# Change log — Task 1 review (revision 2)

The original submission is kept unchanged in `output/original_submission_v1/` (copied by `output/preserve_v1.py` before any edits).

## Checks performed during the review
- Re-read inputs/policy.md and all three raw CSVs. Compared every cleaned row with its source row:
  - dates (DD/MM/YYYY slash dates, `Oct 05 2026`)
  - money parsing (`CAD 240.00`, `$120.00`, `$50.00`)
  - product and channel names (trimmed; letter case ignored)
- Read `cleaned-sales.csv` in full. I had not opened it before submitting the original. No errors found.
- Re-ran `build_task1.py`. All 21 metric assertions passed again, with unchanged values: combined 9 orders; gross 2,820.00; refunds 300.00; net revenue 2,520.00; ads 450.00; contribution 2,070.00; 280.00 per order.
- New scripted checks, all passing:
  - Included + excluded = raw for every file. Rows: 15 = 9 + 6; 7 = 2 + 5; 7 = 5 + 2. Amounts: 4,140.00 = 2,820.00 + 1,320.00; 960.00 = 300.00 + 660.00 (plus 1 unknown); 630.00 = 450.00 + 180.00.
  - exceptions.csv lists exactly the excluded rows.
  - RF05 range, with the refund capped at N1011's 360.00 gross: combined net 2,160.00–2,520.00; contribution 1,710.00–2,070.00; per order 240.00 combined and 120.00 Course at the low end.
- Checked the policy rules against the data:
  - The retained N1008 row is the latest updated_at version (S10).
  - Pending and cancelled orders are excluded, and so is the pending refund.
  - The refund tied to an invalid order (RF04/N9999) is flagged.
  - No refund exceeds its order's gross.
  - No unmapped products.

## Corrections
1. **Unsupported claim in the original handoff.** I had described the reconciliation as "hand-checked", but no tool had checked it. It is now checked by assertions in `build_task1.py`, and summary.md says so.
2. **Misleading column in exceptions.csv.** `amount_impact_cad` was renamed to `recorded_amount_cad`. A new `effect_on_totals` column separates rows that are simply excluded from R06, whose unknown amount makes the totals provisional.
3. **Missing product on refund rows.** cleaned-refunds.csv now shows the linked order's product for R03 (Course), for information only. R05 stays blank because N9999 has no linked order.
4. **summary.md.** Clarified how the checks were done, explained the exceptions.csv columns, and added a revision note.

No figures changed. No source conflict was found that would change a classification.
