补齐 Autocal 测点质量与独立 trial 验证

按已确认窗口比较重复性、加载卸载差异和零载恢复，并逐 trial 留出验证。55 项定向测试、已录视频复核闭环及代码文档检查通过。

apps/autocal/tests/test_session_review.py

@@ -210,3 +210,114 @@ def test_baseline_image_matches_first_reviewed_zero_frame(session):
     assert baseline["status"] == "saved" and baseline["segment_frame"] == 1
     image = cv2.imread(str(saved.parent / baseline["image"]))
     assert 10 < image.mean() < 30
+
+
+@pytest.fixture
+def repeated_session(session):
+    import shutil
+
+    source = session / "trials/trial-001"
+    destination = session / "trials/trial-004"
+    shutil.copytree(source, destination)
+    metadata = json.loads((destination / "session.json").read_text(encoding="utf-8"))
+    metadata["trial_id"] = "trial-004"
+    write_json(destination / "session.json", metadata)
+    with (destination / "forces.csv").open(encoding="utf-8") as handle:
+        forces = list(csv.DictReader(handle))
+    for row in forces:
+        row["FN_raw"] = float(row["FN_raw"]) + 2
+    _csv(destination / "forces.csv", forces)
+    manifest = json.loads((session / "session.json").read_text(encoding="utf-8"))
+    manifest["trials"].append({"trial_id": "trial-004", "site_id": "site-001",
+                               "status": "complete", "directory": "trials/trial-004"})
+    write_json(session / "session.json", manifest)
+    result = review_session(session)
+    selections = {f"{w['trial_id']}:{w['step_index']}": {
+        "t_start_host_mono_ns": w["t_start_host_mono_ns"],
+        "t_end_host_mono_ns": w["t_end_host_mono_ns"], "stability_verified": True,
+    } for site in result["sites"] for w in site["windows"]}
+    return session, selections
+
+
+def test_quality_uses_distinct_trials_and_never_trains_on_held_out_data(repeated_session):
+    import math
+    from orisys_autocal.session_review import load_review
+
+    directory, selections = repeated_session
+    result = review_session(directory, selections=selections)
+    quality = result["sites"][0]["quality"]
+    assert quality["reference_alignment_possible"] is False
+    assert all(row["trials"] == 2 and row["raw_range"] == 2
+               and row["raw_std_between_trials"] == pytest.approx(math.sqrt(2))
+               for row in quality["repeatability"])
+    assert len(quality["hysteresis"]) == 2
+    assert all(row["unload_minus_load_raw"] == 1 for row in quality["hysteresis"])
+    assert all(row["final_minus_initial_raw"] == 0 for row in quality["zero_return"])
+    validation = quality["validation"]
+    assert validation["evaluated_windows"] == 10
+    assert validation["rmse_n"] == pytest.approx(math.sqrt(1.05))
+    assert validation["mae_n"] == pytest.approx(1)
+    assert validation["max_abs_error_n"] == pytest.approx(1.5)
+    assert validation["outside_training_raw_range"] == 3
+    for fold in validation["folds"]:
+        assert fold["held_out_trial"] not in fold["training_trials"]
+        assert len(fold["training_trials"]) == 1
+    assert validation["folds"][0]["fit"]["offset_n"] == pytest.approx(-1.5)
+    assert validation["folds"][1]["fit"]["offset_n"] == pytest.approx(-.5)
+    second = result["sites"][1]["quality"]
+    assert second["validation"]["rmse_n"] is None
+    assert all(row["raw_range"] is None for row in second["repeatability"])
+    saved = save_review(result)
+    assert load_review(directory, saved)["sites"][0]["quality"] == quality
+
+
+def test_quality_ignores_unconfirmed_data_and_unmatched_loads(repeated_session):
+    directory, selections = repeated_session
+    result = review_session(directory)
+    assert result["sites"][0]["quality"]["repeatability"] == []
+    assert result["sites"][0]["quality"]["validation"]["rmse_n"] is None
+    # Keep only one confirmed loading level in the other trial: no usable fit.
+    for key in list(selections):
+        if key.startswith("trial-004:") and key != "trial-004:1":
+            selections[key]["stability_verified"] = False
+    result = review_session(directory, selections=selections)
+    quality = result["sites"][0]["quality"]
+    assert quality["validation"]["folds"][0]["status"] == "insufficient_training"
+    assert len(quality["hysteresis"]) == 1
+    path = directory / "trials/trial-001/reference.csv"
+    with path.open(encoding="utf-8") as handle:
+        rows = list(csv.DictReader(handle))
+    rows[3]["F_ref_normal_N"] = "5.01"
+    _csv(path, rows)
+    quality = review_session(directory, selections=selections)["sites"][0]["quality"]
+    assert quality["hysteresis"] == []
+
+
+def test_quality_dialog_is_reachable_and_shows_validation_tables(qt_app, repeated_session, monkeypatch, tmp_path):
+    from PySide6.QtWidgets import QTableWidget
+    from orisys_autocal.ui.session_review_dialog import SessionReviewDialog
+    from orisys_autocal.ui.quality_review_dialog import QualityReviewDialog
+
+    directory, selections = repeated_session
+    review = SessionReviewDialog(directory)
+    review.review = review_session(directory, selections=selections)
+    review._show_site()
+    opened = []
+    monkeypatch.setattr(QualityReviewDialog, "exec", lambda dialog: opened.append(dialog))
+    review.quality_button.click()
+    assert len(opened) == 1
+    dialog = opened[0]
+    assert dialog.tabs.count() == 4
+    repeat = dialog.findChild(QTableWidget, "quality_repeatability")
+    validation = dialog.findChild(QTableWidget, "quality_validation")
+    assert repeat.rowCount() == 5 and validation.rowCount() == 10
+    assert validation.item(0, 0).text() == "trial-001"
+    assert validation.item(0, 1).text() == "trial-004"
+    dialog.show()
+    qt_app.processEvents()
+    assert dialog.grab().save(str(tmp_path / "repeatability.png"))
+    dialog.tabs.setCurrentIndex(3)
+    qt_app.processEvents()
+    assert dialog.grab().save(str(tmp_path / "validation.png"))
+    dialog.close()
+    review.close()