# Change log — review round (Task 1, Northline Workshops October 2026)

Original submission preserved unchanged in `output/original_submission/`
(cleaned-sales.csv, cleaned-refunds.csv, cleaned-ads.csv, exceptions.csv,
summary.md, reconcile.py). Raw inputs in `inputs/` were never modified.

## Review checks performed and outcomes

1. **Requirement coverage vs PROMPT.md**: all seven metrics reported for
   Workshop, Course and combined; cleaned files preserve source row IDs;
   exceptions.csv explains every exclusion; reconciliation and a practical
   recommendation distinguishing contribution from profit are present. No
   missing requirement found.
2. **Calculations**: re-ran `reconcile.py` and `verify.py`. Totals recomputed
   independently from the written cleaned CSVs and matched hand-derived
   expectations: Workshop 5 orders / 1920.00 gross / 240.00 refunds / 1680.00
   net / 300.00 ads / 1380.00 contribution / 336.00 per order; Course 4 /
   900.00 / 60.00 / 840.00 / 150.00 / 690.00 / 210.00; combined 9 / 2820.00 /
   300.00 / 2520.00 / 450.00 / 2070.00 / 280.00. No arithmetic error found.
3. **Source conflicts**: S09 vs S10 resolved by latest updated_at (S10,
   180.00) per inputs/policy.md; exact duplicates S04/R03/A04 collapsed once
   each. Row conservation re-asserted: sales 15 = 9+1+1+4; refunds 7 =
   2+1+0+4; ads 7 = 5+1+0+1.
4. **Formats/usability**: all cleaned dates YYYY-MM-DD, money two decimals,
   R06 amount preserved empty; CSVs re-read and parsed successfully.
5. **Claims audit**: every stated test (assertions, independent recomputation,
   refund-ceiling checks) corresponds to code actually executed locally;
   no fabricated results. Provisional status correctly driven by R06 only.

## Corrections made (no numeric changes)

1. **Traceability of collapsed/superseded rows** — kept rows in the cleaned
   files now carry back-reference notes so all source row IDs are traceable
   from the cleaned files themselves, not only exceptions.csv:
   - cleaned-sales.csv S03: `collapsed_exact_duplicate:S04`
   - cleaned-sales.csv S10: `supersedes:S09_latest_updated_at`
   - cleaned-refunds.csv R02: `collapsed_exact_duplicate:R03`
   - cleaned-ads.csv A03: `collapsed_exact_duplicate:A04`
2. **Scoped the provisional upper-bound claim in summary.md** — RF05 (R06)
   affects only Course and combined figures; Workshop figures are unaffected.
   Added combined bounds under the policy's refund ≤ paid-gross ceiling
   (N1011 gross 360.00): combined net revenue 2160.00–2520.00, combined
   contribution 1710.00–2070.00 (Course net 480.00–840.00 as before).
3. **Wording fix in summary.md assumptions** — replaced "it contributes 0 to
   the calculable totals" (could be misread as zeroing a missing amount,
   which policy forbids) with "it is excluded from the calculable totals and
   its unknown amount is not treated as zero".

## Post-correction verification (all passed)

- `verify.py` extended with review checks and re-run: back-reference notes
  present; included flags, all non-note fields, and exceptions.csv verified
  byte-equivalent in content to the preserved original submission; all
  original format, total, policy-decision and summary-figure checks still
  pass ("REVIEW CHECKS PASSED", "ALL CHECKS PASSED").
- `output/reconcile.py` refreshed to the corrected generator; re-running it
  reproduces the revised deliverables deterministically.

## Remaining limitations (unchanged)

- Results provisional pending RF05's missing amount (refunds.csv R06).
- RF04 (R05) references nonexistent order N9999; unresolvable from inputs.
- RF05's refund ceiling cannot be checked (amount unknown).
- No COGS/fees/salaries/overhead supplied; contribution is not profit.

Estimated human finish time (estimate, not measured): ~20–30 minutes to
obtain the RF05 amount, investigate N9999, and re-run reconcile.py.
