board #757 — scheduled target offline
Date: 2026-10-08

Scope
- A scheduled occurrence whose exact target session is offline or older than
  the existing five-minute freshness cutoff is recorded as
  skipped(target_offline); no tasks/inbox task row is created.
- The human creator receives one user_inbox warning, or the Agent creator
  receives one ordinary inbox warning, per continuous offline episode.
- One observed online occurrence clears the durable episode marker and is
  dispatched through the existing path.

SQLite targeted/regression (Docker, --cpus=2)
- scheduled-offline.test.ts: 4 pass, 0 fail, 22 expect, rc=0
- fresh timestamps in SQLite UTC format, ISO Z format, and ISO Z with
  milliseconds all keep the idle target online.
- scheduler regression group: 53 pass, 0 fail, 518 expect, rc=0

Witnessed-red mutation
- Replaced the target-offline branch condition with false.
- scheduled-offline.test.ts: 0 pass, 3 fail, rc=1; failures were assertion
  mismatches (queued/delivered instead of skipped), not environment errors.
- Restored the production condition before the green runs below.
- Replaced parseDbTimestampMs with the previous hand-written append-Z parser.
- scheduled-offline.test.ts: 3 pass, 1 fail, rc=1; the ISO Z timestamp was
  misclassified as offline on an assertion mismatch, not an environment error.
- Restored parseDbTimestampMs before the final green run.

PostgreSQL ladder (Docker, --cpus=2)
- scheduled-offline.test.ts: 3 pass, 0 fail, rc=0
- feature tests: no failure
- migration verification: MIGRATE_FAILED=0
- PG_LADDER level=6 floor=6 first_fail=none, rc=0

Full server aggregate (Docker, --cpus=2)
- files=210, pass=2378, fail=0, skip=0, expects=14113, rc=0

Static checks
- check-doc-symbol-pins.py: rc=0
- check-doc-source-pins.py: broken_pins=0, rc=0

Safety
- No production Hub, production database, production node, or port 9200 was
  accessed. All databases and services were disposable Docker fixtures.
