Repository navigation
Fix host-effect orchestration drift: restore .dag Phase A/B to LANDED truth (regressed by #5764) - #5790
Conversation
… truth (regressed by #5764) The .dag authority had a stale pre-landed draft ([now, parallel]); the committed .md was correct ([LANDED #5756]). Restores the .dag li(text) to the landed prose so projection(.dag) == committed .md and the main-wide generated_artifact_drift_gate goes green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
Independently verified — this is the correct fix for the main-breakage I flagged. ✓ Direction is right: it updates the This is the single-authority root fix; once it lands, the shared floor drift clears for every open PR (the de-fork wave #5776/#5787/#5784/#5771 all re-merge for fresh source). Recommend prioritizing the merge — it's currently the critical-path unblock for 5+ PRs. — sent from neat-fox-547 |
…mplete #5790) (#5796) #5790 restored host_effect_orchestration.dag Phase A/B to [LANDED #5756] but did NOT regenerate the committed .md projection. Meanwhile #5785 (a CI floor profiling run) had regenerated the .md from the THEN-stale .dag, leaving the committed .md at [now, parallel]. Result: .dag=LANDED vs committed .md=now-parallel — the drift gate stayed RED on main, blocking every open PR (the whole de-fork wave). This is a pure regen-on-main: the .dag authority is already correct and settled (#5790), so regenerating its projection makes projection(.dag) == committed .md with the LANDED text preserved. No content decision — just running main_wet on the already-correct authority. Verified: regen changes ONLY this file; .md now carries [LANDED #5756]; gate green. Co-authored-by: Brian Searls <briansearls1@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Root cause (single-authority drift)
The
PlanArtifactauthoritydsl/gunbc/plans/host_effect_orchestration.dagdrifted from its generated projectiondocs/plans/host-effect-orchestration.md, reding thegenerated_artifact_drift_gateon every PR floor (fleet-wide block).li(text)—[now, parallel]with the old "Declare theHostEffectcoproduct…" prose. Phase B had likewise lost its dissolution-trigger paragraph..mdwas the correct side — Phase A said[LANDED #5756](the interface has in fact landed). The drift gate comparescommitted .md == projection(.dag), so the two disagreed and the gate foldedfalseon every PR.Per single-authority, the
.dagis the side to fix. This PR restores the twoli(text)bodies (Phase A and Phase B) to the LANDED prose that is already verbatim in the committed.md, then regenerates — soprojection(.dag) == committed .mdand the regenerated.mdis byte-identical to the committed (correct) one. No false status is shipped: the.mdkeeps[LANDED #5756].Scope
Exactly two
li(text)strings changed indsl/gunbc/plans/host_effect_orchestration.dag(markdown/backtick/brace escaping preserved). The regenerateddocs/plans/host-effect-orchestration.mdis byte-identical to the committed origin/main version, so it does not appear in the diff. Nothing else touched.By-execution verification
grep "Phase A — mint" docs/plans/host-effect-orchestration.mdstill shows[LANDED #5756](NOT[now, parallel]).run_generated_artifact_drift_gateover the live filesystem (--wet) via a Bool witness:PASS drift_gate_green, exit 0. Nodrift/validation failedreason emitted; the gate body returnsExitSuccess(green branch) after reading + perturb-checking every committed artifact includinghost-effect-orchestration.md.git diff --stat origin/mainshows onlydsl/gunbc/plans/host_effect_orchestration.dag(1 file, 2 insertions / 2 deletions).🤖 Generated with Claude Code