunes Not Stated down believing it is a quality problem.
• Re-land it without voiding anything — not available. The clause is in an Accepted ADR's Consequences prose, and D-83 permits editing only Coverage Note cells. Any re-landing has to be said in a new document, which is this one.
• Correct ADR-0031 in place — forbidden by the same rule, and by the repository's standing rule that an Accepted ADR is superseded or amended, never edited.
• Extend ADR-0018's D-33 list to name ADR-0031 — rejected for the same reason: ADR-0018 is Accepted. It also would not have helped, since the list already existed and was not read.

This ruling may not be re-decided

If a change contradicts this ADR: stop and raise it. Do not implement over it.

Specifically: do not cite R-04.2 anywhere — it is not a live objective and has not been one since 2026-09-02. Do not quote a Not Stated rate as evidence that the extractive approach works; under R-24's surviving objective it is evidence about coverage and about what workers ask, and nothing else.

Consequences

What this does not fix. Nothing prevents a fourth instance. IDSTATUS gates whether a cited R- id is current, but this citation names an objective inside a finding, and no gate reads prose for claims about a source's contents. The honest position is that this class of defect is caught by adversarial review and by nothing else, which is an argument for running the review regularly rather than once.

A count worth stating plainly. Three claims have now shipped that were already recorded as wrong elsewhere in this repository. Two produced gates or tickets; this one produces neither, because it is not mechanically checkable.

Coverage

Upstream	Landed in	Evidence	Note
R-24	D-93	the finding that deleted the cited objective, and named the surviving one as being about worker needs	supersedes R-04; ADR-0018 D-33 enumerated the five earlier ADRs, this adds the sixth