Repository navigation
Conversation
…e/ skipped (Refs #6042) Gate-class change split out of #6035: no-reader paths (docs, contracts, root *.md) select READERS only, ALL 18->14 of 20. Adds the case-table and readers guards as a nightly LAB step with its own ticket label (lab-affected-readers, one owner). PMAT-Ticket: PMAT-6042 PMAT-Batch: BATCH-aprender-n-receipt Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
noahgift
left a comment
noahgift
left a comment
There was a problem hiding this comment.
Seat 2 (w5382, non-author): APPROVE @4d3423dda71b7849c1151bda2e50a753ac73cd21
Narrowing judged: docs/, non-build contracts/, root *.md, .claude/ and shard-step-costs.tsv skip every crate except READERS plus aprender-core. Any mix with another path, model-families, bindings, BUILD_CONTRACTS or crate files stays ALL. include_str! sites covered: kernel_explain is in CORE (ALL); layer-parity, kernel-fusion, apr-model-capability, softmax are in BUILD_CONTRACTS. Own self-tests pass. Tool is LAB, no gate consumes it yet.
noahgift
left a comment
noahgift
left a comment
There was a problem hiding this comment.
Seat 1 (aprender-w5404, non-author): APPROVE @4d3423dda71b7849c1151bda2e50a753ac73cd21 — GATE-CLASS (narrows scripts/affected_set.sh test selection: docs/, contracts/, root *.md, .claude/, ci/shard-step-costs.tsv give READERS instead of ALL). Measured on this head in a scratch clone: affected_set.sh --self-test rc 0 (12 rows); check_affected_readers.sh --self-test rc 0 and tree run rc 0 (every path-escape crate is in READERS); check_affected_set.sh rc 0 incl. its MUTANT rows (emptied READERS and dropped res= fallback are caught). Read the full diff: build-input exceptions (binding.yaml, model-families, BUILD_CONTRACTS) stay ALL; the new fail-closed paths (missing --diff-from FILE, '..' path, empty metadata/root) all emit ALL. On main nothing but the nightly LAB self-test consumes affected_set.sh, so this PR changes no PR/merge/release gate today; the narrowing bites only when a gate starts consuming it. EMG-7 GREEN (1 commit, PMAT-6042, 0 merges); the new nightly step is LAB (own ticket label, never on a PR path). Privacy scan: only C324 rule names in comments (accepted). Open policy points for the arbiter/quorum, not settled by this seat: the cite is now #6042, not FLOW-010(b) (spec v1.4:94 does not say it); the commit body claim 'ALL 18->14 of 20' is not measured by me. Needs a second non-author seat, planted quorum and arbiter sign.
…mer must run the readers guard (Refs #6042) The READERS-only rows change the tool's output, not any PR run: no PR, merge-queue or release job calls affected_set.sh. READERS is hand-maintained and fails open, so the first consuming PR must run check_affected_readers.sh on its own PR path. PMAT-Ticket: PMAT-6042 PMAT-Batch: BATCH-aprender-n-receipt Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Gate-class: affected-set READERS-only for docs/contracts paths (PMAT-6042)
Split out of #6035.
gate/affected-set-readers-only@4d3423dda7 (w5032), one commit, linear on main 45aa5d1. Gate-class: stays DRAFT until seat 2 + quorum 3/3 + arbiter sign.scripts/affected_set.sh:docs/contracts/*.mdgive READERS, not ALL;.claude/is skipped. Measured: ALL 18 -> 14 of 20 paths.scripts/check_affected_set.sh: case table updated for the new selection (thedocsrow expects READERS only).scripts/check_affected_readers.sh: new readers check..github/workflows/guards-nightly.yml: the case table and readers check run as a nightly LAB step with its own ticket labellab-affected-readers(one owner, C324); the self-test runs first. Nothing is added to a PR, merge-queue or release job.scripts/guards_nightly_manifest.txt: manifest rows for the two LAB steps.The old "FLOW-010 (b)" cite is replaced with #6042. Fragments 296/297 are not touched.
Refs #6042
ont-delta: none no new type, shape or reason; narrows an existing selection
🤖 Generated with Claude Code