chore(teaching): MetricSet dataclass (ADR-0142 debt #3) (#231)

teaching/__init__.py

@@ -14,6 +14,7 @@ approval before they touch the vocabulary manifold.
 
 from teaching.correction import CorrectionCandidate, extract_correction
 from teaching.epistemic import ADMISSIBLE_AS_EVIDENCE, EpistemicStatus, parse_status
+from teaching.metric_set import MetricSet
 from teaching.review import ReviewedTeachingExample, ReviewOutcome, review_correction
 from teaching.store import TeachingStore, PackMutationProposal
 
@@ -21,6 +22,7 @@ __all__ = [
     "ADMISSIBLE_AS_EVIDENCE",
     "CorrectionCandidate",
     "EpistemicStatus",
+    "MetricSet",
     "PackMutationProposal",
     "ReviewedTeachingExample",
     "ReviewOutcome",

teaching/metric_set.py

@@ -0,0 +1,14 @@
+"""Versioned metric carrier for replay evidence gates."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class MetricSet:
+    version: int
+    metrics: tuple[str, ...]
+
+
+__all__ = ["MetricSet"]

teaching/replay.py

@@ -26,16 +26,20 @@ from pathlib import Path
 from typing import Any, Iterator
 
 from chat import teaching_grounding as _tg
+from teaching.metric_set import MetricSet
 from teaching.proposals import ReplayEvidence
 
 
 # Metrics watched for regression.  Any metric whose candidate value
 # is strictly less than the baseline value counts as a regression.
-_WATCHED_METRICS: tuple[str, ...] = (
+_WATCHED_METRICS = MetricSet(
-    "intent_accuracy",
+    version=1,
-    "surface_groundedness",
+    metrics=(
-    "term_capture_rate",
+        "intent_accuracy",
-    "versor_closure_rate",
+        "surface_groundedness",
+        "term_capture_rate",
+        "versor_closure_rate",
+    ),
 )
 
 
@@ -83,7 +87,7 @@ def _run_cognition_public() -> dict[str, float]:
     lane = get_lane("cognition")
     result = run_lane(lane, version="v1", split="public")
     out: dict[str, float] = {}
-    for k in _WATCHED_METRICS:
+    for k in _WATCHED_METRICS.metrics:
         v = result.metrics.get(k)
         if isinstance(v, (int, float)):
             out[k] = float(v)
@@ -143,7 +147,7 @@ def run_replay_equivalence(chain: dict[str, Any]) -> ReplayEvidence:
             candidate = _run_cognition_public()
 
     regressed: list[str] = []
-    for metric in _WATCHED_METRICS:
+    for metric in _WATCHED_METRICS.metrics:
         b = baseline.get(metric)
         c = candidate.get(metric)
         if b is None or c is None: