diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 0f0cbdd..565e7dd 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -12,13 +12,18 @@ jobs:
       matrix:
         python-version: ["3.12", "3.13"]
     steps:
-      - uses: actions/checkout@v4
-      - uses: actions/setup-python@v5
+      # v6/v5 targeted Node 20, which runners now force onto Node 24 with a
+      # deprecation warning; v6 is the Node-24-native pair.
+      - uses: actions/checkout@v6
+      - uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
       - run: pip install -e ".[dev]"
-      - run: ruff check src tests
+      - run: ruff check src tests scripts
       - run: mypy
       # The default suite is fully offline: a conftest socket guard fails any test
       # that reaches for a non-loopback address; playwright/live are deselected.
       - run: python -m pytest -q
+      # Prose-vs-code drift has shipped twice (PR #1, PR #4) because nothing
+      # compared the README/CHANGELOG claims to the manifests and the suite.
+      - run: python scripts/check_consistency.py
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 3d05ade..f3d9457 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -27,10 +27,15 @@ Requires Python >= 3.12. Optional, for the browser-render path and its tests:
 Every change must pass the same gate CI runs. Run it before you push:
 
 ```bash
-.venv/bin/ruff check src tests && .venv/bin/mypy && .venv/bin/python -m pytest
+.venv/bin/ruff check src tests scripts && .venv/bin/mypy && \
+  .venv/bin/python -m pytest && .venv/bin/python scripts/check_consistency.py
 ```
 
 - **`ruff check`** — lint. **`mypy`** — strict type-checking. **`pytest`** — the offline suite.
+- **`check_consistency.py`** — asserts the repo's claims about itself are true: the version in
+  `pyproject.toml`, `__init__.py` and a released `CHANGELOG.md` section all agree, and every
+  "N tests" figure in the README equals what pytest actually collects. If you add tests or
+  bump the version, this is what tells you which prose went stale.
 - The default suite runs **fully offline**: a `conftest.py` socket guard fails any test that
   reaches a non-loopback address, so tests stay deterministic. HTTP is faked with
   `httpx.MockTransport` and fixture HTML.
diff --git a/README.md b/README.md
index c164e30..7b8d213 100644
--- a/README.md
+++ b/README.md
@@ -11,7 +11,7 @@ no telemetry.
 
 > *tearsheet (n.): a page torn from a publication and filed as proof it ran.*
 
-**Trust status:** qualified for heavy usage 2026-07-16 — 243 tests, a falsifiable live
+**Trust status:** qualified for heavy usage 2026-07-16 — 247 tests, a falsifiable live
 eval harness (verdict GREEN), and zero fabrications across the tool's entire recorded
 history. Its documented failure mode is *omission*, and the guards exist to make every
 omission loud. See [Trust](#trust).
@@ -161,7 +161,7 @@ a research tool for reading the public web — not for evading paywalls or bot d
 
 "Can it be trusted for heavy usage?" is a measurement here, not a feeling.
 
-- **Offline suite (243 tests, runs in the gate)**: guard boundary pins, cache-poisoning
+- **Offline suite (247 tests, runs in the gate)**: guard boundary pins, cache-poisoning
   regressions, truncation honesty, charset torture, structure torture, adversarial
   robustness — enforced fully offline by a loopback-only socket guard. The five REAL
   pages that defined the tool's probation (quo, smith.ai, dialpad, heyrosie, a LinkedIn
@@ -195,7 +195,7 @@ for figures you'll quote, treat a suspiciously small extraction of a rich page a
 .venv/bin/ruff check src tests && .venv/bin/mypy && .venv/bin/python -m pytest
 ```
 
-TDD throughout; the default suite (243 tests) runs entirely offline — `httpx.MockTransport`,
+TDD throughout; the default suite (247 tests) runs entirely offline — `httpx.MockTransport`,
 fixture HTML, and a conftest socket guard that fails any test reaching for a non-loopback
 address. Extras: `pytest -m playwright` (real chromium, local server), `pytest -m live`
 (real network). The live trust evaluation lives in `evals/` (see [Trust](#trust)).
diff --git a/scripts/check_consistency.py b/scripts/check_consistency.py
new file mode 100644
index 0000000..b13ac8c
--- /dev/null
+++ b/scripts/check_consistency.py
@@ -0,0 +1,123 @@
+"""Assert the repo's claims about itself are true.
+
+Two classes of drift have shipped here twice, both caught only by luck:
+
+  * PR #1  — "Fix version and test-count inconsistencies"
+  * PR #4  — README said "Since v0.1.5" while both manifests still said 0.1.4
+             and the CHANGELOG carried the change under [Unreleased]
+
+Neither the gate nor CI could see either one, because nothing compared the
+prose to the code. This script does. Run it in the gate.
+
+Checks, all against `pyproject.toml`'s version as canonical:
+  1. `src/tearsheet/__init__.py.__version__` matches it.
+  2. `CHANGELOG.md` has a released `## [X.Y.Z]` section for it — this is the
+     one that catches "documented as shipped while still sitting in
+     [Unreleased]".
+  3. Every "N tests" claim in `README.md` equals the number pytest actually
+     collects for the default (offline) suite.
+
+Exit 0 when everything agrees; 1 with a list of mismatches otherwise.
+"""
+
+from __future__ import annotations
+
+import re
+import subprocess
+import sys
+import tomllib
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+
+# "243 tests", "(243 tests, runs in the gate)", "suite (243 tests)" — any prose
+# claim about how many tests exist. Deliberately broad: a claim this script
+# cannot see is a claim that can rot.
+_README_TEST_CLAIM = re.compile(r"\(?(\d{2,4})\s+tests\b")
+_COLLECTED = re.compile(r"(\d+)(?:/\d+)?\s+tests? collected")
+
+
+def canonical_version() -> str:
+    data = tomllib.loads((REPO_ROOT / "pyproject.toml").read_text())
+    version = data["project"]["version"]
+    if not isinstance(version, str):  # pragma: no cover - malformed manifest
+        raise TypeError("pyproject.toml [project].version must be a string")
+    return version
+
+
+def check_dunder_version(expected: str) -> list[str]:
+    path = REPO_ROOT / "src" / "tearsheet" / "__init__.py"
+    m = re.search(r'__version__\s*=\s*"([^"]+)"', path.read_text())
+    if not m:
+        return [f"{path.name}: no `__version__ = \"X.Y.Z\"` found"]
+    if m.group(1) != expected:
+        return [f"{path.name}: __version__ {m.group(1)!r} != {expected!r}"]
+    return []
+
+
+def check_changelog_released(expected: str) -> list[str]:
+    path = REPO_ROOT / "CHANGELOG.md"
+    text = path.read_text()
+    if re.search(rf"^## \[{re.escape(expected)}\]", text, re.M):
+        return []
+    return [
+        f"{path.name}: no released `## [{expected}]` section — the current version is "
+        "still unreleased, or the section was never added"
+    ]
+
+
+def collected_test_count() -> tuple[int | None, list[str]]:
+    """How many tests the DEFAULT suite collects (what the gate and README mean)."""
+    proc = subprocess.run(  # noqa: S603 - fixed argv, no shell
+        [sys.executable, "-m", "pytest", "--collect-only", "-q"],
+        cwd=REPO_ROOT,
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    m = None
+    for line in reversed(proc.stdout.splitlines()):
+        m = _COLLECTED.search(line)
+        if m:
+            break
+    if not m:
+        return None, ["could not determine collected test count from pytest output"]
+    return int(m.group(1)), []
+
+
+def check_readme_test_counts(actual: int) -> list[str]:
+    path = REPO_ROOT / "README.md"
+    problems = []
+    for lineno, line in enumerate(path.read_text().splitlines(), start=1):
+        for m in _README_TEST_CLAIM.finditer(line):
+            claimed = int(m.group(1))
+            if claimed != actual:
+                problems.append(
+                    f"{path.name}:{lineno}: claims {claimed} tests, pytest collects {actual}"
+                )
+    return problems
+
+
+def main() -> int:
+    expected = canonical_version()
+    problems: list[str] = []
+    problems.extend(check_dunder_version(expected))
+    problems.extend(check_changelog_released(expected))
+
+    actual, count_problems = collected_test_count()
+    problems.extend(count_problems)
+    if actual is not None:
+        problems.extend(check_readme_test_counts(actual))
+
+    if problems:
+        print(f"Consistency check FAILED. pyproject.toml version is {expected!r}.")
+        for p in problems:
+            print(f"  - {p}")
+        return 1
+
+    print(f"Consistency OK. Version {expected} everywhere; README matches {actual} tests.")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())