[MashellHan]

Problem

The scheduled Activities capture scanned broad history during ordinary updates and waited for legacy Token collection before publishing PR and commit observations. Incomplete focused PR refreshes could also lose older same-scope observations.

Change

• Bound ordinary GitHub and ADO PR discovery to the unpaid Beijing-day window: yesterday and today before 00:30, today after 00:30 when yesterday is settled. Persist missed-day debt until both providers complete a validated publication; expire debt outside the retained 60-day window.
• Keep pinned-head continuation for side merges and rewritten history, and batch GitHub delta details. Retain older scoped ADO PRs after a focused page restart and older qualified GitHub private PRs when typed Profile visibility changes.
• Admit only paired, canonical UTC commit-date bounds of at most 60 days through the scoped ADO HTTPS adapter. A failed cold GitHub private PR inventory cannot establish the warm-focus baseline; a completed private scope replaces its prior cache even when the authoritative result is empty, while incomplete scopes retain prior observations.
• Publish the remote snapshot before waiting for legacy Token work while keeping the existing single writer lock and child cleanup. No calendar-triggered full-window worker runs on a settled day.

Validation

• Corrected affected Python synthetic suites: refresh 32/32, ADO 42/42, GitHub Profile 38/38. The unchanged contribution-writer suite passed 15/15 on the initial frozen candidate.
• Corrected check:edit: 2/2 passed; git diff --check and the normal unskipped pre-push hook passed. The hook checked the 11-file PR diff in 0.1 seconds.
• The initial frozen head had 119/119 Python tests and one stable check:ready across 37 files, 1376/1376 in 144.9 seconds (above its 120-second advisory target). These are historical evidence for unchanged inputs; no second broad ready was run for the focused correction. The initial head's CI result is separate and does not accept this changed head.
• The independent initial-head synthetic retention and eight midnight/debt transitions passed. The corrected head adds expired-debt, real protected-adapter synthetic HTTP, cold inventory restart, authoritative zero and partial-scope retention cases.

Pending evidence

The original reviewer requested four P2 corrections on the initial head; corrected-hunk delta review and current-head CI remain with Lead. The 30-second end-to-end publication target has not been measured on the reviewed installed helper. The original delivery/operator lane owns signed helper installation and later natural capture readback. No live collector or App mutation was used for this PR.

QA-Handoff: 94cd89c

[MashellHan]

Lead has adopted this exact head 386faec, tree 33d1c8c270962feedcc6cd27b31fdec9ccaaa765, against main 7dd2918. The same original independent source Reviewer has actually started the frozen eleven-path review.

Current-head CI 43359560552 passed for this head. Source acceptance is pending. A known offline-debt edge is being corrected by the same Backend owner: debt older than the retained 60-day window must recover without repeatedly forcing cold queries, while partial recovery keeps genuine debt. The reviewer will return one finding batch; the resulting corrected head needs only affected checks, delta review and its own required CI.

The stable 119 Python checks, edit 2/2, ready 1376/1376, normal hook and independent retention/midnight probes retain their exact old-head provenance. No unchanged broad rerun is requested. Lead owns current-head CI and protected merge. The separate frontend local-visible-refresh task has its own source owner; UI20 delivery is complete and excluded. Actual 30-second fresh publication/display, helper installation and natural recurring evidence remain pending.

[MashellHan]

Original source review: CHANGES REQUIRED at 386faec / tree 33d1c8c270962feedcc6cd27b31fdec9ccaaa765.

One correction batch is assigned to the same Backend owner:

• ACTIVITIES748-01 (P2): Bound expired finalization debt to the retained window; partial or unpublished recovery must keep payable debt.
• ACTIVITIES748-02 (P2): Admit only the validated optional ADO date-bound pair through the actual protected adapter. Cover real adapter validation with synthetic HTTP, including malformed, duplicate, one-sided and out-of-scope rejection.
• ACTIVITIES748-03 (P2): Do not enable focused private-PR discovery from an incomplete cold baseline. Complete that inventory before narrowing after restart.
• ACTIVITIES748-04 (P2): Persist authoritative empty private-PR results so removed observations do not reappear after restart or the next focused day.

The original reviewer found no additional issue in publication-before-legacy or writer-lock/cleanup ordering. The correction will receive affected checks, the normal hook, a frozen head and one review of the corrected hunks/contracts by the same reviewer. The old-head CI success and saved local checks retain their original provenance; they do not establish acceptance of the corrected head. No second broad ready or review, live collector run or App action is requested.

Canonical original verdict SHA256: 59c727568474068359b99bdadd9bb9e4909f2dad42ee36236f0b41597ebb33cb. Lead retains required current-head CI and protected merge. The actual 30-second fresh publication/display target remains unmeasured.

[MashellHan]

The original independent Reviewer accepted correction head 94cd89c, tree ef3d1be056fab21d0697d4910d556bf839547138, and closed all four ACTIVITIES748 findings. Corrected-hunk review receipt SHA256: 93b6f0758f63b1939b08763ec99bb32ab2161680609ffa0da69b16ed66f6cd89.

Required exact-head CI 43361700882 passed. Fresh branch protection and applicable QA controls were checked before the normal merge. Actual main is f2b361c with the accepted tree, parents 7dd2918 and 94cd89c.

Corrected refresh 32/32, ADO 42/42, Profile 38/38 and edit 2/2 remain scoped evidence. The original broad ready and initial failures retain their original input provenance. PR750 supplies the separately accepted local visible refresh and is being composed onto this main.

Source integration is complete. Signed helper/App delivery, isolated native polling acceptance and later natural source readback remain pending with the original Front installer and original data operator. The 30-second fresh publication/display target remains unmeasured; timeout or retained stale observations will not count as a successful fresh cycle.