Scorecards · SupnWeather

01A forecast is issued before the day settles 02The checker pairs it with the official high 03The scorecard grades it at 08 / 12 / 16 local 04The report publishes — good or bad

Scope

The scoreboard

latest period against the one before

The grade grid

One cell per region per report period, shaded by the selected metric, graded across the three capture hours. Green is good on MAE and coverage; on bias both extremes shade toward red.

Region × period

shade by

Fig 1

MAE by checkpoint hour

How far off the published high was at each capture time. Later checkpoints know more about the day, so the error usually shrinks toward H16.

Per period

Average by checkpoint

Fig 2

Systematic lean

Weighted signed error per capture hour. Above zero the forecasts run warm, below zero cool. A lean that persists for days is calibration, not bad luck.

Per period

Average, signed

Fig 3

The promises

How often the realised high landed inside the issued interval — and how often it hit the exact bucket a market would settle on. The two numbers a market reader cares about most.

Interval cover

Bucket hits

Fig 4

Clockwork and near misses

The share of eligible checkpoints actually captured — a missed checkpoint is a silent hole in verification — and how often the high landed within 1°C of the published centre.

Capture reliability

Within 1°C

Report logs

Every generated report for the selection. Click any row to expand its per-hour breakdown — eligible, captured, missing, unresolved, within-1°C. ⚠ marks thin periods under 8 paired samples.