[zhaozhaozz]

The bot now dispatches collection only when an Issue or PR is opened, reopened or closed, and otherwise refreshes every eight hours. The repository's own schedule ran every 30 minutes on top of that. It now runs once a day (02:17 UTC), only as a fallback for a bot that has stopped dispatching.

Changes

• collect.yml: the cron changes from 17,47 * * * * to 17 2 * * *, with a comment explaining its role.
• The "snapshot is stale" banner moves from 2 hours to 36 hours. Collection without data changes keeps the snapshot time, a new UTC day rebuilds it, and the bot collects at least every eight hours, so a quiet repository's snapshot can legitimately be about a day and a half old. Static asset versions are bumped.
• README, docs/actions-collection.md and docs/dashboard-pages.md describe the new cadence and the banner.

Validation

• python3 -m unittest discover -s tests: 156 run. The only failures are the same 3 that fail on unmodified main (test_lines ×2, test_collect_cost).
• Browser journeys: 23 passed, 1 failed. The failure is the pre-existing E2E run records reuse published JSON and refresh changed versions, which fails identically on main. The banner assertion now expects the 36-hour wording.

Release

Release category: release:internal
User impact: fewer Actions runs. The stale banner appears only when collection has really stopped.
Compatibility and migration: none.