fix(loops): the verdict-capable analyst channel + the worker completion cap#259
Merged
Conversation
…on cap The two measurement channels the autopsies demanded: - StrategyCtx.consult(messages, instruction): the RAW firewalled analyst channel — the critic answers the body's instruction over the compacted trajectory verbatim (no findings extraction), so verdict-shaped steering (CONTINUE/STOP decisions, calibrated predictions) survives the wire. Same firewall as critique (trajectory in, never scores), same billing. The belief arm rewrites onto it with a confidence floor (an uncertain STOP is not obeyed); the E8 corner becomes measurable. - AgenticOptions.maxTokens → runShot: thinking workers (kimi-class) burn unbounded budgets on reasoning and return empty content without a completion cap — the matrix's kimi row measured a broken channel, not the model. WORKER_MAX_TOKENS wired through grid/evolve/steering runners. - steering-modes gains ARMS=refine,belief (targeted re-runs); the test router stub now serves both response-reading styles (json + text — a stub missing text() silently downed every consult/critique leaf in tests).
tangletools
approved these changes
Jun 11, 2026
tangletools
left a comment
tangletools
left a comment
Contributor
There was a problem hiding this comment.
✅ Auto-approved PR — 61461cfd
Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.
tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-11T12:30:22Z
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The two broken measurement channels from tonight's autopsies, fixed in one work-line:
consult()— the raw firewalled analyst channel (the E8 fix)The belief-arm autopsy proved
critique()cannot carry verdict formats —observe()'s findings extraction strips them.StrategyCtx.consult(messages, instruction)is the verdict-capable channel: the critic answers the body's instruction over the compacted trajectory verbatim. Same firewall (trajectory in, never scores), same billing as the critique path. The belief arm rewrites onto it with a calibration floor (STOPobeyed only at confidence ≥0.7); the corner becomes measurable.AgenticOptions.maxTokens— the worker completion cap (the kimi fix)runShotsent nomax_tokens, so thinking workers return empty content everywhere — the matrix's kimi row measured a broken channel, not the model.WORKER_MAX_TOKENSwired through all three runners.Plus:
ARMS=filter on steering-modes for targeted re-runs; the test router stub now serves both response-reading styles (its missingtext()was silently downing consult/critique leaves under test).+1 test (instruction reaches the analyst verbatim; raw reply intact). Suite 790 ✓ · typecheck ✓ · lint ✓. Launching on merge: kimi+glm matrix re-runs, the belief re-run, the deeper-search cost run.