Issue 11 — Source health + observability

[skcache]

Goal

Trust the system without watching a terminal (PRD §17, Issue 11).

Scope

• Structured logs
• Provider health states (healthy / degraded / backoff / down / disabled)
• Poll metrics (p50/p95/p99 latency, per provider)
• Alert failure logging
• Scheduler heartbeat
• Inactive confirmation + observed posting lifetime
• Operational Telegram alerts only for whole-shard / D1 / Telegram / scheduler failures

Non-goals

• Per-fetch metrics rows (aggregate only, PRD §8.6)
• User alerts for single-company blips

Acceptance criteria

• p50/p95/p99 processing metrics recorded
• Observed posting lifetime computable once a role disappears
• 24h soak observability (Issue 12)

Tests

• Metrics aggregation tests
• Health state transition tests

Dependencies

• Issues 5–10

EDN architecture review

No — extends existing source_state / poll_metrics via existing layers.

[skcache]

Depends on: #6 #7 #8 #9 #10 #11