Implement credentialed drill packaging workplan

src/phase_memory/evaluation.py

@@ -10,8 +10,10 @@ from .contracts import graph_from_markitect
 from .models import Diagnostic, MemoryPath
 from .retrieval import activation_quality_report, select_event_path
 from .runtime import PhaseMemoryRuntime
+from .utils import stable_digest, utc_now_iso
 
 EVALUATION_REPORT_SCHEMA = "phase_memory.evaluation.threshold_report.v1"
+EVALUATION_TREND_SCHEMA = "phase_memory.evaluation.trend_artifact.v1"
 
 DEFAULT_THRESHOLDS = {
     "policy_denial_count": 1,
@@ -65,6 +67,54 @@ def evaluation_threshold_report(data: dict[str, Any], *, thresholds: dict[str, f
     }
 
 
+def evaluation_trend_artifact(
+    report: dict[str, Any],
+    *,
+    previous_report: dict[str, Any] | None = None,
+    run_metadata: dict[str, Any] | None = None,
+) -> dict[str, Any]:
+    run_metadata = {
+        "created_at": utc_now_iso(),
+        **dict(run_metadata or {}),
+    }
+    metrics = dict(report.get("metrics") or {})
+    thresholds = dict(report.get("thresholds") or {})
+    previous_metrics = dict((previous_report or {}).get("metrics") or {})
+    threshold_deltas = {
+        key: round(float(metrics.get(key) or 0) - float(threshold), 4)
+        for key, threshold in sorted(thresholds.items())
+    }
+    metric_deltas = {
+        key: round(float(value or 0) - float(previous_metrics.get(key) or 0), 4)
+        for key, value in sorted(metrics.items())
+        if key in previous_metrics
+    }
+    diagnostics = [dict(item) for item in report.get("diagnostics", ())]
+    for key, delta in metric_deltas.items():
+        if delta < 0:
+            diagnostics.append(
+                Diagnostic(
+                    "warn",
+                    "evaluation_metric_regressed",
+                    "Evaluation metric declined from the previous report.",
+                    key,
+                    {"delta": delta, "current": metrics.get(key), "previous": previous_metrics.get(key)},
+                ).to_dict()
+            )
+    artifact_id = f"evaluation-trend:{stable_digest([run_metadata, metrics, thresholds, previous_metrics])}"
+    return {
+        "schema_version": EVALUATION_TREND_SCHEMA,
+        "id": artifact_id,
+        "valid": not any(item.get("severity") == "error" for item in diagnostics),
+        "run": run_metadata,
+        "metrics": metrics,
+        "thresholds": thresholds,
+        "threshold_deltas": threshold_deltas,
+        "metric_deltas": metric_deltas,
+        "report": report,
+        "previous_report_id": (previous_report or {}).get("id", ""),
+        "diagnostics": diagnostics,
+    }
 def _policy_scenario(scenario: dict[str, Any]) -> dict[str, Any]:
     runtime = PhaseMemoryRuntime()
     response = runtime.plan_activation(