Test nonlinear chart-compat provenance

workspace/tests/test_dominance_selector.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import json
 from types import SimpleNamespace
 
 import numpy as np
@@ -18,6 +19,7 @@ from scripts.eval_learned_dominance_selector import (
     _target_value,
 )
 from scripts.eval_nonlinear_dominance_selector import _split_rows, _subset_dataset
+from scripts.eval_nonlinear_dominance_selector import _write_provenance as _write_nonlinear_provenance
 
 
 def test_conformal_quantile_and_fallback_selection() -> None:
@@ -412,3 +414,44 @@ def test_nonlinear_dominance_split_is_row_disjoint() -> None:
     subset = _subset_dataset(dataset, select_rows)
     assert subset["num_rows"] == len(select_rows)
     assert len(subset["samples"]) == 2 * len(select_rows)
+
+
+def test_nonlinear_chart_compat_provenance_records_default_source_index(tmp_path) -> None:
+    calibration_input = tmp_path / "calibration.json"
+    eval_input = tmp_path / "eval.json"
+    train_index = tmp_path / "train_index.json"
+    test_index = tmp_path / "test_index.json"
+    calibration_input.write_text("{}\n")
+    eval_input.write_text("{}\n")
+    train_index.write_text(
+        '{"split":"train","content_hash":"train-content","split_hash":"train-split","retrieval_index_allowed":true}\n'
+    )
+    test_index.write_text(
+        '{"split":"test","content_hash":"test-content","split_hash":"test-split","retrieval_index_allowed":false}\n'
+    )
+    args = SimpleNamespace(
+        calibration_input=calibration_input,
+        calibration_target_index=train_index,
+        eval_input=eval_input,
+        eval_target_index=test_index,
+        source_index=None,
+        checkpoint_template="runs/ctt_residual_base_context_obs_seed{seed}/model.pt",
+        out_dir=tmp_path / "out",
+        k=16,
+        feature_set="chart_compat",
+        selector_chart_feature_mode="base_context_obs",
+        target="utility_margin",
+        model_types="hgb_regressor",
+        selection_frac=0.35,
+        seed=0,
+        bootstrap_samples=10,
+    )
+
+    args.out_dir.mkdir()
+    _write_nonlinear_provenance(args.out_dir, args)
+    data_hash = json.loads((args.out_dir / "data_hash.txt").read_text())
+    split_hash = json.loads((args.out_dir / "split_hash.txt").read_text())
+
+    assert "source_index" in data_hash
+    assert split_hash["source_index"]["split"] == "train"
+    assert split_hash["source_index"]["retrieval_index_allowed"] is True