# Change log — review pass
The original submission is preserved unchanged in output/original-submission/.

## action-plan.csv
- A08 no longer depends on A05 (recipient decision). No source makes the recipient decision a condition of launch approval (04-approved-change.md CR-07). A05 still blocks A10 (sending invitations). A05's due field no longer says "ideally before launch approval".
- Status wording now sticks to the evidence. The old wording implied progress that the inputs don't report:
  - A01: "not started per inputs" → "no assignment recorded as of Oct 20 09:30".
  - A06: "In progress" → "progress not reported". Mei only said the drafts *can* be finished.
  - A12: "Not started per inputs" → "no progress reported".
  - A08: "Not started" → "Pending".
  - A09: "Not started" → "Pending".
  - A14: "Not started" → "Scheduled".
- A10 and A11: the "assumption" labels are now explicit.
- Dependency lists are reordered. No IDs were added.

## decision-log.md
- Added section 6, a dated sequence of next steps (the task asks for "a realistic sequence of next steps").
- Added P4, a coordinator proposal to settle U2 and U4 early. It is marked as not agreed.
- Added a reported fact: Mei can finish drafts without choosing recipients (05-open-questions.txt).
- Section 5 now explains why CR-07's "invitations due Oct 22 12:00" is read as draft copy: D6 conflicts with sending earlier. It also explains why A05 is not a launch-approval gate.

## update-email.txt
- No change. I checked it again against D1–D10 and U1–U7, and the facts match the sources. It is 218 words (whitespace split, header lines included), within the 220 limit.

## Checks performed (output/check.py)
- CSV parses with the exact 7 required columns and 15 unique IDs, and no field is empty except dependencies.
- Every dependency is a real ID and comes earlier in the sequence.
- Every action ID appears in the decision log.
- Email word count is under the limit.
- The revised files were compared with the preserved originals.

## Not checked / limitations
- These checks are structural only. Matching facts to sources was a manual reading of the five input files.
- No messages were sent.
- check.py, fix.py and preserve.py are working scripts in output/ and can be deleted.
