feat(B-0170.4): seed eval-set fixture for count-drift regression coverage

tools/substrate-claim-checker/README.md

@@ -97,6 +97,13 @@ specifically; instances marked count-drift in the memo's
 "Recurring sub-classes" section are this tool's primary regression
 suite.
 
+Frozen on-disk fixtures live in `fixtures/`. Each fixture is a minimal
+markdown file reproducing one historical drift instance, paired with
+a regression test in `fixtures.test.ts`. The fixture seed is one
+count-drift case from PR #1259; additional sub-classes land
+incrementally per B-0170.4. See `fixtures/README.md` for the index
+and the "adding a new fixture" procedure.
+
 ## Hooks integration (planned, not v0)
 
 Per the verify-then-claim memo's mechanization-path section:

tools/substrate-claim-checker/fixtures.test.ts

@@ -0,0 +1,37 @@
+/**
+ * Eval-set fixture regression tests for substrate-claim-checker (B-0170).
+ *
+ * Each test runs an existing `check-*.ts` against a frozen historical
+ * drift fixture under `fixtures/` and asserts that the checker still
+ * detects the empirical drift that prompted the fixture's capture.
+ *
+ * Run with `bun test tools/substrate-claim-checker/fixtures.test.ts`.
+ *
+ * Adding fixtures: see `fixtures/README.md`.
+ */
+
+import { describe, expect, test } from "bun:test";
+import { join } from "node:path";
+import { checkFile as checkCounts } from "./check-counts.ts";
+
+const fixtures = join(import.meta.dir, "fixtures");
+
+describe("eval-set fixtures / count drift", () => {
+  test("count-drift-9-vs-15.md — claim '9 drift instances' vs 15-row table is detected", () => {
+    const result = checkCounts(join(fixtures, "count-drift-9-vs-15.md"));
+    expect(result.ok).toBe(true);
+    // Per PR #3611 review threads (chatgpt-codex-connector + copilot):
+    // assert exact finding count and pin the body claim's line so a
+    // regression in body-claim detection cannot be masked by an
+    // HTML-comment match. The fixture's provenance comment is
+    // intentionally worded to not restate the `<number> <noun>` pair
+    // from the body claim.
+    expect(result.findings.length).toBe(1);
+    const finding = result.findings[0]!;
+    expect(finding.line).toBe(24);
+    expect(finding.claimedCount).toBe(9);
+    expect(finding.actualCount).toBe(15);
+    expect(finding.claim).toContain("drift instances");
+    expect(finding.claimIsMinimum).toBe(false);
+  });
+});

tools/substrate-claim-checker/fixtures/README.md

@@ -0,0 +1,46 @@
+# substrate-claim-checker eval-set fixtures
+
+Frozen historical drift instances from the verify-then-claim discipline
+memo's body table — the canonical eval-set for B-0170.
+
+Each fixture is a small, self-contained markdown file that demonstrates
+ONE drift sub-class as it actually surfaced in shipped substrate. The
+fixtures are NOT pristine examples; they preserve enough of the original
+PR's substance for the checker's behaviour against the historical case
+to be reproducible regression coverage.
+
+## Why on-disk fixtures (not inline test strings)
+
+Inline test strings in `*.test.ts` files cover the synthetic-case axis
+(does the checker work on toy inputs?). Frozen fixtures cover the
+empirical-case axis (does the checker still catch the actual drift the
+PR encountered?). Both axes matter; this directory holds the empirical
+axis.
+
+## Fixture index
+
+| Sub-class | Fixture | Anchor PR / history | Expected finding |
+|---|---|---|---|
+| count drift | `count-drift-9-vs-15.md` | PR #1259 (`review(pr-1257-postmerge): verify-then-claim count drift (9→18+) frontmatter + body + MEMORY.md`) | "9 drift instances" claim vs 15-row table |
+
+Add a new row when a new fixture lands.
+
+## Adding a new fixture
+
+1. Pick a historical drift instance from the verify-then-claim memo's
+   body table (canonical).
+2. Create the smallest markdown file that reproduces the drift pattern.
+   The fixture body should be runnable through the existing checker
+   without external dependencies.
+3. Document the anchor PR (or memo cite) inline at the top of the
+   fixture as an HTML comment so the provenance survives.
+4. Add a regression test in `fixtures.test.ts` that invokes the
+   relevant `check-*.ts` and asserts on the finding count + shape.
+5. Update the fixture index above.
+
+## Composes with
+
+- `tools/substrate-claim-checker/check-counts.ts` and siblings — the
+  checkers the fixtures regress against
+- B-0170 (the parent backlog row this directory contributes to)
+- The verify-then-claim memo's body table (the canonical eval-set)

tools/substrate-claim-checker/fixtures/count-drift-9-vs-15.md

@@ -0,0 +1,42 @@
+<!--
+Eval-set fixture for substrate-claim-checker (B-0170).
+
+Reproduces the count-drift pattern surfaced in PR #1259
+`review(pr-1257-postmerge): verify-then-claim count drift
+(9→18+) frontmatter + body + MEMORY.md` — the body narrative
+claims a smaller count than the body table actually records.
+The exact values are visible inline below.
+
+The fixture is intentionally minimal: only the count claim,
+the table, and enough framing for the checker to find the
+nearest-table within its 50-line window.
+
+NOTE: this comment intentionally avoids restating the exact
+`<number> <noun>` pair from the body claim. Restating it would
+produce a spurious second matching claim from the HTML
+provenance and let regressions in body-claim detection slip
+past `fixtures.test.ts` (per PR #3611 review threads from
+chatgpt-codex-connector + copilot-pull-request-reviewer).
+-->
+
+# Drift catalogue (eval-set fixture)
+
+The catalogue tracks 9 drift instances across the substrate.
+
+| # | Instance | Sub-class |
+|---|---|---|
+| 1 | row-1 | count |
+| 2 | row-2 | count |
+| 3 | row-3 | existence |
+| 4 | row-4 | path-form |
+| 5 | row-5 | cross-surface |
+| 6 | row-6 | count |
+| 7 | row-7 | convention |
+| 8 | row-8 | semantic-equivalence |
+| 9 | row-9 | empirical-output |
+| 10 | row-10 | self-recursive |
+| 11 | row-11 | count |
+| 12 | row-12 | existence |
+| 13 | row-13 | path-form |
+| 14 | row-14 | cross-surface |
+| 15 | row-15 | convention |