How to Evaluate a Football Prediction Site: A 9-Point Checklist

Nearly every football prediction site fails the same four checks: no complete archive, no odds recorded, no probabilities, and a sample too small to mean anything. Run the list below on any service — including this one — and you will know within ten minutes whether there is a model behind the numbers or just confident copy.

1. Is there a complete, timestamped public record?

Not a highlights reel. Every published pick, win and loss, dated before kickoff. If the archive begins partway through the site's life, or picks can be edited after the fact, nothing downstream is checkable. This is the check most sites fail, and failing it should end the evaluation.

2. Is the sample large enough?

A few hundred settled picks is the minimum before a hit rate carries information. Below that, a strong-looking record is well within what randomness produces. Ask how many picks the number is based on; if the site does not say, assume it is small.

3. Are the odds recorded?

A hit rate without prices cannot be converted into profit or loss. Sixty percent at average odds of 1.50 loses money. Forty-five percent at 2.60 makes it. A site reporting accuracy but not odds is reporting the less useful half.

4. Are the outputs probabilities or verdicts?

"Home win" cannot be evaluated except as right or wrong. "Home win, 54%" can be checked for calibration across hundreds of matches. Probabilities are also the only form you can compare against a price — see value bets explained.

5. Is the confidence score meaningful?

If the site grades picks by confidence, the high-confidence group must outperform the low-confidence group in the published record. If it does not, the score is decoration, and it is decoration people are staking money against.

6. Which markets is the record built on?

An impressive number on "over 0.5 goals" or double chance is an easy number. Look for the record broken out by market. A blended figure across easy and hard markets is engineered to look good.

7. Is the methodology described concretely?

"Advanced AI algorithms" is not a methodology. What data goes in? What is the model class? How often does it retrain? What does it not account for? A site unwilling to name its limitations is a site that has not thought about them — ours are on how it works.

8. Are the claims plausible?

9. What is the business model?

A subscription aligns the site with being useful, because people cancel. Affiliate revenue paid per deposit aligns it with volume of betting, which is a different thing. Neither is disqualifying, but you should know which one you are reading, and it should be disclosed.

Applying this to Prodict

The same list, honestly:

• Full public record, every published pick, at picks history.
• Probabilities and stated drivers on every match, not bare verdicts.
• Methodology and its limits described on how it works.
• Subscription-funded; we are not paid per deposit.
• What we do not claim: that we beat the closing line consistently, or that any pick is safe. The model is a research tool, and the archive is there so you can judge it rather than take our word for it.

One last principle

The strongest signal is not what a site claims — it is what it is willing to publish that makes it look worse. A visible losing month, an honest note that a market is not the model's strength, a clear statement of what it cannot see. Sites optimising for conversion never publish those. It is the cheapest tell there is.