Archive probability snapshots and wire them into training

tests/test_probability_snapshot_archive.py

@@ -0,0 +1,117 @@
+import json
+from pathlib import Path
+
+from src.analysis.probability_snapshot_archive import append_probability_snapshot
+
+
+def test_append_probability_snapshot_writes_jsonl(tmp_path: Path):
+    archive_path = tmp_path / "probability_training_snapshots.jsonl"
+
+    append_probability_snapshot(
+        city_name="ankara",
+        local_date="2026-03-20",
+        observation_time="2026-03-20T12:00:00+03:00",
+        temp_symbol="°C",
+        raw_mu=15.2,
+        raw_sigma=1.2,
+        deb_prediction=15.4,
+        ens_data={"p10": 14.8, "median": 15.8, "p90": 17.9},
+        current_forecasts={"ECMWF": 15.8, "GFS": 14.1},
+        max_so_far=15.0,
+        peak_status="before",
+        probabilities=[{"value": 15, "probability": 0.552}],
+        shadow_probabilities=[{"value": 15, "probability": 0.324}],
+        calibration_summary={
+            "engine": "legacy",
+            "mode": "emos_shadow",
+            "calibration_version": "emos-test",
+            "calibration_source": "artifacts/probability_calibration/default.json",
+            "calibrated_mu": 15.1,
+            "calibrated_sigma": 1.25,
+        },
+        archive_path=str(archive_path),
+    )
+
+    lines = archive_path.read_text(encoding="utf-8").strip().splitlines()
+    assert len(lines) == 1
+    payload = json.loads(lines[0])
+    assert payload["city"] == "ankara"
+    assert payload["date"] == "2026-03-20"
+    assert payload["raw_mu"] == 15.2
+    assert payload["ensemble"]["median"] == 15.8
+    assert payload["prob_snapshot"][0]["v"] == 15
+    assert payload["shadow_prob_snapshot"][0]["v"] == 15
+    assert payload["calibration_version"] == "emos-test"
+
+
+def test_append_probability_snapshot_skips_near_duplicate(tmp_path: Path):
+    archive_path = tmp_path / "probability_training_snapshots.jsonl"
+    kwargs = dict(
+        city_name="ankara",
+        local_date="2026-03-20",
+        observation_time="2026-03-20T12:00:00+03:00",
+        temp_symbol="°C",
+        raw_mu=15.2,
+        raw_sigma=1.2,
+        deb_prediction=15.4,
+        ens_data={"p10": 14.8, "median": 15.8, "p90": 17.9},
+        current_forecasts={"ECMWF": 15.8, "GFS": 14.1},
+        max_so_far=15.0,
+        peak_status="before",
+        probabilities=[{"value": 15, "probability": 0.552}],
+        shadow_probabilities=[{"value": 15, "probability": 0.324}],
+        calibration_summary={
+            "engine": "legacy",
+            "mode": "emos_shadow",
+            "calibration_version": "emos-test",
+            "calibration_source": "artifacts/probability_calibration/default.json",
+            "calibrated_mu": 15.1,
+            "calibrated_sigma": 1.25,
+        },
+        archive_path=str(archive_path),
+    )
+
+    append_probability_snapshot(**kwargs)
+    append_probability_snapshot(**kwargs)
+
+    lines = archive_path.read_text(encoding="utf-8").strip().splitlines()
+    assert len(lines) == 1
+
+
+def test_append_probability_snapshot_writes_on_bucket_change(tmp_path: Path):
+    archive_path = tmp_path / "probability_training_snapshots.jsonl"
+    base_kwargs = dict(
+        city_name="ankara",
+        local_date="2026-03-20",
+        observation_time="2026-03-20T12:00:00+03:00",
+        temp_symbol="°C",
+        raw_mu=15.2,
+        raw_sigma=1.2,
+        deb_prediction=15.4,
+        ens_data={"p10": 14.8, "median": 15.8, "p90": 17.9},
+        current_forecasts={"ECMWF": 15.8, "GFS": 14.1},
+        max_so_far=15.0,
+        peak_status="before",
+        shadow_probabilities=[{"value": 15, "probability": 0.324}],
+        calibration_summary={
+            "engine": "legacy",
+            "mode": "emos_shadow",
+            "calibration_version": "emos-test",
+            "calibration_source": "artifacts/probability_calibration/default.json",
+            "calibrated_mu": 15.1,
+            "calibrated_sigma": 1.25,
+        },
+        archive_path=str(archive_path),
+    )
+
+    append_probability_snapshot(
+        probabilities=[{"value": 15, "probability": 0.552}],
+        **base_kwargs,
+    )
+    append_probability_snapshot(
+        probabilities=[{"value": 16, "probability": 0.552}],
+        **base_kwargs,
+    )
+
+    lines = archive_path.read_text(encoding="utf-8").strip().splitlines()
+    assert len(lines) == 2

tests/test_probability_training_dataset.py

@@ -0,0 +1,45 @@
+from scripts.fit_probability_calibration import _extract_samples
+
+
+def test_extract_samples_prefers_snapshot_rows_for_same_city_day():
+    history = {
+        "ankara": {
+            "2026-03-19": {
+                "actual_high": 11.0,
+                "mu": 10.8,
+                "deb_prediction": 10.9,
+                "forecasts": {"ECMWF": 10.5, "GFS": 11.2},
+                "probability_features": {
+                    "ens_median": 10.7,
+                    "ensemble_spread": 0.8,
+                    "peak_status": "before",
+                },
+            }
+        }
+    }
+    snapshot_rows = [
+        {
+            "city": "ankara",
+            "date": "2026-03-19",
+            "timestamp": "2026-03-19T12:00:00+03:00",
+            "raw_mu": 11.2,
+            "raw_sigma": 1.1,
+            "deb_prediction": 11.0,
+            "ensemble": {"p10": 10.0, "median": 11.1, "p90": 12.2},
+            "multi_model": {"ECMWF": 10.5, "GFS": 11.2},
+            "max_so_far": 10.9,
+            "peak_status": "in_window",
+        }
+    ]
+
+    samples, filled = _extract_samples(
+        history,
+        settlement_history={},
+        snapshot_rows=snapshot_rows,
+    )
+
+    assert filled == 0
+    assert len(samples) == 1
+    assert samples[0]["sample_source"] == "snapshot"
+    assert samples[0]["raw_mu"] == 11.2
+    assert samples[0]["peak_flag"] == 0.5