session-memory: Phase 4 Measure — baseline, effectiveness, trend (WP-0009)

Closes the loop. metrics.py: fleet metrics (infra-overhead share, error rate,
schema-thrash, token percentiles, success) + persisted baseline trend. effect.py:
before/after per-pattern effectiveness with an improved verdict per metric.
measure entrypoint with trend + --since effectiveness + JSON. Recorded pre-fix
baseline: 27 sessions, overhead median 11.7%, error rate 0.96, schema-thrash 8.
13 new tests; suite 139/139. Capture->Detect->Curate->Distribute->Measure complete.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

session_memory/measure/effect.py

@@ -0,0 +1,60 @@
+"""Before/after per-pattern effectiveness (PRD §6.5 FR-M1/FR-M2; T02).
+
+Given a change/pattern with an ``applied_at`` date, split sessions into *before*
+and *after* by their start time, aggregate each side, and diff the headline
+metrics — so we can say whether a distributed pattern (e.g. the Read-before-Edit
+reflex, or the State Hub skill) actually moved the numbers, and retire it if not.
+"""
+
+from __future__ import annotations
+
+from .metrics import aggregate
+
+# Metrics where a *lower* value after the change means improvement.
+_LOWER_IS_BETTER = {
+    "infra_overhead_share_median", "infra_overhead_share_p90", "error_rate",
+    "recurring_error_occurrences", "schema_thrash_sessions", "tokens_p50", "tokens_p90",
+}
+# Metrics where a *higher* value is improvement.
+_HIGHER_IS_BETTER = {"success_rate"}
+
+
+def split_by_date(digests: list[dict], applied_at: str) -> tuple[list[dict], list[dict]]:
+    """Partition digests into (before, after) by ``started_at`` vs ``applied_at``."""
+    before, after = [], []
+    for d in digests:
+        ts = d.get("started_at") or ""
+        (after if ts and ts >= applied_at else before).append(d)
+    return before, after
+
+
+def _delta(metric: str, before: float, after: float) -> dict:
+    change = round(after - before, 3)
+    if metric in _LOWER_IS_BETTER:
+        improved = change < 0
+    elif metric in _HIGHER_IS_BETTER:
+        improved = change > 0
+    else:
+        improved = None
+    return {"before": before, "after": after, "change": change, "improved": improved}
+
+
+def effectiveness(digests: list[dict], applied_at: str, *, label: str = "") -> dict:
+    """Compare fleet metrics after ``applied_at`` against the prior period."""
+    before, after = split_by_date(digests, applied_at)
+    b_agg, a_agg = aggregate(before), aggregate(after)
+    metrics = (_LOWER_IS_BETTER | _HIGHER_IS_BETTER)
+    deltas = {}
+    if before and after:
+        for m in metrics:
+            deltas[m] = _delta(m, b_agg.get(m, 0.0), a_agg.get(m, 0.0))
+    return {
+        "label": label,
+        "applied_at": applied_at,
+        "n_before": len(before),
+        "n_after": len(after),
+        "before": b_agg,
+        "after": a_agg,
+        "deltas": deltas,
+        "insufficient_data": not (before and after),
+    }