Prune detail safely through normal collection #100

New Issue

2026-09-27T20:01:11Z

xavierk commented

2026-09-27 20:01:11 +00:00

Parent

Deepen observation-history publication, local-day evidence, and activity selection.

What to build

Actually expire eligible detailed observations after 14 days during normal collection without making history less trustworthy. Keep durable UTC/local summaries, shared boundary evidence, controller and projection prerequisites, successor anchors and all pending-publication recovery inputs until they can be safely replaced or consumed.

Acceptance criteria

Advance controlled time past 14 days through normal collection; eligible detailed samples are removed, with no separate manual maintenance invocation required.
A local-day summary, its recorded boundaries and shared evidence remain visible and truthful after detail expires and readers restart; historical UTC/projection evidence still works as specified.
Pending observations, unfinished intervals and required boundary/continuation anchors are never pruned merely due to age; repair/replay remains possible and idempotent.
Retention runs only after durable trustworthy publication and evidence sufficiency checks, with safe behavior for legacy dates, interruptions and repeated collection.
Use real temporary stores for normal collection-to-reader tests; focused retention tests cover missing replacement evidence and preservation of recovery prerequisites without deployed-store access.

Blocked by

#99 Make historical local-day evidence trustworthy.

## Parent [Deepen observation-history publication, local-day evidence, and activity selection](https://git.bongbetic.com/xavierk/Fenris/issues/95). ## What to build Actually expire eligible detailed observations after 14 days during normal collection without making history less trustworthy. Keep durable UTC/local summaries, shared boundary evidence, controller and projection prerequisites, successor anchors and all pending-publication recovery inputs until they can be safely replaced or consumed. ## Acceptance criteria - [ ] Advance controlled time past 14 days through normal collection; eligible detailed samples are removed, with no separate manual maintenance invocation required. - [ ] A local-day summary, its recorded boundaries and shared evidence remain visible and truthful after detail expires and readers restart; historical UTC/projection evidence still works as specified. - [ ] Pending observations, unfinished intervals and required boundary/continuation anchors are never pruned merely due to age; repair/replay remains possible and idempotent. - [ ] Retention runs only after durable trustworthy publication and evidence sufficiency checks, with safe behavior for legacy dates, interruptions and repeated collection. - [ ] Use real temporary stores for normal collection-to-reader tests; focused retention tests cover missing replacement evidence and preservation of recovery prerequisites without deployed-store access. ## Blocked by - [#99 Make historical local-day evidence trustworthy](https://git.bongbetic.com/xavierk/Fenris/issues/99).

xavierk added the ready-for-agent label 2026-09-27 20:01:11 +00:00

xavierk added a new dependency 2026-09-27 20:01:12 +00:00

#99 Make historical local-day evidence trustworthy

xavierk referenced this issue

2026-09-27 20:01:16 +00:00

Follow and pin live activity #102

xavierk added a new dependency 2026-09-27 20:01:17 +00:00

#102 Follow and pin live activity

xavierk self-assigned this 2026-09-28 07:45:02 +00:00

xavierk commented

2026-09-28 08:40:48 +00:00

Implemented in commit d697536 (fix: prune old detail after publication).

Normal collection now prunes eligible detail after successful publication. The retention pass preserves successor anchors, local-day/UTC replacement evidence, controller context, pending publication inputs, and legacy or incomplete data; an interrupted pass rolls back and can retry.

Verification: focused retention, collection, repair, and issue #99 tests passed (52 total); focused Ruff and compileall passed. The full suite reported 868 passed, 1 skipped, and 21 packaging failures caused by this environment's missing zstd, Podman build exit 100, and package install/removal/purge checks.

Implemented in commit d697536 (fix: prune old detail after publication). Normal collection now prunes eligible detail after successful publication. The retention pass preserves successor anchors, local-day/UTC replacement evidence, controller context, pending publication inputs, and legacy or incomplete data; an interrupted pass rolls back and can retry. Verification: focused retention, collection, repair, and issue #99 tests passed (52 total); focused Ruff and compileall passed. The full suite reported 868 passed, 1 skipped, and 21 packaging failures caused by this environment's missing zstd, Podman build exit 100, and package install/removal/purge checks.

xavierk commented

2026-09-28 09:07:41 +00:00

Final review follow-up in commit 5aeee7b: same-day local totals can replace an interval only when one recorded monitoring period covers that full interval. Added regression coverage for unrelated later activity after a monitoring gap.

Final verification: 60 focused retention, collection, repair, and issue #99 tests passed; focused Ruff, compileall, and git diff checks passed. Full suite: 869 passed, 1 skipped, 21 packaging failures (missing zstd, a Podman build exit 100, and package lifecycle checks).

Final review follow-up in commit 5aeee7b: same-day local totals can replace an interval only when one recorded monitoring period covers that full interval. Added regression coverage for unrelated later activity after a monitoring gap. Final verification: 60 focused retention, collection, repair, and issue #99 tests passed; focused Ruff, compileall, and git diff checks passed. Full suite: 869 passed, 1 skipped, 21 packaging failures (missing zstd, a Podman build exit 100, and package lifecycle checks).

xavierk closed this issue

2026-09-28 09:07:42 +00:00

xavierk referenced this issue from a commit

2026-09-28 09:09:37 +00:00

fix: prune old detail after publication (#100)

xavierk referenced this issue from a commit

2026-09-28 09:09:37 +00:00

fix: guard local-day pruning across monitoring gaps (#100)

Sign in to join this conversation.