Forecast Accuracy

Anyone can post a forecast. The real question is whether it's right — so every prediction here is graded against the official USFWS count that follows, and the result is published, hits and misses alike. That method is the point; the numbers below are the evidence it produces, dated to the season they cover.

The method

The model publishes a forecast for each species ahead of the next refuge survey.
The U.S. Fish & Wildlife Service runs its ground survey and posts the actual counts.
Every prediction is scored against that actual count — no cherry-picking, misses included.
The model recalibrates on the result, so it sharpens as each season unfolds.

The record as of the 2025–26 season

64%

Direction accuracy — through 2025–26

How often the forecast calls the trend right — whether a species is rising, holding, or falling into the next survey. Counts swing from hundreds to over a million, so direction is the honest, useful measure, not chasing an exact number. This is a living figure: it moves as the model grades each new survey, so it reflects the record through the date shown — not a fixed claim.

1,476

Predictions graded

102

Surveys checked

Seasons (2021–2026)

These figures update as the model learns; last recomputed July 2026. The ground truth is public — check us against the USFWS counts.

Why publish this

A forecast you can't check is just an opinion. Publishing the record is the whole point — it's what separates a real, improving model from a confident guess, and it's the one thing the official survey (which only looks backward) doesn't offer. The ground truth is public; so is our grade against it.

During migration season, live per-species grades appear here as each new survey lands. Off-season, the dated record above stands.