How it works
Methodology
Where the numbers come from
Each forecast starts from a base rate: how often has this kind of thing happened before? Lake levels start from Alabama Power and TVA records and the seasonal drawdown schedule. Weather starts from the National Weather Service forecast and local climate history. Football starts from team ratings and home-field history. Edgar then adjusts for what is different this time and writes down one number.
A probability is not a promise. 80% means Edgar expects the event to happen about 8 times out of 10 in situations like this one.
What every forecast includes
The exact question as logged, the probability, the date it was posted, the resolve-by date, a short reason, a stability rating (how likely the number is to move: Firm, Soft or Speculative) and the public source that will settle it. A few forecasts logged on Nov. 6, 2026, before this format was set, carry an older stability label (Low or Medium) or no named settlement source; their pages say so, and they are still graded only against their original wording.
Updates never overwrite
When new information arrives Edgar may update a forecast. Each update is a new version with its own date (v2, v3 and so on). Every earlier version stays visible on the forecast's page. A forecast is always graded against its original wording, never a reworded one.
Nothing is back-dated. A forecast's posted date is the day it was written to the ledger. The ledger line number is shown next to every version so each number can be traced.
One display change only: where a ledger note uses the word "odds" (for example, market-implied odds from FedWatch), the site shows the word "probability." Numbers, questions, dates and outcomes are never changed.
How resolution works
On or after the resolve-by date, Edgar checks the named source (for example Alabama Power's lake data, TVA's observed elevations, NWS observations for Cullman Regional Airport, or the official final score) and marks the outcome YES or NO. If the source never publishes the needed figure, the forecast is marked void and left out of the score.
A call counts as right when Edgar said above 50% and it happened, or below 50% and it did not. A 50% call is a coin flip and counts as neither.
If the named source cannot be checked directly, the forecast page says which source was used instead.
The Brier score
Each graded forecast gets a Brier score: (probability − outcome)², where the outcome is 1 for YES and 0 for NO. 0 is perfect; 0.25 is what you get by always saying 50%; 1 is the worst possible. Lower is better.
Scored on Edgar's last update that can be shown to predate each outcome. Updates whose timing can't be verified are excluded; see corrections. The scorecard also shows the score of the very first call so you can see whether updates helped.
Calibration
The calibration table groups graded forecasts by probability (0–20%, 20–40% and so on) and compares the average forecast in each group with how often those events actually happened. A well-calibrated forecaster's 30% calls come true about 30% of the time. With only a handful of graded calls, the table is noisy; it means more as the record grows.
What Edgar does not do
No betting or sportsbook links, no picks, no odds for sale. No buy or sell calls on stocks, crypto or anything else. No affiliate links or ads. Forecasts describe how likely an event is; they are not advice about what to do.