All checks pass. Deliverables are written and verified.

## Handoff

**Output files**
- `output/recommendation.md` — 643 words (whitespace count, ≤ 650); recommends **O2, Forge Studio 16 (F16-32), Harbour Tech, landed CAD 2,098.88, guaranteed 2026-10-22**, with a conditional fallback via O5 requiring a written delivery guarantee.
- `output/comparison.csv` — all 6 offers, 26 columns: price/shipping/subtotal/12% tax/landed total, S1 specs, effective delivery date + type + source, per-requirement pass flags, verdict, conflict notes, and row-level source citations.
- Working scripts `work_build.py` / `work_verify.py` remain in the workspace (not deliverables); raw `inputs/` untouched.

**Checks actually performed (local Python)**
- Recomputed every landed total with `Decimal` (half-up cents) from the raw S2 CSV and asserted subtotal + tax = total for all rows: O1 2,238.88 · O2 2,098.88 · O3 1,790.88 · O4 1,902.88 · O5 2,126.88 · O6 2,037.28.
- Asserted exactly one qualifying offer (O2); S3 correction applied to O6 (effective date 2026-10-27); O3/O4 resolved against S1, not their listing titles; O5 flagged unconfirmed per the register's "estimates are not commitments" rule.
- Verified CSV structure (6 rows, correct IDs) and word-count limit.

**Key decisions / assumptions**
- "Confirmed delivery" = `guaranteed` type with effective date ≤ 2026-10-23; source precedence per `inputs/00-source-register.md` (S1 for specs, S3 supersedes S2 for O6 only).
- S4 used only as single-run evidence for the exact F16-32 config; no comparative performance claims invented.

**Remaining limitations**
- No live verification of stock, price or store guarantees (offline scenario); O5 fallback depends on a store confirmation not present in the packet.
- S4 leaves battery editing, ProRes and dB noise untested; other machines untested.

**Estimated human finish time:** ~10–15 minutes to review citations, sanity-check the table and place the order (estimate, not measured).