Method

How to Test Any Prediction Tool Yourself

Play Big Small

You do not have to take our word for anything on this site. This is the protocol we use, written so that anybody can run it in an evening with a notebook and no money at risk. It works on apps, bots, channels and pdf systems alike, because it tests the calls rather than the technology.

The whole method

Log a hundred calls before results arrive, log a hundred coin flips over the same rounds, then compare the two.

The protocol

Six Steps, No Stake Required

The order matters more than the tools. Recording a call before the result is what separates a test from a memory.

Run this exactly
  1. Open the tool and a blank sheet. Two columns: round number, and what the tool says.
  2. Write the call down before the round settles. Every round, including the ones you would have skipped.
  3. In a third column, write a coin flip for the same round — heads big, tails small. Flip it before the result too.
  4. Record the actual outcome in a fourth column. No interpretation, no notes about how close it was.
  5. Stop at one hundred rounds. Not at a good moment, not after a win — at one hundred.
  6. Count both columns. Tool correct out of 100, coin correct out of 100.
The control

Why the Coin Column Is the Whole Test

Without it you will conclude something from a number that means nothing on its own. With it, the answer takes ten seconds to read.

Do this

Flip before the result

A coin flipped after you know the outcome is not a control, it is a comment. Flip it at the same moment you record the tool's call.

Do this

Keep every round

Skipping rounds you found unclear is how a 50% record becomes a 70% one. The rounds you would have sat out belong in the count.

Avoid

Judging at 20 rounds

Twenty rounds is noise. A coin will show 13–7 often enough that you will convince yourself of something either way, and the same is true of any tool.

Avoid

Counting money instead of calls

Stake size mixes two questions together. Count calls first — whether the tool is right more often than a coin — and only then look at what it would have cost.

Test mistakes

Five Ways a Test Quietly Fails

Every one of these has happened to us, and each turns a clean result into a story.

Recording the call after seeing the result

The most common failure and the most invisible. Once you know the outcome, an ambiguous signal resolves itself in the tool's favour without any dishonesty being involved. Write it down first, every time.

Letting the tool skip rounds

Many tools issue calls only on rounds they describe as strong. That is fine, but then the test is about those rounds only — so run the coin on exactly the same subset, and count both over the same denominator.

Changing stake size mid-test

Raising stakes after losses turns a test of accuracy into a test of your bankroll. Keep every notional stake identical for all hundred rounds.

Restarting after a bad start

If you begin again whenever the first ten go badly, you are selecting your own results. Start once and finish, however it looks at round thirty.

Testing two tools at once

You will unconsciously follow whichever is doing better and mark the other generously. One tool, one hundred rounds, one sheet.

Reading results

What Your Two Numbers Mean

Assume both columns land somewhere near fifty. Here is how far from fifty is interesting, and how far is ordinary.

46 to 54 correct out of 100The ordinary range for a coin
0100
55 to 60 correctUncommon but well within luck at this length
0100
Above 65 correctWorth extending the test to 500 rounds before believing
0100
Tools that beat the coin column in our logsAcross every category tested so far
nonemany

What a Positive Result Would Actually Mean

If a tool finished clearly ahead of the coin over a hundred rounds, the honest next step is not to stake money on it — it is to run five hundred more. Sequences of independent events produce impressive-looking runs regularly, and the only defence is length. A tool that survives six hundred logged rounds against a control would be the first we have seen, and we would want to publish it.

The far more common outcome is two numbers within a few points of each other, which is exactly what the structure of the game predicts. That result is not a disappointment; it is an evening spent buying certainty for the price of a notebook.

Testing Channels and PDF Systems

The same sheet works. For a channel, log the call as posted, with a timestamp, and ignore any message that arrives after the round closed. For a pdf method, write out the rule in one sentence before you start — if you cannot, that is the finding. Methods that require judgement about which rounds to skip should be tested on the skipped rounds too.

If the underlying reason all of this comes out flat is unclear, it is set out in full in the argument against prediction tools. A summary of what each product category costs is on the home page.

Testing FAQ

No, and you should not. The test measures whether calls are correct, which is visible without a stake. Adding money changes nothing about the result except how much the evening costs.

A hundred settles most questions and takes about an hour on one-minute rounds. Five hundred is where a genuinely unusual result would start to be believable.

Record the first one and note the change. A signal that revises itself continuously is untestable by design, and that is worth knowing on its own.

No. In-app histories are produced by the party being tested and often exclude losing calls. The only record that counts is the one you wrote before results arrived.

Yes, for even-money calls. If you are testing colour or number predictions, use a matching random choice with the same number of options instead.

This website is for informational purposes only and is not a gambling operator: we do not accept bets or payments. 18+. Gambling involves risk and may lead to addiction — please gamble responsibly. Information (bonuses/terms/licenses) may change: check current rules on the operator’s website. We are not responsible for decisions and financial losses related to the use of third-party services.