3.5 KiB
REVIEWED-98 — PENDING-114 — Scripture quoted inside a host text, unmarked: authorize the validated census
Date: 2026-08-08 Decision: (b) AUTHORIZED, with (c) attached exactly as proposed for whatever (b)'s validated recall cannot reach. (a) — fix the Harrison span and stop — REJECTED. Provenance: steward, 2026-08-08, on the executor's PENDING-114. Reasoning recorded here rather than compressed, per PENDING-108/113: a ruling that exists only in conversation evaporates.
Ground 1 — (a) would repeat, in miniature, the error already named in this same thread. Fixing Harrison alone treats a class-shaped problem with a span-shaped fix. That is not an analogy to 118f411; it is the same taxonomy error, identified hours earlier the same day and now written into the register against it.
Ground 2 — the scale context, which PENDING-114 understates. Harrison's quotation is invisible to any sidecar-based detector because it belongs to the intra-line class measured 2026-08-07 at ~6,455 marked quotation runs across 8 sources, ~94% of them intra-line and therefore not expressible at the current section granularity. Harrison is not an isolated miss — it is the first confirmed hit in a population that is this corpus's blind spot for precisely the property the corpus exists to guarantee. First-order gap, not a curiosity.
Ground 3 — why (b) rather than merely thorough. (b) is built so as not to repeat the marker census's own failure. It does not propose trusting a signal; it proposes demonstrating recall against a hand-scored known-positive / known-negative pair — Harrison's actual quotation against Weil's mentions — before any corpus-wide claim is made. That is the discriminating-instance discipline (REVIEWED-83 A1) applied to the instrument before it is applied to the corpus.
Ground 4 — the payoff, and the proposed doctrine earning its keep forward. Whatever (b) finds is marked, not fenced. That is a materially different act from 118f411: attribution removes nothing from the test population, it improves it. A correctly-marked Harrison/Mark instance becomes a better positive control than an unmarked one, not a destroyed one. REVIEWED-96's proposed doctrine — a fix that enforces a property can destroy the population that tests it — is here running forward in time, as a design constraint on remediation rather than as a post-hoc finding. This is the first occasion on which it has done real work.
On (c) as the attached fallback. If the validated detector's recall does not reach some part of the class, disclosing that limit is a legitimate stopping point, not a failure to execute (b) properly — the same standing (c) has in REVIEWED-48's gate-to-abstain, which was ruled doctrine-consistent rather than merely tolerated. (c) is not a consolation branch; it is the honest-degradation route (Constraint #4) and is authorized on its own terms.
Binding on execution. No corpus-wide claim — finding or all-clear — may be reported from the detector until its recall is demonstrated on the hand-scored pair. The marker census already run is a failed instrument, not a count, and is not to be cited as either evidence or reassurance.
Scope. Disposes PENDING-114 only. REVIEWED-96, REVIEWED-97 and PENDING-115 are untouched; PENDING-115 remains a separate blocker on remediation step 3.
Pacing. Execution timing left to the executor as a proportionate-to-energy call, explicitly not decided by the substantive case above.