# Change log — review pass

## What the review found

1. **The original submission contained no deliverables.** Its single attempt to save `output/reconcile.py`
   was cut off mid-file and nothing was written; the script was never run. No cleaned CSVs,
   `exceptions.csv` or `summary.md` existed, and no checks had been performed.
   Record kept in `output/original-submission/README.md` (there were no files to copy).

## What was done in this pass

2. Wrote a complete `output/reconcile.py` and ran it. It produced `cleaned-sales.csv`,
   `cleaned-refunds.csv`, `cleaned-ads.csv`, `exceptions.csv` and `summary.md`.
3. Read all five generated files back and compared them with the raw inputs. Corrections made after
   that read-through:
   - `exceptions.csv` row R03 (duplicate of RF02) had a blank `product`; now filled from its paid order N1006 (Course).
   - `summary.md` section 5: replaced "the only items that can change October's reported figures" with
     wording limited to the two flagged refund records (RF05, RF04).
   - `summary.md` section 5: replaced "overstates what is likely to be kept" (a prediction the inputs
     do not support, since RF03 is only pending) with a statement of the two known amounts not in October.
4. No figure changed between the first run and the corrected run.

## Checks that support the figures

- Script assertions (all passed on the final run): totals recomputed from the written CSVs;
  Workshop + Course = Combined; every raw row ID appears once in a cleaned file or `exceptions.csv`;
  amount tie-outs (sales 4,140.00 − 1,320.00 = 2,820.00; refunds 960.00 − 660.00 = 300.00 plus one
  blank; ads 630.00 − 180.00 = 450.00); all 21 results-table figures equal values worked by hand.
- Raw input files hashed before and after the final run: unchanged.
- Not done: no independent second implementation, and nothing was opened in a spreadsheet application.
