Promote v0.40.2 to production - #1542
Merged
Merged
Conversation
Co-authored-by: Roomote <roomote@roomote.dev>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
* fix: make draft PR notifications contextual * fix: mark delegated PR context untrusted --------- Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @zarnivoop <9827931+zarnivoop@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
* feat: relay Fast child progress through parent * chore: keep Fast relay type internal * fix: reuse Fast relay key on retry * refactor: derive Fast relay identity --------- Co-authored-by: @daniel-lxs <57051444+daniel-lxs@users.noreply.github.com>
* [Fix] Canonicalize Slack day-page slugs so Brain retirement can heal gbrain lowercases slugs before a page row is written, but the Slack collector emitted and tracked slugs carrying the raw uppercase Slack team/channel ids. The inventory therefore split into census rows (lowercase, matching the corpus) and live-tracked rows (mixed-case, matching nothing), and the healing replay's reconciliation, which queries by the emitted prefix, retired zero pages. - Day-page slugs are emitted in gbrain's canonical lowercase form (identical stored pages; only local bookkeeping changes), and slug parsing canonicalizes its input for rows written before the fix. - A per-tick maintenance step (no-op once clean) rewrites mixed-case inventory rows to lowercase, merging case-duplicates by freshest lastSeenAt, and drops sync-state rows of superseded collector versions, which were polluting per-source partition aggregates. - When that rewrite actually changed rows, the deployment ran the healing replay with the mismatch live and retired nothing, so the deep backfill is re-armed by clearing its cursor: the replay re-covers the window and the now-canonical reconciliation retires the duplicates the first pass missed. Timeline evidence source keys keep their original case: they are append-only dedupe keys, and changing them would duplicate timeline rows. * fix: atomically rearm Slack healing replay --------- Co-authored-by: Matt Rubens <2600+mrubens@users.noreply.github.com>
* [Fix] Canonicalize Slack day-page slugs so Brain retirement can heal gbrain lowercases slugs before a page row is written, but the Slack collector emitted and tracked slugs carrying the raw uppercase Slack team/channel ids. The inventory therefore split into census rows (lowercase, matching the corpus) and live-tracked rows (mixed-case, matching nothing), and the healing replay's reconciliation, which queries by the emitted prefix, retired zero pages. - Day-page slugs are emitted in gbrain's canonical lowercase form (identical stored pages; only local bookkeeping changes), and slug parsing canonicalizes its input for rows written before the fix. - A per-tick maintenance step (no-op once clean) rewrites mixed-case inventory rows to lowercase, merging case-duplicates by freshest lastSeenAt, and drops sync-state rows of superseded collector versions, which were polluting per-source partition aggregates. - When that rewrite actually changed rows, the deployment ran the healing replay with the mismatch live and retired nothing, so the deep backfill is re-armed by clearing its cursor: the replay re-covers the window and the now-canonical reconciliation retires the duplicates the first pass missed. Timeline evidence source keys keep their original case: they are append-only dedupe keys, and changing them would duplicate timeline rows. * [Improve] Make Brain source states reflect real ingestion The settings page derived per-source states from raw sync-state rows and misread nearly every source on a live deployment: - 'Backfilling' meant any row with a cursor and no completion stamp, which is the permanent steady state of collectors whose cursor is a rolling checkpoint or mode-state (pull-request facts, member sweeps, Notion's incremental scan) and of the deliberately never-completing Slack and GitHub deep backfills. A replay in progress is now a parent row holding ONLY a cursor; a row with a watermark is a live stream. - Slack and GitHub backfills now record honest completion when every known channel or repository has been read, and re-arm themselves when one the finished walk never read appears. The engine preserves a done step's cursor so the resumed walk reads only the newcomer. - Stream counts were polluted: version-bump orphans, the day-page census, and per-repo rows stranded under a hardcoded superseded id all counted as streams, and 'history read' counted completion stamps that partition rows never carry. Collector ids now live once in BRAIN_COLLECTOR_IDS (types), collectors and readers share them, summaries scope to the current version's rows, GitHub's stream rows are migrated under the current id keeping their positions, and replay progress comes from the walk's own completed set. - The corpus panel no longer tells a mature Brain it 'started ingestion today' when a write burst fills the recency sample; that line is reserved for corpora the untruncated sample proves young. * fix: scope Brain backfill detection to parent rows --------- Co-authored-by: Matt Rubens <2600+mrubens@users.noreply.github.com>
* fix: add Fast inference failure diagnostics * fix: instrument Fast turn latency * feat: capture Fast turn telemetry * fix: align Fast turn diagnostics with logging conventions * fix: measure Fast inference at the prompt boundary --------- Co-authored-by: @daniel-lxs <57051444+daniel-lxs@users.noreply.github.com> Co-authored-by: daniel-lxs <ricciodaniel98@gmail.com>
Co-authored-by: Roomote <roomote@roomote.dev>
* feat: add orchestration model preference * fix: clear inherited orchestration reasoning * fix: default orchestration reasoning to low * fix: apply orchestration reasoning in Fast mode * fix: clear stale model reasoning --------- Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
* improve: browse the full Brain corpus * fix: stop stalled Brain corpus pagination --------- Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: Roomote <roomote@roomote.dev>
* fix: align Fast provider retries with tasks * fix: preserve rate-limit retry overrides --------- Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
Co-authored-by: @mrubens <2600+mrubens@users.noreply.github.com>
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Workflows to automatically generate PRs for you. |
Contributor
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Promote v0.40.2
Frozen at
c626518fff8e0ea76d7e58077980fa7d0fa79712— the commit where0.40.2was versioned. Commits merged todevelopafterward require an explicit candidate refresh or ship in the next release.v0.40.2and triggers the existing GHCRv*image publish (latestchannel).release/v0.40.2branch can be deleted after this PR merges.Changelog
0.40.2 (2026-08-22)
This patch improves Fast coordination, Slack task visibility, pull request review reliability, task feedback, and Brain preview operations.
Highlights
Patch changes