Add bounded evidence to work checkpoint candidates #31

Merged
weichen merged 10 commits from feature/fast-slow-u2-u3 into main 2026-10-19 09:20:24 +00:00
Owner

Summary

This branch defines the source-bound task snapshot used by progressive work cards. Its r0.3 revision lets checkpoint candidates carry optional, bounded source-indexed premises for a direct relationship, a difference, or an unresolved relationship; that revision does not change the persisted task-card shape.

 WorkFindingCandidate
   sources[]
+  evidence?: { mode, premises[], result? }

The shared Dev task card records the approved r0.3 design and exact component SHA b474394.

Evidence

  • Before: The r0.2 A01 model placed all three interface-dependent requirements in the remaining four without a source relationship; the independent semantic verdict was FAIL.
  • After: Candidate validation rejects unknown modes, duplicate/missing roles, out-of-range source indexes/counts, invalid result shapes and arguments over 16 KiB. Core go test ./... -count=1 passed; the generated Frontend/UI mirrors were validated in Dev CI run 3014.
  • Gate: The new r0.3 real-model quality sample is NOT_RUN. All six PRs remain drafts until it passes and the required human stages complete.

Merge Danger

Door: Two-way for the optional r0.3 candidate field. The broader branch also introduces versioned task snapshots, so stored cards must be checked before reverting the whole PR.

Blast Radius: Task snapshot contracts, checkpoint wire validation and generated type mirrors.

## Summary This branch defines the source-bound task snapshot used by progressive work cards. Its r0.3 revision lets checkpoint candidates carry optional, bounded source-indexed premises for a direct relationship, a difference, or an unresolved relationship; that revision does not change the persisted task-card shape. ```diff WorkFindingCandidate sources[] + evidence?: { mode, premises[], result? } ``` The shared [Dev task card](https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/src/branch/feature/fast-slow-u2-u3-plan/.task/feature-fast-slow-u2-u3-plan.md) records the approved r0.3 design and exact component SHA `b474394`. ## Evidence - **Before:** The r0.2 A01 model placed all three interface-dependent requirements in the remaining four without a source relationship; the independent semantic verdict was FAIL. - **After:** Candidate validation rejects unknown modes, duplicate/missing roles, out-of-range source indexes/counts, invalid result shapes and arguments over 16 KiB. Core `go test ./... -count=1` passed; the generated Frontend/UI mirrors were validated in [Dev CI run 3014](https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/actions/runs/3014). - **Gate:** The new r0.3 real-model quality sample is NOT_RUN. All six PRs remain drafts until it passes and the required human stages complete. ## Merge Danger **Door:** Two-way for the optional r0.3 candidate field. The broader branch also introduces versioned task snapshots, so stored cards must be checked before reverting the whole PR. **Blast Radius:** Task snapshot contracts, checkpoint wire validation and generated type mirrors.
weichen added 5 commits 2026-10-13 06:40:19 +00:00
Why: preserve scoped work findings and truthful activity across reloads.

Evidence: task JSON loss and validation/state regressions RED to GREEN; complete Core suite.

Impact: optional card/document wire fields and bounded candidate validation; old Body remains supported.

Refs: fast-slow U2/U3 Task1, R03 R05 R09 R17
Why: Completed tasks require current source authorization and content validation.

Evidence: Optional fields preserve existing contracts; composition verification follows.

Impact: Adds host-owned input source references with no new model authority.

Refs: fast-slow-u2-u3 Tasks 2 and 4
Why: A completed run can still lack readable or fully provided source material.

Evidence: Composition validation follows the backend and UI consumers.

Impact: Adds optional host-owned limitations without changing candidate authority.

Refs: fast-slow-u2-u3 Tasks 4 through 6
Why: Older clients must see the same coverage limits as structured task cards.

Evidence: Core suite and composition checks follow this projection update.

Impact: Projects optional limitations into existing list nodes.

Refs: fast-slow-u2-u3 Task 6
Why: Preserve upstream settings authority and current component admission.

Evidence: Exact authoritative main object fetched before merge; composition replay follows.

Impact: Retain Task contracts alongside scoped settings.

Refs: Agent-System-Dev U2/U3 delivery
weichen added 1 commit 2026-10-13 11:00:49 +00:00
weichen changed title from [WIP] Preserve optional task, finding and source contracts to [WIP] Preserve source-bound task snapshot contracts 2026-10-13 11:03:11 +00:00
weichen added 2 commits 2026-10-13 18:08:44 +00:00
Why: U3 needs bounded, source-indexed premises before the host can judge numeric and membership claims.

Evidence: TestWorkCheckpointEvidenceModesAndBounds failed on missing types, then go test ./... -count=1 and git diff --check passed.

Impact: Extend only the candidate wire contract; persisted task snapshots and display kinds stay unchanged.

Refs: Dev task feature/fast-slow-u2-u3-plan; r0.3 evidence implementation plan Task 1
Why: Generated mirrors gained unused named exports and tripped the frontend knip ratchet.

Evidence: Exact Dev combination failed at 53 unused exports over 52 baseline; Core Go tests passed after narrowing the TS export surface.

Impact: Preserve WorkFindingCandidate.evidence shape while avoiding unused exported helper names.

Refs: Dev task feature/fast-slow-u2-u3-plan; failed combo receipt 275bebe2
weichen changed title from [WIP] Preserve source-bound task snapshot contracts to [WIP] Add bounded evidence to work checkpoint candidates 2026-10-13 18:19:27 +00:00
weichen added 2 commits 2026-10-19 02:59:18 +00:00
weichen marked the pull request as ready for review 2026-10-19 09:20:21 +00:00
weichen merged commit 6fe40b0342 into main 2026-10-19 09:20:24 +00:00
Sign in to join this conversation.
No Reviewers
No labels
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: weichen/Agent-System-Core#31