Classification: correct Confidence: The model's ability to identify 'quiet days' under dry conditions is a strong point. Maintaining high confidence in the 'no-rise' or 'low-rise' classification for dry antecedent states is crucial for reducing false positives.
The model correctly predicted a near-baseline quiet day, with the predicted peak (1.96 ft) falling well within the acceptable error margin of the observed peak (1.84 ft).
| Metric | Predicted | Actual | Error |
|---|---|---|---|
| Peak height | 1.96 ft | 1.84 ft | +0.12 ft |
| Total rise | — | 0.03 ft | — |
| Band | Precip | Predicted Rise | Intensity | Moisture |
|---|---|---|---|---|
| 1 | 0.53" | 0.00 ft | HEAVY | VERY DRY |
| 2 | 0.67" | 0.00 ft | HEAVY | VERY DRY |
| 3 | 0.83" | 0.00 ft | HEAVY | VERY DRY |
| 4 | 1.19" | 0.07 ft | HEAVY | VERY DRY |
| 5 | 1.11" | 0.04 ft | HEAVY | VERY DRY |
Headline: ~45% chance of rising to LOW within the next 12 h. Experimental. Settled outcome: None (reached None) LLM said headline was correct: None Notes: The empirical model indicated a ~45% chance of reaching the 'LOW' tier (3.0 ft). The actual peak was 1.84 ft ('below_low'). While the 'LOW' target was not reached, the model's probability estimate was not a definitive 'false positive' in a binary sense, but the outcome was 'below_low'. The headline is neutral in terms of a direct hit/miss, but the gauge stayed well below the 'LOW' threshold.
The event on September 10, 2026, was characterized by minimal hydrological response despite moderate QPE in the upper bands. The watershed was in a 'VERY DRY' antecedent state, and the gauge recorded only 0.16 inches of precipitation. The hydrograph remained flat with a negligible rise of 0.03 ft, confirming that the dry conditions effectively suppressed runoff from the rainfall that did occur.
The prediction of a 1.96 ft peak was a conservative overestimation of the 1.84 ft actual peak, resulting in a 6.5% error. Given the near-baseline nature of the event and the conservative requirement to avoid false negatives, this margin is well within the 'correct' classification threshold. The model successfully identified that the antecedent moisture was too low to generate a significant rise from the QPE signals.
Predictions issued today (archive): 8 active issuance(s) from 2026-09-10T21:20:02.477959+00:00 to 2026-09-11T04:20:02.773598+00:00; issued peaks ranged 1.81 to 1.96 ft. The prediction graded below is the HIGHEST issued peak of the day (issued 2026-09-11T04:20:02.773598+00:00), not the last card of the day.
Grade-only mode: no adjustments suggested; calibration is fit offline.
The model's ability to identify 'quiet days' under dry conditions is a strong point. Maintaining high confidence in the 'no-rise' or 'low-rise' classification for dry antecedent states is crucial for reducing false positives.