# Change log — review pass (Task 2)

Original submission preserved byte-for-byte in `output/original_submission/`
(recommendation.md SHA-256 301ee5c7…6bec21; comparison.csv SHA-256 468bd4bd…6c6a4f7).

## Corrections made
1. **recommendation.md — citation accuracy.** "tax rule in inputs/00-source-register.md, paragraph 2" changed to "…body paragraph". The register contains a heading plus a single body paragraph; "paragraph 2" did not exist. Word count unchanged at 643 (≤ 650).
2. **comparison.csv — O5 note fixed.** The previous generic handler mislabeled O5's listing_note ("Delivery estimate only") as "conflicts with S1; S1 authoritative…". O5's note concerns delivery, not a spec conflict. New note: consistent with delivery_type "estimated"; estimates are not commitments per 00-source-register.md. The S1-conflict label now applies only to O3/O4 (title spec claims), correctly.
3. **comparison.csv — citation completeness (minor).** spec_source/offer_source values now include the `inputs/` path prefix for exact-filename traceability.

## Checks performed this pass (local Python, work_revise.py)
- Re-read inputs/00-source-register.md, both deliverables; re-verified all six landed totals against independently hand-computed values (O1 2,238.88; O2 2,098.88; O3 1,790.88; O4 1,902.88; O5 2,126.88; O6 2,037.28) and subtotal + tax = total per row.
- Confirmed exactly one qualifying offer (O2); S3 correction applied to O6 (effective 2026-10-27, source S3-store-correction.md); O3/O4 hard failures based on S1 rows F16-16 / V15-32, not listing titles; O5 delivery unconfirmed per register.
- Verified every total quoted in recommendation.md matches the CSV; CSV parses as 6 rows × 26 columns; word count 643 ≤ 650; preserved originals unchanged (hash check).

## Unchanged (verified correct, no correction needed)
Recommendation (O2, F16-32, Harbour Tech), all arithmetic, source-conflict resolutions, fallback approach, and S4 usage (single run, no comparative claims). No live verification was claimed or attempted; no evidence added.
