FO contract keep-moving posture — approval is a trigger, parallel async when independent, no turn-end on async launch with work remaining, captain correction narrows not halts#479
Merged
Conversation
…independent, no turn-end on async, correction narrows not halts Adds the motion-axis discipline to Working Principles as a third blockquote beside the evidence-bar (verification axis) and smallest-sufficient-mechanism (mechanism axis) principles: keep-moving governs cadence, subordinate to both on the axes they own. S1-S3 land as one reconciled blockquote; S4 (correction narrows scope, does not halt the session) folds into Clarification. Resident addition 1177 bytes, within the 1247-byte sprint remainder (sprint-wide 5530 of ~5600 ceiling vs the v0.24.0 baseline). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…andatory offline negative Reconstructs the 0223 Shaping FO false-stop patterns as a live decision scenario: a just-approved gate, two independent ready entities, and a questioned entity, all in one fixture. The host-neutral keepMovingTrace is graded over the FO's tool-call transcript (advance/dispatch/re-shape actions) plus its final message (the permission-question and async-wait turn-ending postures, read from the final message only to avoid the contract-read false-positive). Claude and Codex extractors fill the shared trace; both runner adapters, the Pi coverage entry, and the seed-scenarios doc block are wired per the four-step recipe. The mandatory offline negative proves the grader catches each false-stop pattern in isolation (post-approval permission question, serialized dispatch, dropped re-shape, driving the corrected entity, turn-end-on-async wait, halt-on-correction) while the correct keep-moving trace passes. The live drive against the origin/main baseline is validation's cost — inherits k74g's live-drive-observability proof, no fresh spike. Offline gate green; live-tagged parity/definition guards green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…g-loop surface Cycle-1 codex red diagnosed from run 28843487385: the FO used ZERO spawn_agent; it dispatched via the 0.142.5 standing loop (status --set <e> status=done + wait collab), which codexKeepMovingTrace credited only via spawn_agent -> false-negative 'advanced but did not dispatch'. Same class as zm's cycle-1. Adds the status=done per-entity dispatch surface for approvedDispatched/independentDispatched, with a dialect-regression offline case. The corrected-entity semantics (correctedDriven precision + prose) are held pending an FO decision (escalated). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…-flight check, S4 surface clause Hybrid semantics confirmed (FO, invariant triple): a captain correction means (1) the corrected entity does not drive FORWARD (implementation/done/merge) until the re-shape folds; (2) everything else keeps moving; (3) the folded re-shape SURFACES (gate review / re-presentation) or its rework is honestly named in-flight — silent absorption/park fail. Re-shape mechanism is free. Grader: correctedDriven now keys on the FORWARD status only (a routed rework dispatch or a re-open to ideation never trips it); a correctedAddressed check (final message names the corrected entity + a surface/in-flight disposition) replaces the blanket wait check. Fixed a soundness bug where a bulk multi---set Bash command cross-attributed one entity's status to another (the opus specimen sets four entities in one command) — kmAdvancesToStatus now scopes each --set to its own entity. Replay fixtures distil all three recorded drives (run 28843487385): opus RED (silent wait), sonnet GREEN (honest in-flight), codex GREEN (surfaced gate) — verified against the real transcripts. S4 prose: the pause clause now reads forward-drive (not any-dispatch) and restores the trimmed 'surface the re-shape when folded'; paid for inside vcm's envelope with trims in S1/S2. Resident addition 1235B (<=1247); sprint 5588B. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
Journey cost delta
|
Collaborator
Author
|
Merging on a fully green board — all 9 checks including codex-live (no waiver needed on this merge): the cycle-2 correction-invariant fix turned both claude variants green on vcm's own scenario, and the dialect-robust graders (status=done dispatch surface, forward-only correctedDriven, surfaced-or-in-flight) held across all three hosts. Two validation cycles + detached adversarial grader audit on record. Sprint resident ledger after this merge: 5,588 of ≈5,600 (12-byte margin). |
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.
Installs the keep-moving counterweight: approval triggers the next action, independent work dispatches in parallel, no turn ends with actionable async work remaining, and a captain correction narrows scope instead of halting it.
What changed
Evidence
go test ./...greenReview guidance
The reconciliation seams against the adjacent evidence-bar and mechanism-gate blockquotes are the risk center.
vcm