golem15/msd-core
4
0
Fork 0
Code Issues 1 Pull Requests Actions 2 Packages Projects Releases Wiki Activity

chore(#3421): delete orphan verify-phase workflow, migrate live gates to verifier (#3422)

Browse Source 
* chore(#1892): delete orphan verify-phase workflow, migrate live gates to verifier reference

* test(#1892): retarget structural suites from verify-phase.md to verifier-phase-gates.md

* chore(#1892): reword retired-workflow mentions for removed-but-needed lint

* test(#1892): correct stale surface labels in retargeted suites

* docs(#1892): add verifier-phase-gates row to locale inventories

* chore(#3421): backfill changeset pr number

---------

Co-authored-by: sim <sim@local>
This commit is contained in:
Tom Boucher
2026-08-13 21:22:03 -04:00
committed by GitHub
parent b77b7f8e56
commit d30c99bc92
52 changed files with 313 additions and 752 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/adr/1820-spec-optional-predicate-rail.md
View File

@@ -7,7 +7,7 @@
## Context
The edge-probe and prohibition-probe families surface a phase's unwritten edges and must-NOTs. Before #1820 they ran **only** in `spec-phase` (Step 5.5 / 5.6), so their value reached the verifier **only when a phase authored a SPEC.md**. But the verifier reads `must_haves` from **PLAN.md** frontmatter (`verify-phase.md`), not from SPEC.md — SPEC.md is only an optional source `plan-phase` lifts from. A phase planned without a SPEC therefore shipped with an empty predicate set and the probes' reach was lost. Per the load-bearing premise *verifier reach = spec reach*, that is a silent coverage hole precisely where a spec is thinnest.
The edge-probe and prohibition-probe families surface a phase's unwritten edges and must-NOTs. Before #1820 they ran **only** in `spec-phase` (Step 5.5 / 5.6), so their value reached the verifier **only when a phase authored a SPEC.md**. But the verifier reads `must_haves` from **PLAN.md** frontmatter (`verify-phase` (workflow, retired in #1892)), not from SPEC.md — SPEC.md is only an optional source `plan-phase` lifts from. A phase planned without a SPEC therefore shipped with an empty predicate set and the probes' reach was lost. Per the load-bearing premise *verifier reach = spec reach*, that is a silent coverage hole precisely where a spec is thinnest.
The fix wires the probe generators into `plan-phase`'s `must_haves` authoring as a **spec-optional fallback**: when a section is absent (no SPEC, or a SPEC that omits/empties that section), the fallback runs the probe and authors the predicates directly into PLAN.md. Deciding *whether a section was supplied* requires detecting the SPEC's `## Edge Coverage` / `## Prohibitions (must-NOT)` sections and counting their resolved rows. That detection first lived as ad-hoc `awk` inline in the workflow body, which hard-coded the header strings at the call site and hand-rolled markdown-table row counting — brittleness that produced two real bugs (an exact `^## Prohibitions$` anchor that missed the canonical `## Prohibitions (must-NOT)` heading, and a single-table row-counting assumption). Centralizing the detection made the knowledge testable and shared — and created a new Module seam, which is what makes this an ADR-worthy decision (contributor-standards: *"an ADR is required when a decision introduces a Module seam that other code will depend on"*).