[P2][workflow] Revalidate dependency evidence before loop terminal outcomes #39

New issue

Closed

opened 2026-10-11 16:36:35 +00:00 by kilo · 1 comment

kilo commented

2026-10-11 16:36:35 +00:00

Member

Goal

Extend MI /loop with evidence-based dependency-chain tracing and explicit nonterminal recovery narration without changing MI's repository-local workflow boundaries.

Why

MI already refreshes its own issue queue and returns a proven untouched prerequisite/producer/contract gate to fresh SELECT (#29). Current develop still evaluates explicit dependencies directly; it does not trace backward to the earliest executable predecessor, define UNVERIFIED dependency evidence, or state that technical-recovery progress narration must not hand control back to the owner. Scarcity Router issue #202 and merged PR #203 establish the current precedent.

Scope

Update .kilo/command/loop.md, .kilo/rules/10-task-system.md, applicable rule 50 recovery guidance, and semantic guards in tests/test_workflow_contract.py. Keep dependency evaluation limited to canonical MI issue/producer evidence relevant to the current /loop queue.

Acceptance Criteria

Before an existing WAIT or terminal decision (BLOCKED, STOP_AND_ASK, QUEUE_EMPTY), refresh canonical dependency and producer-contract evidence; traverse explicit edges backward to the earliest executable predecessor. A gated candidate does not block another eligible MI issue.
Missing/inaccessible or unresolved evidence is UNVERIFIED, not readiness and not a proven gate/blocker. Do not claim an empty queue from unknown evidence; name the precise evidence gap and follow existing bounded recovery/terminal rules.
Progress narration during technical recovery is nonterminal; the same authorized invocation continues until an existing terminal condition is met.
Preserve #29 pre-implementation re-selection/history behavior, explicit /loop full-queue authority, standalone selected-issue/PR scope, the review ceiling, reviewer independence, secret safety, and merge authority.
Offline semantic tests cover backward chains, unrelated eligible work, fresh evidence changes, unverified states, terminal queue scope, and nonterminal recovery narration.

Constraints

Workflow-only. No MI product selection/calibration/data changes, live inference, new dependencies/services, cross-repository mutation, or Router-specific model/security requirements. Existing acceptance, reviewer, validation, and authority gates remain unchanged.

Links

Scarcity Router precedent: BioMedical-IT/scarcity-router#202
Integrated correction: BioMedical-IT/scarcity-router#203
Existing MI re-selection contract: #29

## Goal Extend MI `/loop` with evidence-based dependency-chain tracing and explicit nonterminal recovery narration without changing MI's repository-local workflow boundaries. ## Why MI already refreshes its own issue queue and returns a proven untouched prerequisite/producer/contract gate to fresh `SELECT` (`#29`). Current `develop` still evaluates explicit dependencies directly; it does not trace backward to the earliest executable predecessor, define `UNVERIFIED` dependency evidence, or state that technical-recovery progress narration must not hand control back to the owner. Scarcity Router issue #202 and merged PR #203 establish the current precedent. ## Scope Update `.kilo/command/loop.md`, `.kilo/rules/10-task-system.md`, applicable rule 50 recovery guidance, and semantic guards in `tests/test_workflow_contract.py`. Keep dependency evaluation limited to canonical MI issue/producer evidence relevant to the current `/loop` queue. ## Acceptance Criteria 1. Before an existing `WAIT` or terminal decision (`BLOCKED`, `STOP_AND_ASK`, `QUEUE_EMPTY`), refresh canonical dependency and producer-contract evidence; traverse explicit edges backward to the earliest executable predecessor. A gated candidate does not block another eligible MI issue. 2. Missing/inaccessible or unresolved evidence is `UNVERIFIED`, not readiness and not a proven gate/blocker. Do not claim an empty queue from unknown evidence; name the precise evidence gap and follow existing bounded recovery/terminal rules. 3. Progress narration during technical recovery is nonterminal; the same authorized invocation continues until an existing terminal condition is met. 4. Preserve `#29` pre-implementation re-selection/history behavior, explicit `/loop` full-queue authority, standalone selected-issue/PR scope, the review ceiling, reviewer independence, secret safety, and merge authority. 5. Offline semantic tests cover backward chains, unrelated eligible work, fresh evidence changes, unverified states, terminal queue scope, and nonterminal recovery narration. ## Constraints Workflow-only. No MI product selection/calibration/data changes, live inference, new dependencies/services, cross-repository mutation, or Router-specific model/security requirements. Existing acceptance, reviewer, validation, and authority gates remain unchanged. ## Links - Scarcity Router precedent: https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/issues/202 - Integrated correction: https://forgejo.creatidy.com/BioMedical-IT/scarcity-router/pulls/203 - Existing MI re-selection contract: https://forgejo.creatidy.com/Creatidy/model-intelligence/issues/29

kilo referenced this issue

2026-10-11 16:50:01 +00:00

Trace loop dependency evidence before terminal outcomes #40

kilo commented

2026-10-11 16:50:12 +00:00

Author

Member

Implemented and pushed the workflow-only change in PR #40: #40. Base/merge base: e77f7ae13132c9c202c4cf878bb0e42f2f29680d; head: 4d189895a497a3d507c19cd58d08ec8ed637ec55. The change adds backward tracing of explicit MI dependency evidence, UNVERIFIED handling, MI-scoped queue-empty requirements, and nonterminal recovery continuation, with six offline semantic guards. Focused workflow tests: 22 passed. Sanitized offline make check: PASS (178 tests, lint/format, basedpyright, locked sync, whitespace). The tests guard declarative workflow text, not live /loop execution. Issue remains open; PR is open and unmerged.

Implemented and pushed the workflow-only change in PR #40: https://forgejo.creatidy.com/Creatidy/model-intelligence/pulls/40. Base/merge base: `e77f7ae13132c9c202c4cf878bb0e42f2f29680d`; head: `4d189895a497a3d507c19cd58d08ec8ed637ec55`. The change adds backward tracing of explicit MI dependency evidence, `UNVERIFIED` handling, MI-scoped queue-empty requirements, and nonterminal recovery continuation, with six offline semantic guards. Focused workflow tests: 22 passed. Sanitized offline `make check`: PASS (178 tests, lint/format, basedpyright, locked sync, whitespace). The tests guard declarative workflow text, not live `/loop` execution. Issue remains open; PR is open and unmerged.