Deriving verdicts from stored observations…
Drift is judged one run at a time, so a break that lasted four days appears four times in a run ledger. Here it appears once, as the period it was.
From the run where the extraction was flagged to the run where the value stopped being believable.
Derived on this request from 60 observations across 20 runs on 6 sources — 3 episodes, 2 still open, 1 repair adopted.
A warning of zero is reported as zero. Both judges can fire on the same run, and an episode where that happened had no window to act in. Averaging only the comfortable ones would make this a sales number.
One row per period, not per run. Open a row for its warning window and the signals inside it.
| Expand | Source | Field | Period | Runs | Peak | Ended | Warning |
|---|---|---|---|---|---|---|---|
| Aurora Lamp Stand | price | 2026-08-14 → open | 1 | drifted | ongoing | value held | |
| Aurora Desk Lamp (mirror) | price | 2026-08-13 → open | 1 | failed | ongoing |
The question the verification gate could not answer.
| Source | Plan | Adopted | Fields | Runs since | Outcome | Evidence |
|---|---|---|---|---|---|---|
| Aurora Desk Lamp | v2 · deterministic | 2026-08-14 | price | 2 | held | replayed |
Verification proved a repair read the right value from the page in front of it, on the day it was accepted. It could not prove more than that. Held and regressed are what the runs since have shown, and a repair can be correctly verified and still decay — the handle it chose is only as permanent as the page’s next redesign.
Ranked by how often each field's provenance has moved. The bar is the share of comparisons in which it held still.
Only 2 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
2 structural moves in 8 comparisons. Expect the next one — the longest it has held still is 5 runs.
0 structural moves in 3 comparisons. Longest unbroken stretch: 4 runs.
0 structural moves in 3 comparisons. Longest unbroken stretch: 3 runs.
| none |
| Aurora Desk Lamp | price | 2026-08-13 → 2026-08-15 | 2 | drifted | repaired | 1 run |
0 structural moves in 3 comparisons. Longest unbroken stretch: 4 runs.
Only 2 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 2 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
0 structural moves in 8 comparisons. Longest unbroken stretch: 9 runs.
0 structural moves in 8 comparisons. Longest unbroken stretch: 9 runs.
Only 1 comparison recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 1 comparison recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 1 comparison recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 0 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 0 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 0 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 0 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 0 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
Only 0 comparisons recorded. Too little to grade — an absence of evidence, not a clean bill of health.
A repair is not counted as churn. Selectors changing on purpose is what a repair is, and counting it would mean a source got less stable every time it was fixed — the most-repaired field would grade the worst.