fix(ledger): aprender-ci-tools shipped with no surface row, so the bin gate read 29/30 (#5170)

[noahgift]

Closes #5170. 0.71.1 fix row P1.

The red

scripts/surface_audit_bins_gate.sh (release pre-publish dogfood) compares the ledger's binaries with every [[bin]] target in cargo metadata. aprender-ci-tools, the C301 port of the CI helpers, had no row in docs/audits/surface_audit.csv.

RED before (origin/main):

surface ledger: 894 rows, binaries 29/30
RED  [[bin]] aprender-ci-tools ships with no row in docs/audits/surface_audit.csv

GREEN after (this head):

surface ledger: 904 rows, binaries 30/30
GREEN 30/30

The change

• docs/audits/surface_audit.csv: 10 rows for aprender-ci-tools, one per subcommand (publishable-crates, package-include-diff, coverage-report-scope, tarball-shrink-report, tarball-workspace, tarball-build-errors, dag-status, git-patch-id, llama-fit-verdict) plus --version. Each row cites its declaration line in crates/aprender-ci-tools/src/main.rs. Cluster: 12,test-harness. Gated: none. --version is quality 6, not the 7 the observed --version rows carry, because it was not run for this change.
• The denominator grows and the numerator does not, so every stated ratio moves from 144/894 (16.1%) to 144/904 (15.9%): the totals in contracts/apr-dogfood-coverage-v1.yaml (rows, ratio, binaries 30, 29 at zero), a per-binary row, the history line, and the T2 lines in the contract and in the apr-dogfood skill. Without these, G2.6 T2 pairing in check_dogfood_coverage.sh goes RED on the stale 144/894.
• Roadmap entry PMAT-5170 and the regenerated roadmap.yaml.

Checks run on this head

• surface_audit_bins_gate.sh: GREEN 30/30 (was RED 29/30)
• check_surface_audit_bins.sh (PR-time case table): pass
• check_dogfood_coverage.sh vs origin/main: PASS (G2.2 no row lost, G2.3 144/904, G2.5, G2.6 T2 pairing PASS, G2.4)
• pv validate contracts/apr-dogfood-coverage-v1.yaml: valid, 0 errors
• make readme-sync-check roadmap-aggregate-check: ok (README count unchanged)
• cargo fmt --all -- --check: clean
• No contract added or removed and no typed relation changed, so the ONT witness is unchanged.

ont-delta: none (ledger rows and restated coverage ratios; no ontology node, relation or contract id added or removed)

🤖 Generated with Claude Code

[noahgift]

seat 1, aprender-a7 APPROVE @ 605094d

Re-ran on a detached checkout of this head:

• scripts/surface_audit_bins_gate.sh (live cargo metadata): surface ledger: 904 rows, binaries 30/30 / GREEN 30/30, rc 0.
• Same gate with --csv = main's ledger: binaries 29/30, RED [[bin]] aprender-ci-tools ships with no row, rc 1. So the row is what turns it green.
• The ten rows match the binary: enum Cmd in crates/aprender-ci-tools/src/main.rs has exactly 9 variants, at lines 27/30/38/45/54/67/76/83/92 as cited, plus version at :15. All carry cluster 12 / test-harness, the id every other test-harness row uses.
• Coverage figures: 144/904 = 0.1593, binaries 30 with 29 at zero, and the per-binary entry aprender-ci-tools: {total: 10, covered: 0} are consistent; the T2 pair is updated on all three lines (contract invariant, falsification text, SKILL.md).
• cargo fmt --all -- --check clean.

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=5175 head=acec3ba6e90fe244044f6ad9651ebce5bf99a6d8 verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

[noahgift]

seat 2, df APPROVE @ 605094d (seat 2, aprender-df)

Read at this head:

• The Cmd enum in crates/aprender-ci-tools/src/main.rs has exactly nine variants, at lines 27, 30, 38, 45, 54, 67, 76, 83 and 92. The ten new ledger rows cite those same nine lines plus version = … at :15. Each row name is the clap kebab-case of its variant, so no variant is missing and no row is extra.
• The crate has no explicit [[bin]], so the binary name is the package name, aprender-ci-tools, which matches the ledger key.
• The arithmetic is consistent: 144/904 = 0.1593 (15.9%). binaries 30, of which 29 at zero, with the new per-binary row {total: 10, covered: 0}. The contract's T2 lines and the SKILL.md T2 line state the same pair, so G2.6 pairing is satisfied.
• Quality 6 on --version is the honest grade, because the PR says that row was not run.

Not re-run by me: surface_audit_bins_gate.sh, which needs cargo metadata. I rely on the RED 29/30 → GREEN 30/30 pair in the body, plus this head's CI.

[noahgift]

seat 2, aprender-df APPROVE @ acec3ba

Re-seat on the new head; my review on 605094d is void.

Delta 605094d..acec3ba (one commit, 2 files):

• contracts/apr-dogfood-coverage-v1.yaml: test-harness total 51 -> 61 and q5_6 total 706 -> 716. Both are +10, matching the 10 new aprender-ci-tools rows in docs/audits/surface_audit.csv; the covered counts are unchanged, as expected for rows that add uncovered surface.
• contracts/contracts.nt: regenerated. It adds 10 auditRow triples (one per new CSV row) and 6 ledgerOrphan triples, and it drops 3 unledgeredCommand triples for commands that now have a row. That is consistent with the CSV change. No shapes change.

The earlier review of the CSV rows still applies: the 10 cited line numbers match the subcommand enum in crates/aprender-ci-tools/src/main.rs.

[noahgift]

seat 1, aprender-a7 APPROVE @ acec3ba

My review on 605094d is void; this one replaces it. The delta since then is one commit: contracts.nt regenerated, plus two baselines in apr-dogfood-coverage-v1.yaml.

• Baselines: I recounted docs/audits/surface_audit.csv with a quote-aware CSV split (10 rows carry quoted commas, so a plain -F, miscounts). It gives 904 rows, test-harness 61 and q5_6 716, matching the new values. The quality buckets sum to 80+49+716+38+21 = 904.
• scripts/check_dogfood_coverage.sh on this head: PASS. It does not read the per-cluster or per-quality totals: with test-harness put back to 51 it still PASSes. Those two lines are checked only by hand or by dogfood_baseline.py, which is why I recounted.
• The P1 fix itself (bins gate GREEN 30/30 vs main's RED 29/30, ten rows matching main.rs) is unchanged from my first review.

Non-blocking finding, outside this PR's scope: the regenerated graph gives aprender-ci-tools 6 bin:ledgerOrphan triples: dag-status, git-patch-id, llama-fit-verdict, tarball-build-errors, tarball-shrink-report and tarball-workspace. The ledger rows are right: enum Cmd has those variants. So the ontology binary extractor sees only 3 of the 9 subcommands, the same 3 that were unledgeredCommand on main. No shape targets this binary yet, so nothing goes red. A bin-aprender-ci-tools--* contract with the usual ledgerOrphan maxCount 0 would go red on it. Worth a ticket against crates/aprender-contracts/src/ontology/extract/binary.rs.