Files
msd-core/tests/emitted-drift-acks/1956-cross-artifact-fact-drift.json
Tom Boucher b7431a9259 feat(#1956): flag cross-artifact fact drift in the plan drift guard (#3259)
* test(#1956): failing-first contract for cross-artifact fact-drift pass

* feat(#1956): flag cross-artifact fact drift in the plan drift guard

* fix(#1956): correct config-key assertion and bidirectional lifecycle-lag exemption

* docs(#1956): document the cross-artifact axis in the architecture reference

* feat(#1956): decide the phase-status drift axis deterministically

* fix(#1956): scope the progress-table lookup, abstain without a position section, rank deferred

* docs(#1956): backfill changeset pr number

---------

Co-authored-by: sim <sim@local>
2026-08-09 14:23:35 -04:00

7 lines
1.8 KiB
JSON

{
"version": 1,
"paths": {
"plan-review-convergence.md": "#1956: the plan drift guard gains a second axis — a cross-artifact fact-drift pass sitting immediately after the existing source-grounding pass, under the SAME `plan_review.source_grounding` gate. Growth is the new section only (26195 -> 30672 bytes, +4477); no existing text was rewritten and no new config key was introduced. The addition is deliberately inline rather than extracted: ADR-1610 Decision 4 names eager `@`-import relocation as proxy-gaming (it shrinks the measured file while leaving loaded context unchanged or larger), legitimate extraction is Read-at-step lazy, and this pass has no lazy-read seam — it is prose the orchestrator must already hold when it runs the guard. The file remains far inside its DEFAULT tier hard cap of 40960 bytes (tests/workflow-size-budget.test.cjs), with ~11.4 KB of headroom. Content justification: source-grounding proves a plan's cited SYMBOLS exist in source; nothing proved that the same FACT stated in two planning artifacts still agrees. Because each phase runs in a fresh context an agent typically reads one artifact and trusts it, so a stale duplicate — a phase marked complete in STATE.md but in progress in ROADMAP.md, a success criterion the plan restates with a different outcome, a term used against its CONTEXT.md definition — silently steers it wrong. The pass keys on contradicting knowledge rather than similar-looking text (the false-positive mode the issue's own research comment names), gates on a three-way conjunction, defers the axes gsd-plan-checker already owns (Dimensions 1, 7b, 9) so nothing is reported twice, and is advisory only: it never sets hardBlock and contributes to neither HIGH_COUNT nor ACTIONABLE_COUNT, so a project carrying pre-existing drift can still converge."
}
}