Reliability, performance, and safer regeneration

A two-week sprint focused on making the platform trustworthy and fast as it scaled. The throughline was reliability: GitHub sync now recovers from transient failures, and the most-visited dashboard screen stopped re-fetching on every visit.

• Reliability — added retry handling + a dead-letter path to background sync so activity stops silently going missing (acme/platform-api #482).
• Performance — cached recent activity so the dashboard loads instantly on return instead of re-fetching (acme/web-app #311).
• Maintainability — reworked summary generation so entries regenerate cleanly without duplicating state (acme/platform-api #479).

Made background jobs idempotent by keying on the activity id — the decision that made retries safe rather than the retry loop itself. Chose to surface exhausted jobs in a dead-letter record instead of swallowing the failure.

The work timeline stays complete under GitHub rate-limiting, and the dashboard's most-visited screen is no longer a cold fetch every visit. Reliability work like this is invisible when it works and very damaging when it doesn't.

Reviewed the authentication cleanup PRs and helped settle how sessions are refreshed and invalidated.

• Add a metric/alert on dead-letter volume so silent failures surface quickly.
• Extend the caching pattern to the activity detail view.

• acme/platform-api #482 — Improve GitHub sync reliability with retry handling
• acme/web-app #311 — Reduce dashboard loading friction by caching recent activity
• acme/platform-api #479 — Refactor summary generation flow