Quarterly re-test record
Copy this record when a new quarter opens and leave the previous issue intact. A changed camera angle can move a line crossing even when the software and the test file keep their names, so carry the viewpoint and the measurement rule alongside the version number.
| Entry | Fill at re-test |
|---|---|
| Quarter, issue and previous issue | [enter quarter, dated run ID and prior record] |
| Scope | [site, enrolled cameras, task, field permissions and schedule] |
| Input | [manifest release and hash, field session IDs, family split and camera view] |
| System | [device, version, plan, ingest method and configuration snapshot] |
| Changes | [rule, model, firmware, network, retention, clock or viewpoint change] |
| Coverage | [monitored/scheduled seconds, outage intervals and excluded references] |
| Event counts | [matched, missed, unmatched, duplicate outputs and unscorable records] |
| Rates | [recall, precision, false alerts/hour and negative-window fraction, with raw denominators] |
| Live delivery | [endpoint, matched deliveries, missing receipts, median, p95 and clock uncertainty] |
| Task-specific result | [report cell, numerator, denominator, unit, raw file and reviewer] |
| Disposition | [repeatable, changed configuration, insufficient evidence or outside scope] |
When the numbers move
Repeat the old input on the old configuration if it is still available. Change one factor, rerun, and record the difference with its run ID. If the old configuration is gone, describe the break in the series. A difference between quarters can still go on record even when no one can isolate its cause.
For the technical preflight, replay C02 and F01 from the same release manifest and compare their checksums and time origin against the prior packet. Once the preflight is signed, move on to the field run and attach its own annotation set. A quarterly record closes when another reviewer can reproduce the reported value from the retained evidence.
Record identifiers · L01 · L02