dogfood: pre-publish reads fmt, clippy, test and coverage from the commit's CI (#4672)

contracts/dogfood-runner-unification-v1.yaml

@@ -9,6 +9,7 @@ metadata:
   - 'scripts/check_verifier_pinning.sh — the gate: static scan + behavioural probe of both pins'
   - 'scripts/check_dogfood_shim.sh — the gate: one runner, one prose, and the user-scope copy stays a gated shim'
   - 'scripts/check_dogfood_row_order.sh — the guard (report mode until three green nights): long rows last, no early stop, a red cheap row keeps the verdict red, row times (aprender#4672)'
+  - 'scripts/check_dogfood_reads_ci.sh + scripts/release/sha_checks.sh — pre-publish reads fmt and test from the CI of the same commit and coverage from the receipt cut_tag requires (tag_coverage_gate.sh --resolve); no green receipt for that sha is red; clippy stays measured (aprender#4672)'
   - 'scripts/.dogfood-shim-reference.sh — the reviewed text of the user-scope shim'
   - '.claude/skills/dogfood/SKILL.md — the one canonical prose, with an explicit `name:` (#2332)'
   - 'aprender#2361 — a user-scope skill shadowed the repo one; hardening it edited a file that never ran'
@@ -443,6 +444,37 @@ falsification_tests:
   test: 'bash scripts/check_dogfood_row_order.sh --self-test   # the five timing rows'
   if_fails: 'a release-day check over its budget passes its own timing row, or a missing measurement reads as on time'
   mutation: 'the six planted timing rows (ok, over, at budget, hole, absent, a budget of 15m) must give rc 0, 1, 0, 2, 2, 2 (observed 2026-10-04)'
+- id: F-DOG1-032
+  rule: in --phase pre-publish, fmt and test read the commit's CI and coverage reads the receipt cut_tag requires; none re-runs; clippy stays measured
+  prediction: >
+    In scripts/dogfood.sh the fmt and test rows are ci_mark calls inside the pre-publish
+    branch (fmt reads `ci / gate`, test reads `workspace-test`), and the coverage row is
+    cov_mark, which PASSes only when `tag_coverage_gate.sh --resolve HEAD` exits 0: the
+    coverage-nightly receipt (#4734) that autopilot cut_tag already requires before the tag.
+    Their measuring commands sit only in that branch's else, so every other phase still
+    measures them. clippy is measured in every phase, because the `ci / gate` lint is
+    narrower (default features, unused-variables allowed, a lib-only fallback).
+    `workspace-test` is nextest --lib over the workspace; the root facade's doctests and
+    bin tests are not in it. Reviews and a C314 round on aprender#4886 settled these.
+  test: 'bash scripts/check_dogfood_reads_ci.sh   # R2'
+  if_fails: 'the release dogfood re-runs work CI already did for the same commit, a row reads the wrong check, clippy reads a narrower lint, or pre-publish passes a coverage row it never read'
+  mutation: 'put `gate fmt` back in pre-publish (rerun), point test at `ci / gate` (wrongspec), read clippy from CI (clippyread), pass coverage unread (covpass), drop the coverage row (covmissing); each must turn RED (observed 2026-10-06 in --self-test)'
+- id: F-DOG1-033
+  rule: no green receipt for the exact sha is RED, never a skip
+  prediction: >
+    The runner's own CI-read block, ci_mark() and cov_mark(), lifted and driven against
+    planted check rows and a planted coverage reader for a real commit, give PASS only
+    when each check, or the coverage receipt, is green for THAT sha, and FAIL on a red
+    check for the sha, a missing check, a green check on a different sha, a refused or
+    other-sha coverage receipt, a failed GitHub read, missing readers and a tree with
+    uncommitted tracked changes. scripts/release/sha_checks.sh holds the same rule on its
+    own (newest run wins, a queued rerun counts as newest, runs that disagree at one start
+    time are red, pending, cancelled and skipped are red, a same-named check in another
+    workflow does not count); tag_coverage_gate.sh --self-test holds it for coverage (no
+    receipt, other sha, below COV_FLOOR, gh failing).
+  test: 'bash scripts/check_dogfood_reads_ci.sh --self-test; bash scripts/release/sha_checks.sh --self-test; bash scripts/release/tag_coverage_gate.sh --self-test'
+  if_fails: 'a check or receipt that is red, absent, or for another commit lets the release dogfood say GO'
+  mutation: 'let a dirty tree read HEAD (nodirty), make the no-receipt fallback a PASS (skipread), pass a refused coverage receipt (covskip), resolve another sha (covwrongsha); each must turn RED; sha_checks.sh plants 21 rows (observed 2026-10-06)'
 proof_obligations:
 - id: DRU-INV-001
   type: invariant

evidence/pr-review/4886/6bac6be648bdcee2c8b8c0c7c640d1419d34703b/findings.sarif

@@ -0,0 +1,136 @@
+{
+  "$schema": "https://docs.oasis-open.org/sarif/sarif/v2.1.0/errata01/os/schemas/sarif-schema-2.1.0.json",
+  "version": "2.1.0",
+  "runs": [
+    {
+      "tool": {"driver": {"name": "pmat"}},
+      "invocations": [{"executionSuccessful": true, "toolExecutionNotifications": []}],
+      "results": [
+        {
+          "ruleId": "duplication_scan_no_hits",
+          "level": "note",
+          "message": {"text": "The lexical duplication scan over head, all 1540 unmerged sibling branches and the merge-base-to-main range (0 files) found 0 hits for the 13 symbols the diff adds. The diff changes no Rust file, so the pmat semantic index (Rust-only) was queried but cannot reach it."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["bash", "scripts/pr_review_duplication_scan.sh", "--base", "88bf12c8710dd29617a28add806a37c3edac42d2", "--head", "6bac6be648bdcee2c8b8c0c7c640d1419d34703b", "--json", "<out>"],
+            "exit_code": 0,
+            "stdout_sha256": "35d6586b94bd0567dd5909f59e9a6a8041e90d521e3fdcd466b93441535574fc",
+            "failure_scenario": "A re-implementation under a different name is invisible to a name match; that residual is not covered.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "diff_patch_id_run_by_parent",
+          "level": "note",
+          "message": {"text": "diff_patch_id 30be26cb97a09421c8df64ea78e5fa9bd2505a7d over 88bf12c871..6bac6be648 (evidence/pr-review/4886/ excluded) was computed by the PARENT session, not by this lane, under a cop ruling: subagent-lock.sh exempts the parent by design. Control by the parent: the same function over e95ae8280f..deb4f19d5e gives 6ff6a07990, equal to the A2 line CI printed. The stdout_sha256 here is the digest of the reported id plus a newline, taken by this lane from the value relayed to it; this lane did not observe that stdout."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["bash", "-c", ". scripts/lib/pr_review_patch_id.sh && prpid_compute <worktree> 88bf12c8710dd29617a28add806a37c3edac42d2 6bac6be648bdcee2c8b8c0c7c640d1419d34703b 4886"],
+            "exit_code": 0,
+            "stdout_sha256": "4ea3e412e1fbb421c8243f3851823ac59632a5d513949ef8636ebfb92f6aa850",
+            "failure_scenario": "If the relayed id were wrong, Arm 4 would find no receipt bound to the subject diff and refuse the merge, so the error cannot arm a merge that should not land.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "lane_hook_refusals",
+          "level": "note",
+          "message": {"text": "This lane was refused by the PreToolUse hook subagent-lock.sh (a review lane runs read-only git only) on two git subcommands: patch-id (via prpid_compute) and archive (the disposable tree for the agy lane). Neither was routed around: the patch-id was relayed by the parent; the disposable tree was built with tar from the clean worktree (tracked files, no .git, with evidence/, target/ and .pmat/ left out), a file copy and not a git operation."},
+          "properties": {
+            "grounding": "asserted",
+            "rationale": "The refusals are the hook output this lane received verbatim; no command ran, so there is no exit code or stdout to measure.",
+            "failure_scenario": "A reader of the receipt would otherwise take the tar-built agy workspace for a git-archive tree, and the patch-id for one this lane computed.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "dead_elif_pre_publish_coverage",
+          "level": "warning",
+          "message": {"text": "scripts/dogfood.sh line 1857, `elif [ \"$DOGFOOD_PHASE\" = pre-publish ]`, sits inside the `else` (line 1819) of the new `if [ \"$DOGFOOD_PHASE\" = pre-publish ]` at line 1816, which closes at line 1865. It can never be true, so the #3957 F1b branch and its FAIL message are dead code; pre-publish coverage is now decided only by cov_mark."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["sed", "-n", "1816p;1819p;1857p;1865p", "scripts/dogfood.sh"],
+            "exit_code": 0,
+            "stdout_sha256": "24dcf29e1f7cec413070882991cd914f534409f6da1bbcee4e21703639b01afa",
+            "failure_scenario": "A later edit trusts the F1b branch (coverage below floor in pre-publish is RED with its owed-by message) to run; it does not.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "ci_read_narrower_than_measured_test",
+          "level": "note",
+          "message": {"text": "In pre-publish the test row reads `workspace-test` (nextest --lib over the workspace) where the replaced row ran `cargo test $FEATS` (doctests, integration targets, bin tests). The runner comment in the 2c block discloses this; the receipt row says only `read, not re-run`."},
+          "properties": {
+            "grounding": "asserted",
+            "rationale": "Taken from the diff comment in scripts/dogfood.sh and CLAUDE.md Testing (only a subset of integration targets is wired into CI). Not measured by running anything.",
+            "failure_scenario": "A doctest or integration-only failure that CI does not run passes the pre-publish test row.",
+            "precision_class": "advisory"
+          }
+        }
+      ]
+    },
+    {
+      "tool": {"driver": {"name": "cargo-mutants"}},
+      "invocations": [{"executionSuccessful": true, "toolExecutionNotifications": []}],
+      "results": [
+        {
+          "ruleId": "committed_plants_killed",
+          "level": "note",
+          "message": {"text": "All 9 committed planted defects of check_dogfood_reads_ci.sh are killed (--self-test exits 0); sha_checks.sh --self-test is 21/21."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["bash", "scripts/check_dogfood_reads_ci.sh", "--self-test"],
+            "exit_code": 0,
+            "stdout_sha256": "542142ef51e2f92a7d6370d1e99d028e0e81b1e4b2ef29ee09f04a2e63adb258",
+            "failure_scenario": "If a plant survived, a runner defect the guard claims to catch would pass unnoticed.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "extra_mutant_cov_inverted_killed",
+          "level": "note",
+          "message": {"text": "Extra mutant: cov_mark records PASS when the reader's rc is non-zero (-eq 0 changed to -ne 0), in a scratch copy of the guard tree. Killed: the guard exits 1 with 8 FAIL lines."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["bash", "<scratch copy, rc test inverted>/scripts/check_dogfood_reads_ci.sh"],
+            "exit_code": 1,
+            "stdout_sha256": "66bab24d53e48cbc40b95cf25f5836e858a2f4d0a88c6a5006da929a6059c8df",
+            "failure_scenario": "If it survived, an inverted coverage verdict would pass.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "extra_mutant_git_dir_survives",
+          "level": "warning",
+          "message": {"text": "Extra mutant outside the committed set: cov_mark without the GIT_DIR prefix on the tag_coverage_gate.sh call (the commit message says the gate's git reads go through GIT_DIR, with no cd). The guard on a scratch copy with that edit still exits 0: R3 drives cov_mark against a planted reader that does not read git."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["bash", "<scratch copy, GIT_DIR prefix removed>/scripts/check_dogfood_reads_ci.sh"],
+            "exit_code": 0,
+            "stdout_sha256": "bd261c161c3a1c825e81c5cbf508bf68441b1f404b40fdb1f83a8a0f48bcfe60",
+            "failure_scenario": "A later edit drops GIT_DIR and the real tag_coverage_gate.sh reads git from the caller's cwd; the guard stays green and the commit message's claim is untested.",
+            "precision_class": "advisory"
+          }
+        },
+        {
+          "ruleId": "extra_mutant_spec_substring_survives",
+          "level": "warning",
+          "message": {"text": "Extra mutant outside the committed set: ci_mark matching the sha_checks line by a substring of the spec instead of the ` <spec> -- ` field. The guard on a scratch copy still exits 0. Likely behaviourally equivalent today (the two specs share no text); that is not proven."},
+          "properties": {
+            "grounding": "measured",
+            "command": ["bash", "<scratch copy, grep -F widened to the bare spec>/scripts/check_dogfood_reads_ci.sh"],
+            "exit_code": 0,
+            "stdout_sha256": "bd261c161c3a1c825e81c5cbf508bf68441b1f404b40fdb1f83a8a0f48bcfe60",
+            "failure_scenario": "A future spec that is a substring of another check's name would read that check's line and could PASS on it.",
+            "precision_class": "advisory"
+          }
+        }
+      ]
+    },
+    {
+      "tool": {"driver": {"name": "antigravity"}},
+      "invocations": [{"executionSuccessful": true, "toolExecutionNotifications": []}],
+      "results": []
+    }
+  ]
+}

evidence/pr-review/4886/6bac6be648bdcee2c8b8c0c7c640d1419d34703b/receipt.intoto.jsonl

@@ -0,0 +1 @@
+{"_type":"https://in-toto.io/Statement/v1","subject":[{"name":"git+https://github.com/paiml/aprender","digest":{"sha1":"6bac6be648bdcee2c8b8c0c7c640d1419d34703b"}}],"predicateType":"https://paiml.dev/attestations/pr-review/v2","predicate":{"skill_version":"2.1.0","attestation_level":"L1-self","pr":4886,"base_sha":"88bf12c8710dd29617a28add806a37c3edac42d2","head_sha":"6bac6be648bdcee2c8b8c0c7c640d1419d34703b","author_actor":{"kind":"agent","id":"agent:claude-opus-5-5/build-kaizen-4672-2c"},"reviewer_actor":{"kind":"agent","id":"agent:claude-sonnet-5-5/pr-review-4886-lane"},"affected_crates":[],"verdict":"FINDINGS","consultations":{"pmat":{"status":"consulted","transport":"cli","transport_unavailable":["mcp: not attempted this run (cli answered)"],"index_commit":"6bac6be648bdcee2c8b8c0c7c640d1419d34703b","index_is_ancestor":true,"index_worktree_dirty":false,"complexity_delta":[],"tdg_delta":[],"satd_introduced":[],"cache_hits":0,"duplication_hits":[],"duplication_coverage":{"rust":"lexical","shell":"lexical","python":"lexical","config":"lexical","docs":"lexical","other":"lexical","sibling_branches":"lexical","merge_base_to_main":"lexical"},"duplication_horizon":["head=6bac6be648bdcee2c8b8c0c7c640d1419d34703b","siblings=refs/remotes/origin/* unmerged into origin/main","merge_base_to_main=88bf12c8710dd29617a28add806a37c3edac42d2..refs/remotes/origin/main"],"horizon_branches_total":1540,"horizon_branches_scanned":1540,"merge_base_to_main_files":0,"symbols_searched":13,"queries":[{"q":"pre-publish coverage reads receipt tag_coverage_gate","command":["pmat","query","pre-publish coverage reads receipt tag_coverage_gate","--limit","10"],"cwd":"the review worktree at head","results":10,"note":"The stale index was rebuilt at head (94427 functions) by this query. The semantic index is Rust-only and the diff changes no Rust. No TODO/FIXME/HACK/XXX on any added line.","stdout_sha256":"0c2f0cb4a5ba3de18e72e2138df7d266883a839a2f4b44fb0539dcf7e65726ad"}]},"cuda":{"status":"not-triggered","trigger_reason":"no changed path or commit message matches a section 3.B pattern (guard --match-path on all 4 paths and --match-message both exit 1) and the diff makes no device-behaviour claim: a shell runner, a guard, a helper and a contract"},"crux":{"status":"not-triggered","trigger_reason":"the guard --match-crux-surface predicate matches none of the diff added lines, and the diff changes no subcommand, flag, route, MCP tool, config key or output format"},"mutation":{"status":"consulted","scope":"guard","attempted":12,"killed":10,"survivors":[{"mutant":"cov_mark without the GIT_DIR prefix on the tag_coverage_gate.sh call","file":"scripts/dogfood.sh","line":697,"killed":false},{"mutant":"ci_mark matches the sha_checks line by a bare spec substring","file":"scripts/dogfood.sh","line":687,"killed":false}]},"antigravity":{"status":"consulted","attempted":1,"agy_version":"agy 1.3.0","binary_path":"/home/noah/.local/bin/agy","model_id":"gemini-3.1-pro-high","model_family":"google/gemini (requested id; unverifiable from the receipt)","exit_code":0,"duration_seconds":109.296312252,"agy_status":"SUCCESS","usage":{"input_tokens":108355,"output_tokens":8950,"thinking_tokens":7055,"cache_read_tokens":758376,"total_tokens":117305},"output_check":{"structured_output_present":true,"reviewed":true,"schema_valid":true},"reverified_by_primary":false,"divergence":{"agreed":0,"agy_only":0,"primary_only":2,"contradicted":0},"findings":[]}},"findings_ref":{"path":"findings.sarif","sha256":"3925c7448eb668ad6c20319a8a3757ac829d6be324761ab801eeafdd235fa900"},"cost":{"input_tokens":108355,"output_tokens":8950,"wall_seconds":1500},"diff_patch_id":"30be26cb97a09421c8df64ea78e5fa9bd2505a7d","diff_patch_id_algo":"git-patch-id-verbatim/pinned-diff-v1","notes":["diff_patch_id was computed by the parent session under a cop ruling (subagent-lock.sh exempts the parent); control e95ae8280f..deb4f19d5e gives 6ff6a07990, equal to the CI A2 line. This lane was refused by subagent-lock.sh on patch-id and on archive and did not route around either; see SARIF rules diff_patch_id_run_by_parent and lane_hook_refusals."]}}