# Change log — review pass (Task 8)

## Verdict
No scheduling error found. The chosen plan (main review Tue 2026-10-27, 14:00–14:45 PT / 21:00–21:45 UTC, Room C; Theo prep 13:30–14:00 PT; setup 13:50–14:00 PT; remote Maya+Jules follow-up 14:45–15:00 PT) satisfies every hard constraint and is the earliest feasible start. Two minor documentation/usability corrections were made; schedule.csv and invitation-draft.txt are unchanged.

## Checks performed in this review (all local, scripts included)
1. **Preservation**: byte-identical copies of all four first-submission files saved to `output/original_submission/`; input file hashes recorded (inputs never written).
2. **Stronger feasibility proof** (`review_audit.py`): re-ran the constraint search at **1-minute resolution** across 09:00–17:00. Feasible starts: 14:00–14:05 (Room C only) and 14:10–14:15 (Room B only). Earliest = 14:00, Room C — confirms the original 5-minute-grid result. Also proved all input boundaries/durations are 5-minute aligned, so the original grid could not miss an earlier start.
3. **Date/UTC**: 2026-10-27 is a Tuesday; every schedule.csv row's UTC = PT + 7h re-verified.
4. **Invitation requirements**: contains required people (Maya, Theo, Jules), optional (Nora, with reason she cannot attend), room (C), agenda, PT+UTC times, and DRAFT/not-sent marker.
5. **Risk note**: word count 433 (< 500); rejected alternatives and the single unresolved decision (beta-flag approval by an unidentified security owner, per inputs/update.txt) present; no constraint quietly relaxed — the morning 10:00 slot is rejected because no room covers setup+review (inputs/rooms.csv + inputs/update.txt).
6. **Source conflicts**: none found between constraints.md, update.txt and the CSVs; the update's room-calendar rule for setup was already applied.
7. **Action claims**: nothing sent or booked; all "verified" claims correspond to executed scripts.

## Corrections made
- `risk-note.md`: verification sentence upgraded from "5-minute-grid check" to the 1-minute-resolution result; later feasible starts now stated as exact ranges (14:01–14:05 C; 14:10–14:15 B) instead of three sample times. 413 → 433 words, still < 500.
- `check_schedule.py` (output copy): added RUN LOCATION header — it must be run from the workspace root because it reads `inputs/` relatively.
- Added `review_audit.py` (this review's evidence script) and this change log.
- `review_audit.py`: preservation step now reports identical-vs-revised per file instead of asserting equality (the assertion was only valid before the intentional revisions).

## Unchanged assumptions/limitations (still flagged in risk note)
- Theo's prep "no other meeting" checked only against availability windows (no separate calendar supplied).
- "Same available window" = Maya's/Jules's windows containing the review.
- Zero setup buffer at Room C's 13:50 opening; 5-min end buffer; fallback Room B 14:10 PT.
- Security owner remains unidentified — the one decision still needed.
