fix: CRITICAL timestamp bug causing 313 vs 336 hour mismatch + validation

**ROOT CAUSE**: Forecast timestamps started at run_date + 1 hour instead of + 1 day
- Forecasted: Sept 30 01:00 to Oct 14 00:00 (336 hours)
- Actuals: Oct 1 00:00 to Oct 14 23:00 (336 hours)
- Overlap: Oct 1 00:00 to Oct 14 00:00 (313 hours only)
- Missing: 23 hours from Oct 14

**FIX 1**: src/forecasting/chronos_inference.py:279
- Changed: timedelta(hours=1) → timedelta(days=1)
- Result: Forecast now covers Oct 1 00:00 to Oct 14 23:00 (correct 336 hours)

**FIX 2**: scripts/evaluate_october_2024.py - Add validation checks:
- Warn if n_hours != 336 (missing data)
- Flag MAE < 0.001 MW as suspicious (data leakage)
- Detect constant forecasts (MAE std < 0.01 across days)

This explains why 84% of borders showed invalid results:
- 13 borders: Zero MAE (need to re-evaluate with correct timestamps)
- 13 borders: Constant forecasts (model behavior issue)
- 6 Poland borders: Identical near-zero MAE (feature issue)

Previous evaluation INVALID - need to re-run with corrected timestamps

scripts/evaluate_october_2024.py

@@ -131,10 +131,18 @@ def main():
             print(f"      [{i:2d}/{len(borders)}] {border:15s} - SKIPPED (no valid data)")
             continue
 
+        # Validation: Check for expected 336 hours (14 days × 24 hours)
+        if len(valid_data) != 336:
+            print(f"      [{i:2d}/{len(borders)}] {border:15s} - WARNING: Only {len(valid_data)}/336 hours (missing {336-len(valid_data)} hours)")
+
         # Calculate overall metrics
         mae = (valid_data[forecast_col] - valid_data[target_col]).abs().mean()
         rmse = ((valid_data[forecast_col] - valid_data[target_col])**2).mean()**0.5
 
+        # Validation: Flag suspicious MAE values
+        if mae < 0.001:
+            print(f"      [{i:2d}/{len(borders)}] {border:15s} - WARNING: Suspiciously low MAE = {mae:.2e} MW (possible data leakage)")
+
         # Calculate per-day MAE (D+1 through D+14)
         per_day_mae = []
         for day in range(1, 15):
@@ -165,6 +173,14 @@ def main():
             day_num = day_idx + 1
             result_dict[f'mae_d{day_num}'] = per_day_mae[day_idx] if len(per_day_mae) > day_idx else np.nan
 
+        # Validation: Check for constant forecasts (same MAE across all days)
+        valid_mae_values = [m for m in per_day_mae if not np.isnan(m)]
+        if len(valid_mae_values) >= 5:  # Need at least 5 days to check
+            mae_std = np.std(valid_mae_values)
+            mae_mean = np.mean(valid_mae_values)
+            if mae_std < 0.01 and mae_mean > 0:  # Essentially zero variation but non-zero MAE
+                print(f"      [{i:2d}/{len(borders)}] {border:15s} - WARNING: Constant forecast detected (MAE std={mae_std:.2e})")
+
         results.append(result_dict)
 
         # Status indicator

src/forecasting/chronos_inference.py

@@ -276,7 +276,7 @@ class ChronosInferencePipeline:
         """
         # Create forecast timestamps
         run_datetime = datetime.strptime(results['run_date'], "%Y-%m-%d")
-        forecast_start = run_datetime + timedelta(hours=1)
+        forecast_start = run_datetime + timedelta(days=1)  # Next day at midnight, not +1 hour
         forecast_hours = results['forecast_days'] * 24
 
         timestamps = [