# Review and change log

## Revised deliverables
- `output/recommendation.md` — 578 whitespace-delimited words (limit: 650).
- `output/comparison.csv` — all six offers, authoritative hardware, original and effective delivery terms, landed costs, failures/unknowns and record-level citations.

The original three submitted files are preserved separately in `output/original_submission/`: `recommendation.md`, `comparison.csv`, and `build_comparison.py`. Their SHA-256 hashes are recorded in `output/review_checks.txt`; the review script checked that the archived bytes remained unchanged during revision. Raw inputs were not edited.

## Changes
1. Clarified that discounts are already applied, rather than describing the tax calculation as having “no additional fees or discounts.” This matches `inputs/00-source-register.md`, “Scenario sources — captured October 19, 2026, 09:00 Pacific.” No amounts changed.
2. Made the fallback explicitly require **arrival in Vancouver by October 23, 2026**, not merely obtaining a commitment by that date. O5 still does not qualify on the supplied evidence: `inputs/S2-store-offers.csv`, O5, provides only an estimate.
3. Brought the existing GPU-class interpretation into the opening recommendation and renamed CSV fields to `hardware_pass_under_gpu_interpretation` and `qualifies_under_gpu_interpretation`. This avoids implying that the source explicitly supplies a discrete/integrated flag. The interpretation itself is unchanged; `inputs/S1-manufacturer-specs.csv`, all SKU records, lists named GPUs and VRAM but no class flag.
4. Specified UTF-8 explicitly in the regenerated CSV's supporting Python script.

**Decision unchanged:** O2 / F16-32 remains the sole qualifying offer under the disclosed GPU interpretation, at CAD 2,098.88 with CAD 101.12 budget headroom. [inputs/S1-manufacturer-specs.csv, F16-32; inputs/S2-store-offers.csv, O2; inputs/00-source-register.md, “Scenario sources — captured October 19, 2026, 09:00 Pacific”]

## Checks actually performed
- Re-read the assignment and all five supplied source files, comparing their authority rules and records against the submission.
- Independently recalculated each offer's shipping-inclusive tax, total and budget headroom using integer cents; all six agree with the original Decimal calculations.
- Checked every CSV hardware field against its exact S1 SKU and store/listing fields against S2. O3's RAM/SSD and O4's VRAM remain rejected despite conflicting titles.
- Verified O5 remains unconfirmed and O6 uses October 27, preserving its original October 23 field for traceability. S3 expressly supersedes O6; its stated supersession controls despite the source-register capture timestamp. [inputs/S3-store-correction.md, “S3 — Direct Aspen correction — October 19, 2026, 08:45 Pacific”, O6 / A15-32]
- Parsed the revised CSV as UTF-8; checked unique headers, consistent field counts, exactly six unique offers, all assessment flags, and source-file existence. Confirmed the Markdown landed totals match the CSV and the recommendation remains below 650 words.
- Reviewed the performance paragraph against `inputs/S4-independent-test.md`, “S4 — Edit Bench lab note — October 17, 2026”: exact tested SKU, single plugged-in run, timeline, export time and limitations are preserved. No benchmark was performed by this submission or review.
- Read back the revised recommendation. No graphical spreadsheet or Markdown rendering test was performed.

## Handoff and limitations
Supporting files: `output/build_comparison.py` (generator), `output/review_submission.py` (preservation/revision/audit), and `output/review_checks.txt` (executed check results). Run scripts from the workspace root with Python's standard library.

No seller contact, checkout, purchase, external research or new hardware evidence was obtained. GPU classification remains a disclosed interpretation; O5 cannot be treated as a confirmed fallback, and S4 cannot establish comparative performance. Estimated human work: **5–10 minutes** for a final purchase-term/GPU-class review, excluding any seller response time; this is an estimate, not a measured result. No further document correction is identified by this review.
