docs: reconcile the September 21 backlog and module audit

[kyleve]

Summary

Reconcile all 12 backlogs with source through a32b8b05 (PR #318) and refresh the September 21 module audit from the September 7 coverage boundary, 5f65b9f0. INBOX is empty.

• Archive the per-intent perform() item through its documented-limitation option; no new intent runtime coverage is claimed. Correct the already-completed foreground welcome reference to PR fix(Where): refresh region welcomes on foreground activation #314.
• Narrow the iOS 27/HistoryObserver, current-region marker, welcome snapshot, and intent logging requests to their remaining work. Preserve existing priorities and origins.
• File one new P2: the coalesced-location cancellation test waits for one driver request without proving both waiters registered. This is a static synchronization finding, not a reproduced CI failure.
• Correct invalidated module guidance and welcome comments, update moved citations, and distinguish architecture validation already included by ./test --everything from the CI checks it omits.
• Report 27 modules, 717 source files, 374 test/support files, 50 image suites, and 506 references, with per-module verified-source results and explicit limitations.

Review focus

HistoryObserver PR #319 remains open at the audit boundary. Welcome semantic captures now exist; scrolling, iPad, and first-greeting coverage remain open. The benchmark repository still exists after the saved September 9 downgrade date, but billing and integration state are unverified. Snapshot timing remains historical; the new counts do not constitute new performance measurements.

Testing

• ./swiftformat --lint — passed, 0/1,148 files need formatting; 125 skipped. Optional cache write was sandbox-blocked without affecting lint.
• ./shellcheck — passed.
• ./attribution --check — passed, 12 credits current.
• ./snapshot-shards check — passed, 50 suites assigned (13/15/18 + 4 intake).
• ./sync-agents and git diff --check — passed; generated mirrors remain ignored.
• Independently verified all 27 inventory rows, reference totals, and report links against tracked files.
• Skipped ./test, architecture execution, simulator/image tests, and retained Python/Ruby suites: only Markdown and Swift comments changed. No fresh visual or runtime validation. This macOS-hosted audit remains static analysis plus the listed checks; the skill's Linux/Tuist/simulator limitations are preserved. Prior PR results are historical evidence only.

Posted by an AI agent on kve's behalf.