Scheduling reliability: re-arm bounded recovery cycles

[hiidkaboutyou-spec]

Scope

Small follow-up repair for the confirmed one-shot Daily scheduling watchdog defect. No retrieval, source, cursor, translation, media, Telegram, Fanfic, or public-publishing behavior is changed.

Root cause

The recovery Daily is created with the repository GITHUB_TOKEN. GitHub allows that workflow_dispatch run itself, but token-triggered events do not generally create downstream workflow runs; production confirmed the recovery completion did not produce the workflow_run watchdog successor. Relying on workflow_run alone therefore made recovery one-shot.

Repair

Keep workflow_run as the fast primary watchdog arm and add an independent off-minute periodic watchdog heartbeat (4,16,28,40,52). The heartbeat only inspects Daily workflow state and dispatches one main.yml mode=live recovery when the latest relevant Daily is stale and no active Daily exists. Failed automated recovery is a hard stop for periodic chaining. API/dispatch failures fail safe.

Safety

• Daily runtime concurrency unchanged.
• Watchdog concurrency unchanged and cancel-in-progress.
• No Daily→Daily self-dispatch.
• No watchdog→watchdog recursion.
• No PAT/new secret/second host.
• Recovery still enters the normal production Daily workflow.
• Private-review/source-authority/Phase 2/Fanfic surfaces unchanged.

Validation requested

Canonical validation plus workflow YAML/CI on this exact head. Do not merge until green and reviewed.