[codex] add evidence-bound failure clustering

src/ofw/__init__.py

@@ -26,6 +26,21 @@
     Sha256Digest,
     WorkspaceFile,
 )
+from ofw.diagnosis import (
+    ClusterRevisionRef,
+    ClusterState,
+    DiagnosisError,
+    DiagnosisErrorCode,
+    DiagnosisResult,
+    DiagnosisRun,
+    EvidenceAnchor,
+    EvidenceAnchorKind,
+    FailureCluster,
+    MechanismKey,
+    PythonDiagnoser,
+    Severity,
+    TraceDiagnosis,
+)
 from ofw.harness import EditableFile, Harness, Subagent, Tool, editable
 from ofw.mine import (
     Mine,
@@ -109,6 +124,7 @@ class _OfwNamespace:
     ServiceName = ServiceName
     MiningPolicy = MiningPolicy
     ScoreName = ScoreName
+    PythonDiagnoser = PythonDiagnoser
 
     def editable(self, path: Path) -> EditableFile:
         return editable(path)
@@ -158,6 +174,8 @@ def read_snapshot_content(
     "AssetAccess",
     "CanaryCase",
     "CaseId",
+    "ClusterState",
+    "ClusterRevisionRef",
     "ComponentKind",
     "CollectionError",
     "CollectionErrorCode",
@@ -166,7 +184,14 @@ def read_snapshot_content(
     "CommandVerifier",
     "ContentCaptureMode",
     "DockerCompose",
+    "DiagnosisResult",
+    "DiagnosisRun",
+    "DiagnosisError",
+    "DiagnosisErrorCode",
     "EditableFile",
+    "EvidenceAnchor",
+    "EvidenceAnchorKind",
+    "FailureCluster",
     "GitCommit",
     "Harness",
     "HarnessAsset",
@@ -187,6 +212,7 @@ def read_snapshot_content(
     "MineError",
     "MineErrorCode",
     "MiningPolicy",
+    "MechanismKey",
     "ObservationContent",
     "ObservationContentField",
     "ObservationContentHit",
@@ -199,20 +225,23 @@ def read_snapshot_content(
     "ProcessLimits",
     "PythonEntrypoint",
     "PythonLoop",
+    "PythonDiagnoser",
     "PythonVerifier",
     "RunErrorCode",
     "RunResult",
     "RunStatus",
     "ScoreName",
     "Sha256Digest",
     "ServiceName",
+    "Severity",
     "SecretEnvironmentVariable",
     "SnapshotContentReference",
     "Subagent",
     "Tool",
     "TraceWindow",
     "TracePartition",
     "TraceQualityThreshold",
+    "TraceDiagnosis",
     "VerifierResult",
     "VerifierVerdict",
     "WorkspaceFile",

src/ofw/_diagnosis_runner.py

@@ -0,0 +1,46 @@
+"""Timed child-process entrypoint for a file-backed trace diagnoser."""
+
+from __future__ import annotations
+
+import importlib
+import inspect
+import sys
+from collections.abc import Callable
+from typing import cast
+
+from pydantic import TypeAdapter, ValidationError
+
+from ofw.diagnosis import TraceDiagnosis
+from ofw.mine import TraceSnapshot
+
+_SNAPSHOT_ADAPTER: TypeAdapter[TraceSnapshot] = TypeAdapter(TraceSnapshot)
+_DIAGNOSIS_ADAPTER: TypeAdapter[TraceDiagnosis] = TypeAdapter(TraceDiagnosis)
+
+
+def main() -> int:
+    if len(sys.argv) != 3:
+        return 2
+    payload: str = sys.stdin.read()
+    try:
+        snapshot = _SNAPSHOT_ADAPTER.validate_json(payload)
+    except ValidationError:
+        return 2
+    module = importlib.import_module(sys.argv[1])
+    functions = tuple(
+        function
+        for name, function in inspect.getmembers(module, inspect.isfunction)
+        if name == sys.argv[2]
+    )
+    if len(functions) != 1:
+        return 2
+    function: Callable[[TraceSnapshot], TraceDiagnosis] = cast(
+        Callable[[TraceSnapshot], TraceDiagnosis],
+        functions[0],
+    )
+    diagnosis: TraceDiagnosis = function(snapshot)
+    sys.stdout.write(_DIAGNOSIS_ADAPTER.dump_json(diagnosis).decode())
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())