auto-sync 2026-07-04T03:59:09Z workspace (part 5)

workspace/scripts/slurm/train_ctt_feature_proxy.sbatch

@@ -60,12 +60,15 @@ VAL_OUT="runs/${NAME_PREFIX}_seed${SEED}_val_proxy"
   --k "${CTT_PROXY_K:-16}" \
   --max-target-charts "${CTT_MAX_TARGET_CHARTS:-69}" \
   --neighbors "${CTT_PROXY_NEIGHBORS:-8}" \
-  --thresholds "${CTT_PROXY_THRESHOLDS:-0.20,0.40}"
+  --thresholds "${CTT_PROXY_THRESHOLDS:-0.20,0.40}" \
+  --no-markdown-report
 
 "$PYTHON" scripts/build_ctt_proxy_comparison.py \
-  --out-dir runs/ctt_val_proxy_comparison
+  --out-dir runs/ctt_val_proxy_comparison \
+  --no-markdown-report
 
 "$PYTHON" scripts/summarize_ctt_runs.py \
   --run-root runs \
   --out-csv runs/summary_ctt.csv \
-  --out-md runs/summary_ctt.md
+  --out-md runs/summary_ctt.md \
+  --no-markdown-report

workspace/scripts/summarize_ctt_runs.py

@@ -39,6 +39,11 @@ def main(argv: list[str] | None = None) -> int:
     parser.add_argument("--run-root", type=Path, default=Path("runs"))
     parser.add_argument("--out-csv", type=Path, default=Path("runs/summary_ctt.csv"))
     parser.add_argument("--out-md", type=Path, default=Path("runs/summary_ctt.md"))
+    parser.add_argument(
+        "--no-markdown-report",
+        action="store_true",
+        help="Do not write --out-md; the persistent prose summary lives in README.md.",
+    )
     args = parser.parse_args(argv)
 
     rows = []
@@ -99,8 +104,14 @@ def main(argv: list[str] | None = None) -> int:
         writer.writeheader()
         for row in rows:
             writer.writerow({column: row.get(column, "n/a") for column in COLUMNS})
-    args.out_md.write_text(_markdown(rows) + "\n")
-    print(json.dumps({"csv": str(args.out_csv), "md": str(args.out_md), "rows": len(rows)}, indent=2))
+    if args.no_markdown_report:
+        if args.out_md.exists():
+            args.out_md.unlink()
+        md_path: str | None = None
+    else:
+        args.out_md.write_text(_markdown(rows) + "\n")
+        md_path = str(args.out_md)
+    print(json.dumps({"csv": str(args.out_csv), "md": md_path, "rows": len(rows)}, indent=2))
     return 0
 
 

workspace/tests/test_dominance_selector.py

@@ -21,8 +21,10 @@ from scripts.eval_learned_dominance_selector import (
     _feature_names,
     _fit_select_ridge,
     _pairwise_calibration_summary,
+    _score_shape_matrix,
     _source_evidence_feature,
     _target_value,
+    _write_report_artifact,
 )
 from scripts.eval_nonlinear_dominance_selector import _split_rows, _subset_dataset
 from scripts.eval_nonlinear_dominance_selector import _write_provenance as _write_nonlinear_provenance
@@ -443,6 +445,61 @@ def test_learned_dominance_tangent_features_and_success_weighted_target() -> Non
     assert source_evidence[1] == pytest.approx(np.log1p(2))
     assert source_evidence[-1] == 1.0
     assert _target_value("success_weighted_margin", utility_margin=0.2, candidate_success=1.0) == 1.2
+    assert _target_value(
+        "success_weighted_margin",
+        utility_margin=0.2,
+        candidate_success=1.0,
+        success_bonus=2.5,
+    ) == 2.7
+
+
+def test_score_shape_feature_set_and_markdown_suppression(tmp_path) -> None:
+    score_shape = _score_shape_matrix([3.0, 1.0, 2.0])
+    feature = _candidate_feature(
+        score=3.0,
+        base_score=0.25,
+        score_mean=2.0,
+        score_std=1.0,
+        candidate_index=0,
+        candidate_type="ctt_transport_rank0",
+        tangent=np.zeros(21, dtype=float),
+        num_candidates=3,
+        feature_set="score_context_chart_compat",
+        context={
+            "target_task_id": "PickCube-v1",
+            "source_task_id": "PickCube-v1",
+            "instruction": "Pick up the cube.",
+        },
+        chart_compat=np.ones(22, dtype=float),
+        score_shape=score_shape[0],
+    )
+
+    assert score_shape.shape == (3, 8)
+    assert score_shape[0, 0] == 0.0
+    assert score_shape[0, 2] == 0.0
+    assert score_shape[0, 3] == 1.0
+    assert score_shape[1, 0] == 1.0
+    assert len(feature) == len(_feature_names("score_context_chart_compat"))
+    assert len(feature) == 10 + 8 + 27 + 22
+
+    metrics = {
+        "num_calibration_rows": 1,
+        "num_eval_rows": 1,
+        "selected_lambda": 0.0,
+        "tau": 0.0,
+        "threshold_scope": "global",
+        "fit_objective": "pointwise",
+        "feature_set": "basic",
+        "target": "utility_margin",
+        "eval_summary": {},
+        "calibration_summary": {},
+    }
+    report_path = tmp_path / "report.md"
+    report_path.write_text("old\n")
+    _write_report_artifact(tmp_path, metrics, no_markdown_report=True)
+    assert not report_path.exists()
+    _write_report_artifact(tmp_path, metrics, no_markdown_report=False)
+    assert report_path.exists()
 
 
 def test_chart_compat_feature_is_deployment_visible_and_fixed_width() -> None: