Updated August 27, 2026. Reviewed by David.
Quick answer
This page is about evaluating a trend claim — the tests that separate a real pattern from a number produced by searching. For what betting trends are and which categories tend to survive, start with betting trends.
One rule does most of the work: ask how many patterns were examined before this one was published.
The multiple-comparisons problem
Take any results database and apply situational filters — home, road, favourite, underdog, rested, division, primetime, after a loss. A handful of filters generates thousands of combinations. In a dataset that size, extreme records appear by chance alone, with no predictive content whatsoever.
Publishing the extremes and discarding the rest produces an unlimited supply of impressive statistics. This is not dishonesty; it is what searching a large space does. But it means the striking record you are reading is, by default, the winner of a lottery you were not shown.
Four tests
| Test | Question | Failure looks like |
|---|---|---|
| Pre-registration | Was the filter chosen before looking? | Filters that feel oddly specific. |
| Mechanism | Why would this be true? | No causal story, only a record. |
| Split-half | Does it hold in both halves of the sample? | Strong early, absent later. |
| Closing line | Does it beat the close, not just win? | Wins without beating the number. |
Why closing-line performance is the strongest test
Results are noisy; prices are not. A strategy that consistently takes better numbers than the closing line is demonstrating something real about its information, even across a stretch of losing bets. A strategy that wins without beating the close has probably been lucky.
This is why PropsBot grades against the closing number in its performance methodology, and why closing line value is the metric worth watching in anyone’s published record, including ours.
Warning signs in published trends
- No denominator. A record without a sample size, or without the number of filters tested.
- Suspiciously precise filters. “Road underdogs of 3 to 6.5 after a bye” is a search result, not a hypothesis.
- Unfalsifiable framing. If nothing would disprove it, it is a story.
- Results-only grading. Win rates quoted with no reference to the prices taken.
How PropsBot applies this to itself
The same standard, or the published record means nothing. Every pick is timestamped at publication with the posted line, graded against the actual outcome, and passes are published alongside plays. A service that only shows its winners is running the same selection process that produces bogus trends.
Where to go next
- Betting trends — what they are and which survive.
- Public betting trends — crowd behaviour patterns.
- Closing line value — the test that matters most.
- Public track record — graded per sport (NFL 16.8% ROI on 8,571 High ROI Signal picks, as of September 29, 2026).
Frequently asked questions
How do you tell if a betting trend is real?
Four tests: was the filter chosen before examining the data, is there a mechanism explaining it, does it hold when the sample is split in half, and does it beat the closing line rather than merely winning.
Why do most published betting trends fail?
Because they are found by searching thousands of filter combinations. In a search that large, extreme records occur by chance, and publishing only the extremes guarantees a supply of patterns with no predictive value.
What sample size does a betting trend need?
There is no single threshold. What matters more is the denominator of the search: a record surfaced from thousands of tested combinations requires far more evidence than one predicted in advance.
Why grade against the closing line instead of results?
Results are noisy over any realistic sample. Consistently beating the closing number demonstrates genuine information even through losing stretches, while winning without beating the close is usually variance.
Model output is research, not a guarantee. Lines move; verify the current price before betting. 21+. Please gamble responsibly.
