docs: the README still said fifteen rules, and now a test says otherwise

README.md

@@ -552,7 +552,7 @@ Hardening beyond these honesty notes is tracked in
 | **Agent frameworks** | [CrewAI](adapters/crewai/) · [OpenSRE](adapters/opensre/) · Microsoft AGT audit files (Semantic Kernel, AutoGen, LangGraph via AGT) | LangChain · AutoGen native |
 | **MCP transport** | Stdio audit proxy (wrap any MCP server command) | SSE / streamable HTTP proxy |
 | **Observability / SIEM** | Loki · Grafana · Elastic ECS · Splunk HEC · CloudEvents v1.0 (Knative, EventBridge, Event Grid, Dapr, Kafka) · generic webhook | Datadog · New Relic |
-| **Detection formats** | In-engine sequence rules · LogQL · Elastic · Splunk · [Sigma pack](docs/integrations/sigma/README.md) (4 rules) | STIX/TAXII export |
+| **Detection formats** | In-engine sequence rules · LogQL · Elastic · Splunk · [Sigma pack](docs/integrations/sigma/README.md) (23 rules) | STIX/TAXII export |
 | **Policy engines** | Regex DLP manifest (`agentmetry/policies/dlp/`) · tool allow/deny YAML (`agentmetry/policies/tool/`) | OPA / Rego policy-as-code |
 | **Compliance docs** | [ISO 42001 mapping](docs/compliance/iso-42001-mapping.md) · [AI Act checklist](docs/compliance/ai-act-deployer-checklist.md) | SOC 2 evidence templates |
 
@@ -570,12 +570,17 @@ Matching is **ordered within a session**, not a threshold on one row.
 `credential-exfil` requires credential access (T1552) *then* network egress
 (TA0011), in that order. Reversed, it does not fire.
 
-Fifteen built-in rules ship today (plus YAML count rules), covering credential exfiltration, guardrail bypass,
+Fourteen published rules ship today, plus one held back as experimental and
+the YAML count rules. They cover credential exfiltration, guardrail bypass,
 download cradles, supply-chain merges, recon-then-collect, and both published
 [Agent Data Injection](https://arxiv.org/abs/2607.05120) chains.
 
 **[Every rule, how ordered matching works, and the research behind it →](docs/detection-rules.md)**
 
+That page also maps this project against the
+[OWASP Agentic Skills Top 10](https://owasp.org/www-project-agentic-skills-top-10/):
+two risks covered, three partial, and five that are somebody else's job.
+
 ```http
 GET /api/v1/audit/detections/{correlation_id}
 ```

apps/orchestrator/tests/test_readme_claims.py

@@ -179,3 +179,46 @@ def test_readme_does_not_reintroduce_the_overstatement():
     # happens without updating the README, the invitation stops working.
     assert "agentmetry/api/routes/audit.py" in text
     assert "agentmetry/core/audit/identity.py" in text
+
+def test_readme_rule_count_matches_the_registry():
+    """0.7.0 delisted a rule and the README kept saying fifteen.
+
+    That release existed to make published claims true. It fixed
+    `detection-rules.md`, the Sigma pack and the whitepaper, and missed the most
+    read document in the repository, where the number sat spelled out as a word
+    and so matched no numeric search.
+    """
+    from agentmetry.core.audit.detection.rules import BUILTIN_RULE_IDS
+
+    text = _readme().lower()
+    words = {
+        12: "twelve", 13: "thirteen", 14: "fourteen",
+        15: "fifteen", 16: "sixteen", 17: "seventeen",
+    }
+    published = len(BUILTIN_RULE_IDS)
+    correct = words[published]
+
+    for count, word in words.items():
+        if count == published:
+            continue
+        for phrase in (f"{word} built-in rules", f"{word} published rules", f"{count} built-in rules"):
+            assert phrase not in text, (
+                f"README says {phrase!r} but the registry publishes {published}. "
+                f"Use {correct!r}, and remember an experimental rule is not published."
+            )
+
+
+def test_readme_sigma_count_matches_the_pack():
+    """The pack is generated, so its size moves without anyone editing prose."""
+    import re
+
+    sigma_dir = Path(__file__).resolve().parents[3] / "docs" / "integrations" / "sigma"
+    if not sigma_dir.is_dir():
+        pytest.skip("sigma pack not present")
+    shipped = len(list(sigma_dir.glob("*.yml")))
+
+    match = re.search(r"Sigma pack\]\([^)]+\)\s*\((\d+) rules\)", _readme())
+    assert match, "README no longer states a Sigma rule count; update this test or the README"
+    assert int(match.group(1)) == shipped, (
+        f"README claims {match.group(1)} Sigma rules, pack ships {shipped}"
+    )