0703 — Collector progress uses manager lease history

Context

Worker instance logging [0700] [0702] supplies independent manager coverage. Collector progress must use the existing scheduled alert and recording paths without requiring another observation series or user-maintained pending state.

Decision

• Read the latest collector completion and manager snapshots intersecting the pending window, as of database time. Order snapshots by original creation time descending, then expiry and id descending; include boundary intervals.
• Recent completion is healthy. Otherwise, no current lease coverage is insufficient evidence. Join touching/overlapping intervals backward; uncovered gaps break pending, including overnight shutdowns.
• Unhealthy duration begins at the later of coverage start and completion plus MaximumAge, bounded by the read window. Absent completion uses coverage start. Share the pending-duration decision with sampled-history alerts.
• Register metrics_collector_progress (VK0101) as a system-owned scheduled alert, with the normal consumed policy and atomic Record/classify path. Insufficient evidence logs a warning and leaves recorded alerts unchanged.
• Add SystemConfig.MetricsCollectorProgressAlert and the system alert selector. Default cadence is one minute, pending two minutes, and MaximumAge zero resolves to max(two minutes, three declared collector poll intervals). A positive MaximumAge overrides this; DisablePending still requires coverage.
• Exact lease gaps need no MaximumGap config. Existing topic alert schedules remain hourly. Completion age uses the metric’s Unix-second value.

Consequences

No additional measurement writes or pending-state storage are needed. Current declarations can change the default age before a running collector next claims its updated rate. Lease coverage retains [0700]‘s last-expiry tradeoff and does not prove successful execution. No internal alert can report while all checking machinery is stopped. Real-worker labs remain a separate checkpoint.