# Change log — review pass

The original submission is kept unchanged in `output/original_submission/` (recommendation.md, comparison.csv, build.py), copied by `preserve.py`.

## Checks performed (all run with local Python)
- **Totals:** `verify.py` recomputes all six landed totals in integer cents, independently of `build.py`. All six match: O1 2238.88, O2 2098.88, O3 1790.88, O4 1902.88, O5 2126.88, O6 2037.28.
- **Coverage and specs:** every S2 offer appears in the CSV, and the RAM, SSD and VRAM values match S1 for each SKU.
- **O1 price threshold:** 1964.28 × 1.12 = 2199.9936, which is within budget. 1964.29 × 1.12 = 2200.0048, which is over.
- **Citations and length:** every filename cited in recommendation.md exists. The file is 643 words by whitespace count, within the 650 limit.
- **Unchanged results:** the recommendation (O2) and the pass/fail result for each offer didn't change. I found no calculation error.

## Corrections and improvements
1. **CSV traceability.** I added columns citing the exact source rows for each offer (`spec_source`, `price_source`). `delivery_source` now names the S3 heading for O6.
2. **Unknowns and conflicts.** I added an `unknowns_or_conflicts` column. It records the S2 listing titles for O3 and O4 that conflict with S1, the open question of whether O5's delivery will be guaranteed, and the O6 conflict between S2 and S3, including the 08:45 vs 09:00 timestamp point.
3. **Test evidence in the CSV.** I added a `test_evidence_S4` column. It shows that only F16-32 was tested, in one run.
4. **Matching status labels.** The CSV status for O5 is now "FAILS (delivery unconfirmed)", matching recommendation.md.
5. **Missing assumption.** I added the assumption that listed shipping is mandatory and taxed. The register taxes "mandatory shipping", and S2 doesn't say whether shipping is mandatory.
6. **Clearer conflict note.** recommendation.md now says that the S2/S3 conflict over O6 decides the outcome (O6 would qualify under S2) and why S3 governs. I also added heading citations for S3 and S4.
7. **Accuracy of the earlier handoff.** The first handoff said I "re-read the generated CSV". In fact I had only printed the offer IDs, totals and status. The full CSV is now checked by `verify.py`.

## Workspace notes
- `wc.py` and `fix.py` are small helper scripts. They aren't deliverables.
- I didn't edit any input files.
