The original submission delivered nothing: the one file I tried to save was cut off and never written, and no script had been run. All deliverables below were produced in this review pass, and the reconciliation script's built-in checks passed on the final run.

## October 2026 results (CAD, tax excluded)

| Measure | Workshop | Course | Combined |
|---|---:|---:|---:|
| Paid orders | 5 | 4 | 9 |
| Gross sales | 1,920.00 | 900.00 | 2,820.00 |
| Applied refunds | 240.00 | 60.00 ‡ | 300.00 ‡ |
| Net revenue | 1,680.00 | 840.00 ‡ | 2,520.00 ‡ |
| Advertising spend | 300.00 | 150.00 | 450.00 |
| Contribution after advertising | 1,380.00 | 690.00 ‡ | 2,070.00 ‡ |
| Net revenue per paid order | 336.00 | 210.00 ‡ | 280.00 ‡ |

‡ Provisional. Settled refund RF05 (row R06, Course order N1011) has no amount, so it is treated as unknown and left out, as `inputs/policy.md` para 3 requires. Because a refund cannot exceed the order's 360.00 gross, Course net revenue will land between 480.00 and 840.00, and combined contribution between 1,710.00 and 2,070.00. That is a bound, not an estimate.

Contribution after advertising is not profit: no COGS, payment fees, salaries or overhead were supplied (`inputs/policy.md` para 1).

## Reconciliation

| Source | Raw rows | Included | Excluded | Excluded row IDs |
|---|---:|---:|---:|---|
| `inputs/sales.csv` | 15 | 9 | 6 | S04, S05, S08, S09, S11, S15 |
| `inputs/refunds.csv` | 7 | 2 | 5 | R03, R04, R05, R06, R07 |
| `inputs/ads.csv` | 7 | 5 | 2 | A04, A07 |

Amounts tie out:
- **Sales:** 4,140.00 raw − 1,320.00 excluded = 2,820.00.
- **Refunds:** 960.00 raw − 660.00 excluded = 300.00, plus the one blank amount.
- **Ads:** 630.00 raw − 180.00 excluded = 450.00.

## Recommendation

- **Ad budget:** do not shift it between products on these numbers. Both return 5.60 of calculable net revenue per ad dollar, the Course figure is provisional, and the sample is nine orders.
- **Before closing October:** resolve the RF05 amount and RF04 (row R05), a 120.00 settled refund against order N9999, which is not in the sales file.
- **Workshop refunds:** three of five paid Workshop orders have a refund record. Counting the 240.00 pending (RF03) and the 240.00 settled on 2 November (RF06), refund activity is 720.00, or 37.5% of Workshop gross, against 12.5% recognised in October. The inputs give no refund reasons.

## Change log

1. Recorded the empty original state in `output/original-submission/README.md`; there were no files to copy.
2. Wrote and ran a complete `reconcile.py`, which generated the five required deliverables.
3. Read all generated files back against the raw inputs and made three corrections:
   - Filled a blank product on exception row R03 (now Course, from order N1006).
   - Narrowed a claim that RF05 and RF04 were "the only items" that could change October.
   - Replaced a prediction that Workshop revenue "overstates what is likely to be kept" with the two known amounts.
4. No figure changed between the first run and the corrected run.

## Handoff

**Checks actually performed**
- Script assertions passed on the final run:
  - totals are recomputed from the written CSVs;
  - Workshop plus Course equals Combined;
  - every raw row ID appears exactly once, in a cleaned file or in `exceptions.csv`;
  - the amount tie-outs above hold;
  - all 21 table figures match values I worked by hand.
- Raw input files were hashed before and after the final run and were unchanged.
- I read `summary.md` and all four CSVs in full after the first run. After the corrections I re-checked only the changed lines (R03 product and the two reworded summary sentences).

**Not done**
- No independent second implementation.
- Nothing was opened in a spreadsheet application.

**Remaining limitations**
- Course and combined net figures stay provisional until the RF05 amount is obtained.
- RF04 / N9999 cannot be resolved from the inputs.
- The policy file has a single heading, so citations use paragraph numbers (para 1–3).
- Cleaned files hold only records that enter the totals; R05 and R06 are kept in `exceptions.csv` with their raw values.

**Estimated human time to finish:** about 20–30 minutes to obtain the RF05 amount, trace N9999 and rerun the script. This is an estimate, not a measurement.

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

`work/patch.py`, outside `output/`, is the helper that applied the three corrections and reran the script.