How to Spot a Fake Tipster — Seven Checks That Take Five Minutes

The decisive test is not the strike rate on display. It is whether the record was written before kick-off, somewhere it cannot be edited afterwards. Almost everything that looks like proof in this market — screenshots, a percentage in a bio, a run of winners — is produced after the result is already known, and therefore proves nothing at all.

Seven checks follow. None requires statistical training, and all of them can be done before you pay anything.

1. When was the record created

Ask yourself one question: if this source wanted to remove a selection that lost, could it? In a message channel, a social post, or a gallery of screenshots, the answer is yes, and nobody outside would ever know.

A record you can lean on is written before kick-off and stored where a row is not edited and not deleted, including when it is embarrassing. That is why every prediction we publish is given an identifier, a timestamp and a signature at publication time, not at settlement. The full payload and signing scheme are set out in the guide on what a verified prediction record looks like.

2. Are the losses shown

A source that displays only what worked is not showing data. It is showing a selection of data. Without the misses there is no denominator, and without a denominator there is no strike rate — only a story.

Look for a single table where hits and misses sit together, by date, with no holes in the timeline. A two-week gap in the middle of a season is almost always a period that was removed. Our record shows settled predictions unfiltered, weaker stretches included.

3. Is the price at entry shown

This is the check almost nobody runs, and it separates analysis from storytelling. The same selection can be a good decision at one price and a poor decision at another.

An arithmetic illustration, not a performance figure: suppose a model puts an outcome at 40%. Fair odds for 40% are 2.50. Someone who entered at 2.80 took more than the probability justifies; someone who entered the very same selection at 2.20 took less. Both decisions look identical in a results table that reports only "won" or "lost" — which is precisely why such a table does not measure decision quality, even when every number in it is entirely genuine.

It follows that two sources cannot be compared on strike rate alone. A source hitting often at short prices can be worse than a source hitting less often at long ones. The concept is covered separately in fair odds and the no-vig price, and the vocabulary is in the glossary.

4. Is there an abstention state

A source with a selection every day, every round, in every league is selling availability, not analysis. Most fixtures are not an opportunity; that is arithmetic rather than modesty. Anyone who never says "there is nothing to do here" is claiming to find an advantage everywhere, which is the least plausible claim in the category.

For us, abstention is a declared state rather than an absence of content, and it is shown with the reason that produced it. That side is covered in when to pass a bet.

5. Is a probability stated

"Home win" is a selection, not a prediction. A prediction says how likely something is, which makes it checkable afterwards: when the source said 40%, did it happen roughly 40% of the time? A selection without a probability cannot be examined at all — you can only count winners, and counting winners without prices and without a denominator measures nothing.

Our model publishes a calibrated probability for every selection, and the method is described in how the model works.

6. Is it shown what changed

A good prediction updates when new information arrives — a lineup, an injury, a price move. The question is what happens to the previous version. If it disappears, there is no way to tell whether the update was made before the match or after it. If it is retained and marked, the whole sequence stays visible.

7. Can you verify it yourself

The last check is also the strongest: can an outsider examine the record without trusting the publisher. When both the signed data and the signature are published, you can recompute the signature and compare. Change a single digit and you get a different result. Without a mechanism like that, every claim about history rests on trust alone.

What a good answer looks like

The checks above are questions, and you can simply ask them. What matters is the shape of the answer, not its content.

A good answer is specific and immediate: a link to the record, an explanation of what each row stores, and an explicit statement of what is not measured. A source that says "we do not measure that" gains credibility rather than losing it, because anyone with a number for every question is probably filling gaps with estimates.

A poor answer moves the conversation: it offers a trial instead of data, tells you about one large win, explains that the method is proprietary and therefore cannot be shown, or describes the person asking as someone who does not understand. A method can be proprietary. A results record cannot be both secret and evidence.

Watch what happens after a weak run, too. A source that explains a weak run purely through external factors — referees, variance, injuries — and never through anything it did itself will explain the next one the same way.

What you can measure once the record is real

Suppose a source passes all seven. Now questions that were previously pointless become worth asking:

• Calibration — when it said 40%, did it happen near 40% of the time? This is a test across groups of predictions, not single ones, so it needs a continuous series rather than a chosen sample.
• Continuity — do the weak stretches appear in the table, or does the season begin just after one.
• Sample depth — how many predictions have settled in total. A small sample is not evidence either way.
• What is absent — a metric that is not measured should not appear as an estimate. A declared gap beats a guess.

None of this promises future profitability, and that is worth saying plainly: a good record makes a source checkable, not certain.

About paying

Payment is neither a sign of quality nor a sign of a problem — but it moves the evidence question forward, not backward. A source asking for money before you can see any record is asking for trust on the basis of trust.

Two practical checks before paying: is there an open part you can examine without an account, and is the historical record accessible without paying. Past results are not merchandise; they are the evidence the request for future money rests on. A source that locks its history is selling you the evidence along with the product.

What to do with the answers

A source that fails checks 1 and 2 does not need examining further — there is nothing to measure. One that passes 1 and 2 but fails 3 is reporting real data without the context that gives it meaning. One that passes all seven is not necessarily profitable, but it is at least checkable, and that is the precondition for every other discussion.

These checks work on us too, which is the point. The full comparison between a model and a tipster is in model vs tipster, and the separation between decision quality and outcome is in good decision, bad outcome.

Analytical predictions are not a promise of an outcome, and no method removes risk. This content is intended for readers aged 18 and over. If betting stops being a choice, support is available — see responsible use.