# Review change log

## Outcome

No calculation, eligibility or source-conflict correction was required. The review independently derived inclusion decisions from the raw CSVs and policy, rather than relying on the initial script's manually specified exclusion list. All 21 reported metric values (seven metrics × Workshop, Course and combined) matched.

The governing source remains `inputs/policy.md`, **Northline Workshops accounting note — October 2026**. No new evidence was introduced.

## Changes made

- **Preserved the original submission**: the five original deliverables are stored byte-for-byte in `output/original-submission/`. Original-file hashes are recorded in `output/review-checks.json`.
- **Clarified `output/summary.md`**: exceptions should be joined by source filename plus `row_id`; the repeated `exclusion_code` is a category, not a unique join key.
- **Corrected ambiguous reconciliation labels**: columns containing “+ unknown” no longer call the entire amount “known.” Added the explicit known-only refund bridge, 960.00 = 300.00 + 660.00, while keeping the missing R06 amount unresolved (`inputs/refunds.csv`, R06).
- **Clarified the applied-refunds label** as known, eligible deductions and added a review-update section.
- **Added `output/review-checks.json`** with executed check results, independently calculated totals, amount bridges, and preservation hashes.
- **No CSV content changed**: `cleaned-sales.csv`, `cleaned-refunds.csv`, `cleaned-ads.csv` and `exceptions.csv` remain byte-identical to the original submission.

## Checks actually performed

- Parsed all four output CSVs with Python's standard-library CSV reader; checked unique headers, rectangular rows, source-field transformations and source-row coverage.
- Independently applied unique-ID, latest-update, status, date, valid-paid-order and missing-amount rules to the raw files. Confirmed S10 supersedes S09 (`inputs/sales.csv`); duplicates S04, R03 and A04 remain excluded in their respective source files.
- Confirmed all 29 source rows are retained, with 16 included and 13 excluded; verified one exception for each exclusion and the existence of referenced row IDs.
- Recomputed product and combined metrics using decimal arithmetic and compared them to all 21 displayed metric cells.
- Rechecked known monetary bridges and cumulative known settled refund caps, including November R07. Confirmed R06's amount and R05's product remain blank (`inputs/refunds.csv`).
- Compared original and current CSV bytes and verified raw input hashes were unchanged during this review.
- Reviewed the initial script and its recorded successful execution against the initial summary's test claims. Those claims were supported; the new independent rule-based checks add stronger validation of eligibility decisions.

CSV usability was tested by parsing, not by opening a spreadsheet application. No external verification or additional source access was performed.

## Revised handoff

Current deliverables:

- `output/cleaned-sales.csv`
- `output/cleaned-refunds.csv`
- `output/cleaned-ads.csv`
- `output/exceptions.csv`
- `output/summary.md`
- `output/change-log.md`
- `output/review-checks.json`

Original deliverables: the same five original filenames under `output/original-submission/`.

Remaining limitations: RF05's amount is missing and RF04 references unmatched order N9999 (`inputs/refunds.csv`, R06 and R05). Calculable net revenue remains CAD 2,520.00 and contribution after advertising CAD 2,070.00, both provisional. Full business profit cannot be determined without the cost categories absent under the accounting policy.

Estimated human effort to finalize remains **15–25 minutes after supporting records are available**, excluding waiting time and compiling a full cost ledger. This is an estimate, not a measured result.
