governance: steward places REVIEWED-83 Amendment 1

Placed by the steward 2026-08-01. Committed by the executor to get it off one disk —
preservation, not modification (Constraint 1; /wrap-up §6.5 boundary).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WuMjg3ipEVa3n8CoSzoyvc

REVIEWED.md

@@ -858,6 +858,17 @@ brief. No work is blocked meanwhile; `wake-digest.py --brief` remains the fallba
 **If corrections satisfied:** land as a MINOR supersession of the §Tiering & Fence evidence-tier table (new V-DPDF row, independence stated as a requirement) plus `graduation-spec.yaml` declared data (discriminator, thresholds, extractor pair). Prior version frozen byte-identical; bounded-diff proof at landing. Tag commits REVIEWED-83.
 **Awaiting:** steward placement, then the Q3 demonstration before Harrison's graduation stamp releases.
 
+### REVIEWED-83 — AMENDMENT 1 (2026-08-01): Q3's first half was affirmed on a control that could not have detected the defect
+**Date:** 2026-08-01
+**Decision:** AUTHORIZED — amendment placed. Q1, Q2, Q4, Q5 stand unchanged. Q3's *second* half was already corrected by REVIEWED-84; this corrects the *first*.
+**What Q3 affirmed:** *"The classifier's positive/negative control (16/44 in-corpus; 62/59/164 across the 385-book library) is complete and sufficient for the tier-routing question."*
+**What was measured (2026-08-01, closing PENDING-85):** it is not sufficient. Two of the sixteen born-digital verdicts are wrong, and both works hold canonicals, so the in-corpus figure is **14/44**, not 16/44. `the-arcades-project-walter-benjamin-pdf` is a scan OCR'd by Adobe Paper Capture with ClearScan; `tschichold-form-book` is a scan OCR'd by ABBYY FineReader and re-typeset. Both were confirmed by eye at page level, not inferred from metadata.
+**Why the control could not have caught it — and the shape is the one REVIEWED-84 already named.** The classifier's controls exercise its *decision rule*: given three signals, does it decide correctly? They cannot test whether three signals are *enough*. For OCR-then-re-typeset PDFs the triad's third leg is structurally absent — ClearScan discards the page bitmap, so there is no page image at any size, and ABBYY re-typesets into real embedded fonts, so no font signature betrays it either. REVIEWED-84 found that adding independence cannot fix an operator that discards position. This finds that adding controls cannot fix a triad that lacks a signal. **In both cases the control was correct and sat at the wrong layer.**
+**A consequence for the classifier's own design premise.** Its docstring rules that *"metadata is testimony; structure is evidence."* For this family that is inverted: structure carries nothing and the producer/creator string is the only cheap discriminator. Any remedy adds a fourth signal — producer/creator strings plus an OCR-error probe over the extracted text — never a threshold adjustment. The threshold hypothesis carried in PENDING-85 was tested and refuted.
+**Consequential correction, routed not applied.** The `0 of 17` two-column exposure quoted in ratified spec v2.9.0 and in `chamber-library/CLAUDE.md` rests on this classifier. Re-running it gives 16, not 17, and two of those sixteen are OCR'd, so the honest figure is **0 of 14**. **The zero is unaffected** — removing members from a population cannot create a two-column instance — so REVIEWED-84's argument strengthens and only the denominator is wrong. Correcting ratified text is not the executor's act; recorded here for the steward to route.
+**What remains the jurist's to rule, and is not decided by this amendment:** whether V-DPDF ratification now requires the classifier to be re-demonstrated against the OCR-then-re-typeset family before the tier routes any work.
+**If AUTHORIZED:** proceed. Tag commits REVIEWED-83-A1.
+
 ## REVIEWED-84 — PENDING-87 — Order attestation: REVIEWED-83 Q3 corrected, not merely reformulated (design-gate passed with required conditions)
 **Date:** 2026-07-29
 **Decision:** AUTHORIZED — proceed per the ruling. This ruling corrects REVIEWED-83's own Q3, on the jurist's error, not the executor's.