Add SciForge visualization pipeline

agents/figure_agent.py

@@ -129,7 +129,14 @@ def render_metric_figure_artifacts(
     )
     used_fallback = False
     try:
-        subprocess.run([RUNTIME_PYTHON, str(script)], check=True, cwd=str(out_svg.parent), timeout=120)
+        subprocess.run(
+            [RUNTIME_PYTHON, str(script)],
+            check=True,
+            cwd=str(out_svg.parent),
+            timeout=120,
+            capture_output=True,
+            text=True,
+        )
     except (subprocess.CalledProcessError, FileNotFoundError, subprocess.TimeoutExpired):
         used_fallback = True
         out_svg.write_text(

agents/validation_loop.py

@@ -22,6 +22,7 @@
 
 from agents import codex_executor
 from agents import experiment_supervisor
+from agents import visualization_agent
 from agents.workspace_layout import ensure_run_workspace, plan_file_path, promote_canonical_run, write_latest_status
 from contracts import DeepInsightSpec, ExperimentIterationPacket, ExperimentSpec
 from config import (
@@ -511,6 +512,77 @@ def _record_artifact(
     )
 
 
+def _generate_validation_figures(
+    run_id: int,
+    workdir: Path,
+    *,
+    insight: dict,
+    metric_name: str,
+    baseline_metric_value: float | None,
+    best_metric_value: float | None = None,
+    verdict: str | None = None,
+    summary_path: Path | None = None,
+) -> list[dict]:
+    """Generate validation-loop figure artifacts for a completed run."""
+    try:
+        bundle = visualization_agent.generate_visualization_bundle(
+            run_id=run_id,
+            workdir=workdir,
+            insight=insight,
+            metric_name=metric_name,
+            baseline_metric_value=baseline_metric_value,
+            best_metric_value=best_metric_value,
+            verdict=verdict,
+            summary_path=summary_path,
+        )
+    except Exception as exc:
+        print(f"[LOOP] Figure generation skipped for run {run_id}: {exc}", flush=True)
+        return []
+
+    assets = [dict(asset) for asset in bundle.get("assets") or [] if isinstance(asset, dict)]
+    for asset in assets:
+        path = Path(str(asset.get("path") or ""))
+        if not path.exists():
+            continue
+        asset_kind = str(asset.get("asset_kind") or "")
+        artifact_type = "plot" if asset_kind in {"svg", "pdf", "png", "jpg", "jpeg"} else "source_data"
+        try:
+            _record_artifact(
+                run_id,
+                artifact_type,
+                path,
+                metric_key=asset.get("metric_name") or metric_name,
+                metadata={
+                    "figure_id": asset.get("figure_id"),
+                    "figure_kind": asset.get("figure_kind"),
+                    "asset_kind": asset_kind,
+                    "caption": asset.get("caption"),
+                    "source": asset.get("source"),
+                    **(asset.get("metadata") if isinstance(asset.get("metadata"), dict) else {}),
+                },
+            )
+        except Exception as exc:
+            print(f"[LOOP] Figure artifact registration skipped for {path}: {exc}", flush=True)
+
+    for key, contract_type in (
+        ("manifest_path", "ValidationFigureManifest"),
+        ("references_path", "ValidationFigureReferences"),
+    ):
+        raw_path = str(bundle.get(key) or "").strip()
+        if raw_path and Path(raw_path).exists():
+            try:
+                _record_artifact(
+                    run_id,
+                    "source_data",
+                    Path(raw_path),
+                    metric_key=metric_name,
+                    metadata={"contract_type": contract_type},
+                )
+            except Exception as exc:
+                print(f"[LOOP] Figure manifest registration skipped for {raw_path}: {exc}", flush=True)
+    return assets
+
+
 def _read_experiment_spec(
     run: dict,
     insight: dict,
@@ -1234,24 +1306,32 @@ def run_validation_loop(run_id: int, execution_context: dict | None = None) -> d
     promote_canonical_run(insight_id, run_id, insight=insight)
 
     summary_path = workdir / "results" / "validation_summary.json"
-    summary_path.write_text(
-        json.dumps(
-            {
-                "run_id": run_id,
-                "verdict": verdict,
-                "baseline": baseline,
-                "best_value": best_value,
-                "effect_size": effect,
-                "effect_pct": effect_pct,
-                "iterations_total": iter_num,
-                "iterations_kept": total_kept,
-                "environment_report": environment_report,
-                "stop_reason": stop_reason,
-            },
-            indent=2,
-        ),
-        encoding="utf-8",
+    summary_payload = {
+        "run_id": run_id,
+        "verdict": verdict,
+        "baseline": baseline,
+        "best_value": best_value,
+        "effect_size": effect,
+        "effect_pct": effect_pct,
+        "iterations_total": iter_num,
+        "iterations_kept": total_kept,
+        "environment_report": environment_report,
+        "stop_reason": stop_reason,
+    }
+    summary_path.write_text(json.dumps(summary_payload, indent=2), encoding="utf-8")
+    figure_assets = _generate_validation_figures(
+        run_id,
+        workdir,
+        insight=insight,
+        metric_name=metric_name,
+        baseline_metric_value=baseline,
+        best_metric_value=best_value,
+        verdict=verdict,
+        summary_path=summary_path,
     )
+    if figure_assets:
+        summary_payload["figure_artifacts"] = figure_assets
+        summary_path.write_text(json.dumps(summary_payload, indent=2), encoding="utf-8")
     _record_artifact(
         run_id,
         "source_data",
@@ -1274,6 +1354,7 @@ def run_validation_loop(run_id: int, execution_context: dict | None = None) -> d
             "iterations_total": iter_num,
             "iterations_kept": total_kept,
             "summary_path": str(summary_path),
+            "figure_artifacts": figure_assets,
         },
         run_id=run_id,
         insight=insight,
@@ -1294,4 +1375,5 @@ def run_validation_loop(run_id: int, execution_context: dict | None = None) -> d
         "total_seconds": total_time,
         "environment_report": environment_report,
         "stop_reason": stop_reason,
+        "figure_artifacts": figure_assets,
     }