Fuel Detective Summary
================================================================================
Generated UTC: 2026-03-27T16:08:48.893257+00:00
Data source: UK fuel station full-history dataset
Stations analyzed: 7,347
Unique station IDs: 7,347
Duplicate station rows in output: 0
Median E10 (ppl): 149.900
Median diesel (ppl): 179.900
Average staleness (days): 78.887
Median expected days to change: 14.95
High change probability rows (>0.70): 1,017
High anomaly rows: 8
Diesel cheaper than petrol rows: 7
Stale rows (>60 days): 6,399
Fill-now opportunities: 16
Fill-soon opportunities: 128

Market intelligence
--------------------------------------------------------------------------------
Brand stats rows: 37
Postcode sector stats rows: 1,282
Market regime rows: 1,282
Market regime breakdown:
Mixed: 520 (40.6%)
Supermarket Dominated: 461 (36.0%)
Fragmented: 113 (8.8%)
Competitive: 99 (7.7%)
Local Monopoly: 89 (6.9%)
Cheapest positioning brand: COSTCO WHOLESALE (score=-41.408)
Most expensive positioning brand: WELCOME BREAK (score=60.500)
Largest postcode sector sample: CR0 (stations=14)

Model diagnostics
--------------------------------------------------------------------------------
Target mode: price_change
Target horizon (days): 7.0
Target positive rate: 0.6482
Class balance: class_0=5113, class_1=9422
Training rows: 1,145
Validation rows: 2,907
Validation ROC-AUC (last fold): 0.7218
Walk-forward ROC-AUC mean: 0.7259
Walk-forward ROC-AUC std: 0.0239
Validation Brier score (last fold): 0.2224
Walk-forward Brier mean: 0.2103
Walk-forward Brier std: 0.0180

Walk-forward fold metrics
--------------------------------------------------------------------------------
fold_1: train_rows=618 validation_rows=2907 roc_auc=0.7570 brier=0.1850
fold_2: train_rows=848 validation_rows=2907 roc_auc=0.6990 brier=0.2237
fold_3: train_rows=1145 validation_rows=2907 roc_auc=0.7218 brier=0.2224

Classification report
--------------------------------------------------------------------------------
                      precision    recall  f1-score   support

        stable_state      0.770     0.505     0.610      1518
price_change_next_7d      0.607     0.835     0.703      1389

            accuracy                          0.663      2907
           macro avg      0.688     0.670     0.656      2907
        weighted avg      0.692     0.663     0.654      2907

Top feature importance
--------------------------------------------------------------------------------
diesel: 0.128710
e10: 0.100503
diesel_premium: 0.096218
diesel_petrol_ratio: 0.096027
premium_spread: 0.081717
fuel_price_std: 0.080488
e5: 0.078227
brand_encoded: 0.068623
price_vs_competitor_avg: 0.054480
nearest_competitor_distance_km: 0.045602