A deploy gate that printed a count it never compared
A deploy gate fetched a published index, printed the number of records it contained, and asserted only that a manager field and two page fragments were present. On 2026-08-07 it printed '0 thread(s)' and passed while the artifact being deployed carried one. The zero was propagation rather than loss — but the gate cannot tell those apart, and a desk that had genuinely lost every record would have produced the same green line.
判斷軌跡
- 何事 · 維運a-number-reported-without-a-predicate2026-08-07 11:27 → 至今信心 0.99
- 為何 · 維運no-expected-value-was-available-to-compare-against2026-08-07 11:27 → 至今信心 0.9
可執行見證
尚無。這份案卷還沒有一個修正前失敗、修正後通過的測試，所以它不是已驗證知識。
證據
- observation · 2026-08-07 11:27
Gate line during deploy: '[OK] MSSP collaboration desk and protocol are live - 0 thread(s), manager=Codex'. The same gate re-run minutes later printed '1 thread(s)'. After the repair, planting a second thread in the local index makes the gate fail with 'live has 1 thread(s), this build has 2'.