Repository navigation
fix(compaction): drop thinking blocks from the live tail - #1316
Closed
TheGreatAxios wants to merge 1 commit into
Conversation
Uncapped thinking on the newest pair was pinning post-fold occupancy near 45%. The live copy drops thinking and redacted_thinking; the archive keeps the full bodies. Thinking-only turns become the existing omitted marker so adapters do not 400 on empty content.
TheGreatAxios
commented
Oct 4, 2026
TheGreatAxios
left a comment
TheGreatAxios
left a comment
Collaborator
Author
There was a problem hiding this comment.
Review · Comment
Live tail drops thinking/redacted_thinking so a 40k-token thinking fixture folds from 40,144 live tokens (main / #1315) to 2,375. Newest tool pair stays; thinking-only turns become THINKING_ONLY_OMITTED. Warden: approve. Planner: hold-with-follow-up.
Findings
src/session/compactor.ts:731—excerptTailTurnalso strips thinking from the newest assistant that still hastool_call. Anthropic requires that thinking besidetool_useon the post-compact continuation. This is a known follow-up (do not reverse the occupancy drop; do not excerpt thinking). Follow-up: https://linear.app/abklabs/issue/CL-9904
bun test src/context-compactor.test.ts src/session/compactor-pairing.test.ts — 45 pass. Compaction baseline e2e/metrics pass. Occupancy eval vs origin/main: 40144 with thinking vs 2375 without.
TheGreatAxios
deleted the branch
cl-9902-keep-compaction-growth-latch-after-under-threshold-folds
October 5, 2026 01:44
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
thinkingandredacted_thinkingblocks so uncapped newest-pair thinking cannot pin post-fold occupancy near 45%. Full thinking stays in the archived transcript.[thinking-only turn omitted]marker so adapters do not 400 on empty content. Newest tool-call/result pairs still stay for resume.live_tokens(post-fold live-turn estimate). Tail budget stays 2500.On a 40k-token thinking fixture: origin/main and the latch-only branch keep 40,144 live tokens with thinking; this branch folds to 2,375 live tokens (under 10% of 128k).
Verification
bun test src/context-compactor.test.ts src/session/compactor-pairing.test.ts— 45 passbun test ./evals/compaction/metrics.test.ts ./e2e/compaction-baseline.test.tspassFixes CL-9899