The review found no errors in the figures or the four data files; I corrected four statements in `summary.md` and two claims about checks. The first submission is preserved unchanged in `output/original-submission/`.

Headline results are as before: 9 paid orders, net revenue 2,520.00 CAD and contribution after advertising 2,070.00 CAD combined, with Course and combined figures still provisional because refund RF05 has no amount.

## Change log

| # | Where | Problem in first submission | Correction |
|---|---|---|---|
| 1 | `summary.md` §5, recommendation 3 | "Workshop carries the larger refund exposure" was not supported. Workshop's known refunds are 720.00 of 1,920.00 (37.5%, counting pending RF03 and November-settled RF06). Course is 60.00 plus the unknown RF05, which at its 360.00 ceiling would make 420.00 of 900.00 (46.7%). | Now states both figures and says the higher refund rate cannot be determined until RF05 is known. |
| 2 | `summary.md` §4, RF04 (refunds.csv R05) | I flagged the unmatched order N9999 but did not say it conflicts with policy.md's "every refund here relates to an October order". | Conflict stated as unresolved; the refund is still flagged and excluded as policy.md requires. |
| 3 | `summary.md` §5, "not profit" paragraph | I said policy.md attributes sales to ads. It only assigns ad spend to the product named on each ad row. | Reworded; the inputs do not show which sales the ads produced. |
| 4 | `summary.md` §5, recommendation 1 | Only RF05 was named as able to lower net revenue per ad dollar. | Added that RF04 (120.00) could also lower either product's figure. |
| 5 | `reconcile.py` comment | Said raw inputs were confirmed untouched; the code only checked row counts. | Comment corrected, and a real comparison added to `verify.py`. |
| 6 | Verification script | The first handoff cited a check whose script was not in `output/`. | Extended and delivered as `output/verify.py`. |

## Checks run for this revision

All were run through `output/verify.py`, which completed without error:

- **Headline figures:** recomputed from the cleaned CSVs alone and matched to the table in `summary.md`.
- **Policy rules on cleaned files:**
  - Order, refund and ad IDs are unique.
  - Sales are all paid and dated in October.
  - Refunds are all settled in October, matched to an included paid order and not above its gross.
  - Ads are all dated in October.
  - Dates are YYYY-MM-DD and money is two decimals.
  - RF05's amount is still blank in `exceptions.csv`.
- **Row reconciliation:** every source row is in exactly one of a cleaned file or `exceptions.csv` (sales 15 = 9 + 6, refunds 7 = 2 + 5, ads 7 = 5 + 2).
- **Inputs:** the three CSVs in `inputs/` are identical to the listing in the assignment prompt, ignoring a trailing newline.
- **Against the first submission:** the four data CSVs are byte-identical; `summary.md` differs in five lines, which are the four corrections above plus a revision note in the header.

I also read the revised `summary.md` in full as text. It was not viewed in a rendered Markdown view.

## Remaining limitations

- RF05's amount and RF04's order are unresolved, so Course and combined results stay provisional.
- Nothing was checked against a bank, payment processor or ad platform; none is in the inputs.

Estimated human time to finish is about 15–20 minutes of review, plus however long it takes to obtain the RF05 amount and trace N9999. This is an estimate, not a measurement.

Files are in `output/`:
- summary.md
- cleaned-sales.csv
- cleaned-refunds.csv
- cleaned-ads.csv
- exceptions.csv
- reconcile.py
- verify.py
- CHANGELOG.md
- original-submission/