Show latest Krea2 evidence in tuning report

docs/sxcp-eval-loop.md

@@ -51,7 +51,9 @@ python tools/krea2_tuning_report.py
 ```
 
 The report includes atlas references plus prompt cues and avoid cues for the
-next fixed-seed test candidate.
+next fixed-seed test candidate. It also shows the latest durable evidence for
+variants that already have fixed-seed results, including the evidence id, seed,
+decision, candidate prompt summary, and observation.
 
 ## Optional Command Hook
 

krea2_tuning_report.py

@@ -23,6 +23,23 @@ def _coverage_state(status: str, accepted_count: int) -> str:
     return "tracked"
 
 
+def _latest_evidence(entries: list[dict[str, Any]], *, result: str | None = None) -> dict[str, Any]:
+    filtered = [entry for entry in entries if result is None or entry.get("result") == result]
+    if not filtered:
+        return {}
+    entry = filtered[-1]
+    return {
+        "id": entry.get("id") or "",
+        "seed": entry.get("seed"),
+        "result": entry.get("result") or "",
+        "decision": entry.get("decision") or "",
+        "baseline_prompt_summary": entry.get("baseline_prompt_summary") or "",
+        "candidate_prompt_summary": entry.get("candidate_prompt_summary") or "",
+        "observation": entry.get("observation") or "",
+        "commit": entry.get("commit") or "",
+    }
+
+
 def coverage_rows() -> list[dict[str, Any]]:
     rows: list[dict[str, Any]] = []
     for variant in krea2_pose_variant_catalog.variants():
@@ -39,6 +56,8 @@ def coverage_rows() -> list[dict[str, Any]]:
                 "coverage_state": _coverage_state(status, len(accepted)),
                 "accepted_evidence_count": len(accepted),
                 "total_evidence_count": len(evidence),
+                "latest_evidence": _latest_evidence(evidence),
+                "latest_accepted_evidence": _latest_evidence(evidence, result="accepted"),
                 "reference_count": len(variant.get("reference_images") or []),
                 "guide_section": (variant.get("evidence") or {}).get("guide_section", ""),
             }
@@ -212,6 +231,21 @@ def markdown_report(atlas_root: str | Path | None = None) -> str:
         lines.append(
             f"| {row['key']} | {row['status']} | {row['accepted_evidence_count']}/{row['total_evidence_count']} | {row['coverage_state']} |"
         )
+    evidence_rows = [row for row in coverage_rows() if row.get("latest_evidence")]
+    if evidence_rows:
+        lines.extend(["", "## Latest Evidence", ""])
+        for row in evidence_rows:
+            evidence = row.get("latest_evidence") or {}
+            seed = evidence.get("seed")
+            seed_text = f"seed {seed}" if isinstance(seed, int) else "seed unknown"
+            commit = evidence.get("commit") or "uncommitted"
+            lines.append(
+                f"- {row['key']}: {evidence.get('id') or 'unnamed'} ({evidence.get('result') or 'unknown'}, {seed_text}, {evidence.get('decision') or 'unknown'}, commit {commit})"
+            )
+            if evidence.get("candidate_prompt_summary"):
+                lines.append(f"  Candidate: {evidence['candidate_prompt_summary']}")
+            if evidence.get("observation"):
+                lines.append(f"  Observation: {evidence['observation']}")
     summary = coverage_summary()
     if summary["next_test_candidates"]:
         lines.extend(

tools/prompt_smoke.py

@@ -6998,6 +6998,12 @@ def smoke_krea2_tuning_report_policy() -> None:
     boobjob = by_key.get("pov_boobjob_upright_cleavage") or {}
     _expect(boobjob.get("coverage_state") == "proven_with_evidence", "Boobjob report should be proven with evidence")
     _expect(boobjob.get("accepted_evidence_count", 0) >= 1, "Boobjob report lost accepted evidence count")
+    boobjob_latest = boobjob.get("latest_evidence") or {}
+    _expect(boobjob_latest.get("id") == "boobjob-7302-upright-cleavage", "Boobjob report lost latest evidence id")
+    _expect(boobjob_latest.get("seed") == 7302, "Boobjob report lost latest fixed seed")
+    _expect(boobjob_latest.get("result") == "accepted", "Boobjob report lost latest evidence result")
+    _expect(boobjob_latest.get("decision") == "generator_patch", "Boobjob report lost latest evidence decision")
+    _expect("upright frontal boobjob geometry" in str(boobjob_latest.get("candidate_prompt_summary") or ""), "Boobjob report lost latest candidate summary")
     ballsucking = by_key.get("pov_ballsucking_low_head") or {}
     _expect(ballsucking.get("coverage_state") == "needs_fixed_seed_tests", "Ballsucking report should need fixed-seed tests")
     _expect(ballsucking.get("accepted_evidence_count") == 0, "Ballsucking report should not have accepted evidence yet")
@@ -7293,6 +7299,11 @@ def smoke_krea2_tuning_report_policy() -> None:
         _expect("custom_pose" in atlas_markdown, "Krea2 tuning report markdown lost unmapped atlas folder")
         _expect("pov_custom_pose_candidate" in atlas_markdown, "Krea2 tuning report markdown lost suggested gap key")
     markdown = krea2_tuning_report.markdown_report()
+    _expect("## Latest Evidence" in markdown, "Krea2 tuning report markdown lost latest evidence section")
+    _expect("boobjob-7302-upright-cleavage" in markdown, "Krea2 tuning report markdown lost boobjob evidence id")
+    _expect("seed 7302" in markdown, "Krea2 tuning report markdown lost evidence seed")
+    _expect("generator_patch" in markdown, "Krea2 tuning report markdown lost evidence decision")
+    _expect("upright frontal boobjob geometry" in markdown, "Krea2 tuning report markdown lost evidence prompt summary")
     _expect("pov_ballsucking_low_head" in markdown, "Krea2 tuning report markdown lost candidate variant")
     _expect("pov_footjob_frontal_sole_stroke" in markdown, "Krea2 tuning report markdown lost footjob candidate variant")
     _expect("pov_fingering_reclined_open_thighs" in markdown, "Krea2 tuning report markdown lost fingering candidate variant")