Scheduled Actions stability

[sarthakagrawal927]

Work

1. Scheduled Actions stability — keep seed/enrich/embed and digest workflows green across Turso dimension changes and API rate limits.

Context

Migrated from PROJECT_STATUS.md on 2026-07-29 when Fleet made GitHub Issues the sole operational work queue.

This issue is the to-do. Link the implementing pull request with Closes #<issue>; the linked pull request makes the work in progress, and merge + issue closure makes it done.

[sarthakagrawal927]

The bounded workflow hardening is already in main: migration self-heals embedding-dimension drift, database calls retry transient failures, seed/enrich/embed jobs have concurrency and hard limits, and the last scheduled seed plus weekly digest runs completed successfully.

The remaining part—keeping automatic database-backed runs enabled—is externally blocked by the Turso row-read allowance. Automatic seed/digest schedules are intentionally paused behind the documented circuit breaker until there is an explicit provider-side row-read budget and alert. Marking this blocked rather than running or changing production infrastructure.

[sarthakagrawal927]

Status check (2026-08-08)

Product/app issues from this week are fixed and deployed (auth, rate-limit false positives, My Projects fleet snapshot, loading shells, CI lint).

This issue (scheduled Actions)

Still blocked by ops policy, not by missing code:

• Turso is retired / rollback-held; D1 is authoritative.
• Automatic seed/digest schedules remain intentionally paused (row-read circuit breaker from the Turso era). Re-enabling needs an explicit operator decision for D1 budget + schedule config — not a code fix.

Bounded hardening already on main: migration self-heal, retries, concurrency caps on seed/enrich/embed.

No production schedule change made in this pass.

[sarthakagrawal927]

Unblocked and implemented in #92.

The Turso row-read circuit breaker that held this issue no longer applies — Turso is retired, D1 is authoritative, and the six most recent scheduled seed runs (2026-08-09 → 2026-08-14) completed successfully. Removing the blocked label.

The remaining in-repo gap was failure visibility: a failed scheduled run produced no signal outside Actions history. #92 adds an alert job that keeps one open scheduled-job-failure tracking issue — opened/updated on a failed scheduled run, closed automatically by the next successful one.