fix(funding): the manifest guard billed contributors for my public copy

apps/orchestrator/tests/test_funding_manifest.py

@@ -10,9 +10,17 @@
 for the version string. This does the same for the manifest, so a figure that
 drifts fails here instead of in front of a funder.
 
-Only the claims that a command can settle are pinned. The download count and
-the test count are stamped with a date in the manifest itself and are left
-alone, because a number that says when it was true does not become false.
+Only claims that move on a *release* are pinned: the version and the licence.
+Anything that moves when somebody contributes is not.
+
+That distinction was learned the expensive way. The first version of this file
+pinned the benchmark case count too, and the next contributor PR to add corpus
+cases went red on `funding.json`, a file they had never touched and had no
+business editing. A guard that taxes contributors for the maintainer's public
+copy is worse than the drift it prevents, so the manifest no longer quotes
+counts at all and this no longer checks them. The README quotes them, and
+`test_readme_claims.py` guards that, where the change and the claim live in the
+same pull request.
 """
 
 from __future__ import annotations
@@ -23,7 +31,6 @@
 
 import pytest
 
-from agentmetry.core.audit.detection.benchmark import load_corpus
 from agentmetry.core.version import __version__
 
 MANIFEST = Path(__file__).resolve().parents[3] / "funding.json"
@@ -64,14 +71,6 @@ def test_quoted_version_matches_the_package():
     )
 
 
-def test_quoted_benchmark_case_count_matches_the_corpus():
-    quoted = re.search(r"(\d+) case detection benchmark", _project_description())
-    assert quoted, "funding.json no longer quotes a case count; update this test or the manifest"
-    assert int(quoted.group(1)) == len(load_corpus()), (
-        "funding.json quotes a different case count than the corpus holds"
-    )
-
-
 def test_declared_licence_matches_the_package_metadata():
     pyproject = (Path(__file__).resolve().parents[1] / "pyproject.toml").read_text(
         encoding="utf-8"
@@ -83,3 +82,15 @@ def test_declared_licence_matches_the_package_metadata():
     assert licences == [f"spdx:{declared.group(1)}"], (
         "funding.json declares a different licence than the package does"
     )
+
+
+def test_the_manifest_quotes_no_count_that_a_contribution_can_move():
+    """The regression that made this change necessary.
+
+    A number here is a number somebody else's pull request can invalidate.
+    """
+    description = _project_description()
+    assert not re.search(r"\d[\d,]* (?:tests|cases|case )", description), (
+        "funding.json quotes a count that moves when somebody contributes; the "
+        "README is the place for those"
+    )

funding.json

@@ -15,7 +15,7 @@
     {
       "guid": "agentmetry",
       "name": "Agentmetry",
-      "description": "AI coding agents run shell commands, edit files and call third-party tools hundreds of times an hour, largely unattended. What they actually did is recorded nowhere by default. The agent transcript is a conversation rather than a record of effects, the editor keeps no durable log, and endpoint security sees only that one process spawned another, which is what a coding agent looks like when it is working correctly.\n\nAgentmetry records what an agent did at the tool boundary, correlates sequences into detections, and writes a hash-chained JSONL trail the operator owns, forwarding into whatever SIEM they already run. It is local-first: there is no vendor cloud, and no telemetry reaches the maintainer. Tool arguments are hashed inside the hook process, so plaintext never crosses the wire.\n\nVerifiable as of 2026-09-02: version 0.7.0 on PyPI, Apache-2.0, 1,186 tests, a 54 case detection benchmark at zero misses and zero false positives, MITRE ATT&CK and ATLAS mappings, and four SIEM adapters. Every one of those figures is printed by a command in the README.\n\nFunding would go to making the format underneath it common rather than ours: a standalone agent session record specification, mappings into Elastic Common Schema, OCSF, Splunk CIM and Google SecOps UDM, and a conformance suite so a third party can verify their own mapping without reading our Python. The success condition is other vendors adopting the fields, including ones who compete with us.",
+      "description": "AI coding agents run shell commands, edit files and call third-party tools hundreds of times an hour, largely unattended. What they actually did is recorded nowhere by default. The agent transcript is a conversation rather than a record of effects, the editor keeps no durable log, and endpoint security sees only that one process spawned another, which is what a coding agent looks like when it is working correctly.\n\nAgentmetry records what an agent did at the tool boundary, correlates sequences into detections, and writes a hash-chained JSONL trail the operator owns, forwarding into whatever SIEM they already run. It is local-first: there is no vendor cloud, and no telemetry reaches the maintainer. Tool arguments are hashed inside the hook process, so plaintext never crosses the wire.\n\nVerifiable as of 2026-09-04: version 0.7.0 on PyPI, Apache-2.0, a detection benchmark that runs at zero misses and zero false positives, MITRE ATT&CK and ATLAS mappings, and four SIEM adapters. Case and test counts are deliberately not quoted here, because they move every time somebody contributes. The README prints the current ones and a test checks that it still prints them correctly.\n\nFunding would go to making the format underneath it common rather than ours: a standalone agent session record specification, mappings into Elastic Common Schema, OCSF, Splunk CIM and Google SecOps UDM, and a conformance suite so a third party can verify their own mapping without reading our Python. The success condition is other vendors adopting the fields, including ones who compete with us.",
       "webpageUrl": {
         "url": "https://agentmetry.ai",
         "wellKnown": "https://agentmetry.ai/.well-known/funding-manifest-urls"