Sports handicappers — “touts” — sell picks on the strength of a record they publish themselves. Almost nobody checks those records against what actually happened. Since January 2026 we have been capturing picks at the moment they are posted publicly, then grading them against official results. This report is what that data says. The full dataset is free to download and the queries behind every number are published.
Three findings, each covered in its own section below.
Every pick in this dataset was published in public before the event started, captured by an automated collector, and graded afterwards against the official result. Nothing is self-reported. Nothing is entered by hand after the fact.
That last point is the whole reason the dataset exists. A handicapper's own record is written by the handicapper. Ours is written by a scraper that does not know who is winning and a settlement job that does not care. When a capper deletes a losing post — which happens — we already have it.
| FIELD | VALUE |
|---|---|
| Collection window | 2026-01-04 → 2026-08-24 |
| Total picks captured | 31,972 |
| Distinct authors captured | 4,441 |
| Picks reaching a final result | 9,893 |
| Picks in the clean analysis subset | 6,906 |
| Handicappers in the clean subset | 1,384 |
| Total staked (clean subset) | 10,932.20 units |
The gap between 9,893 settled picks and the 6,906 we analysed is deliberate, and section 8 explains exactly what came out and why. Short version: parlays stored with combined odds, and rows where the odds or stake failed to parse. Leaving them in would have made the numbers look considerably more interesting and considerably less true.
| METRIC | VALUE |
|---|---|
| Graded picks | 6,906 |
| Won | 3,494 |
| Lost | 3,255 |
| Push / void | 157 |
| Win rate (pushes excluded) | 51.77% |
| Net result | −63.58 units |
| Return on investment | −0.58% |
A 51.77% win rate sounds like a winning operation. It is not, and the reason is the price. At a standard −110, a bettor needs 52.38% just to break even; the vigorish is the gap. This population landed at 51.77% — roughly six tenths of a percentage point short, which is what produces the −0.58% return.
−0.58% is not a catastrophe, and we are not going to dress it up as one. A crowd of handicappers landing within a percentage point of break-even is a crowd that is picking better than random and then handing the difference to the sportsbook. The interesting question is not “are they idiots” — they are not — it is whether any individual among them is reliably beating the price. That is what sections 3 and 4 test.
Split the 1,384 handicappers by how many graded picks we hold for each, and ask what share of each group is showing a profit.
| MIN GRADED PICKS | CAPPERS | PICKS | PROFITABLE | % PROFITABLE | GROUP ROI |
|---|---|---|---|---|---|
| 1 or more | 1,384 | 6,906 | 637 | 46.0% | −0.58% |
| 5 or more | 315 | 5,117 | 166 | 52.7% | +1.50% |
| 10 or more | 147 | 4,056 | 84 | 57.1% | +1.69% |
| 20 or more | 62 | 2,894 | 33 | 53.2% | +1.21% |
| 30 or more | 36 | 2,271 | 18 | 50.0% | −2.02% |
| 50 or more | 17 | 1,572 | 8 | 47.1% | −3.40% |
| 100 or more | 4 | 711 | 2 | 50.0% | −2.13% |
From 10 graded picks onward the share of profitable handicappers falls monotonically: 57.1%, then 53.2%, then 50.0%, then 47.1%. Group ROI follows it down, from +1.69% to −3.40%.
This is what regression toward the vig looks like. At small samples, luck manufactures winners — a capper who has posted six picks and hit four of them is “up” in exactly the way a coin can come up heads four times in six. As the sample grows there is less room for luck to hide, and the group settles toward the number the price implies. The handicappers with the longest public track records in our data are, as a group, losing money.
Only four handicappers in this dataset have 100 or more graded picks. That row is in the table for completeness and should not be quoted as a finding — n=4 supports no conclusion whatsoever. The trend we are describing rests on the 10+, 20+, 30+ and 50+ rows, where n is 147, 62, 36 and 17 respectively. Those are modest samples too, and we say so in section 9.
The decay in section 3 could have a comfortable explanation: maybe the good handicappers are simply diluted by a flood of bad ones as the sample grows. So we tested persistence directly — within each individual, does an early winning run tell you anything about what comes next?
We took every handicapper with at least 30 graded picks (36 of them), ordered each history by match date — the date the event was played — and compared profit over their first 10 picks with profit over their next 20. Every one of the 36 has at least three distinct match dates (median far higher), so each history has a real chronology to split.
| GROUP | CAPPERS | PROFITABLE OVER NEXT 20 | RATE |
|---|---|---|---|
| Profitable over first 10 | 17 | 7 | 41.2% |
| Unprofitable over first 10 | 19 | 9 | 47.4% |
In this captured sample, the profitable-starter group had a lower observed proportion of profitable next-20-pick records (7 of 17) than the unprofitable-starter group (9 of 19). Over those next 20 picks, the profitable-starter group returned −0.19%. These small groups do not establish a difference in future profitability.
This test depends entirely on getting each capper's sequence right, so the ordering key is worth stating. We order by match date, not by the row's database creation timestamp. The creation timestamp records when our collector wrote the row, which for backfilled history is not when the pick was made — the earliest creation timestamp in the table is late March 2026, months after the January start of the collection window, and eleven of these 36 handicappers have their entire history written on a single creation date. Sequencing on that field would order those eleven arbitrarily and quietly corrupt the result. Match date has no such problem here: every one of the 36 spans at least three distinct match dates.
This matters commercially, because a 10-pick hot streak is the standard sales asset of the industry. It is long enough to screenshot and short enough to happen by chance constantly. In a population of 1,384 handicappers, hot 10-pick runs are not rare events. They are guaranteed events, and they are guaranteed to belong to someone every single week.
Bucketing all 6,906 graded picks by the price at which they were posted:
| PRICE BAND | PICKS | WIN RATE | ROI |
|---|---|---|---|
| Heavy favorite (−200 or shorter) | 114 | 58.2% | −9.97% |
| Clear favorite (−199 to −130) | 1,807 | 58.6% | −2.78% |
| Near pick-em (−129 to −100) | 3,140 | 51.0% | −3.22% |
| Modest underdog (+100 to +149) | 1,412 | 47.1% | +0.35% |
| Underdog (+150 or longer) | 433 | 42.5% | +27.40% |
The pattern is clean across the four bands that carry real sample: the shorter the price, the worse the return. Handicappers hit favorites often — 58.6% on clear favorites — but not often enough to cover what those favorites cost. Meanwhile the underdog bands, where they win far less often, are the only ones that made money.
This runs opposite to the classic favorite-longshot bias found in public betting markets, where recreational bettors overbet longshots and longshots are therefore the worst value. Among published handicappers we see the mirror image: the commercial demand for a high win rate pushes them onto short prices, and short prices are where their edge disappears.
+27.40% on 433 underdog picks is arithmetically consistent with the 42.5% win rate at an average price of +217 — we checked, and it is not a parlay artifact (section 8). But returns at long prices are extremely high-variance, and 433 picks is not enough to call this a durable edge. Treat it as a description of what happened in this window, not a strategy.
Leagues with at least 100 graded picks in the clean subset.
| LEAGUE | PICKS | CAPPERS | WIN RATE | ROI |
|---|---|---|---|---|
| MLB | 2,170 | 489 | 52.6% | +2.90% |
| Football (association) | 1,231 | 266 | 49.9% | −1.63% |
| NBA | 945 | 264 | 51.9% | −5.94% |
| NCAAB | 556 | 135 | 51.8% | −5.06% |
| Tennis | 425 | 145 | 52.9% | +0.32% |
| NHL | 380 | 103 | 49.7% | +2.64% |
| Cricket | 140 | 32 | 56.9% | +0.11% |
| NFL | 137 | 73 | 54.5% | +4.31% |
| Esports | 137 | 30 | 52.6% | −4.02% |
| WNBA | 113 | 46 | 48.6% | +9.46% |
Basketball is where the money went. NBA and NCAAB together account for 1,501 graded picks at −5.94% and −5.06%. Both show win rates just above 51% — respectable-looking numbers attached to clearly losing returns, which is the favorite tax from section 5 showing up sport by sport.
Sample sizes below roughly 400 picks (Cricket, NFL, Esports, WNBA) are thin enough that the ROI column should be read as noise until the next update. NFL in particular is low here only because the collection window covers an offseason and no full season.
| SOURCE | PICKS | WIN RATE | ROI |
|---|---|---|---|
| Pick-of-the-day posts | 4,860 | 51.1% | −3.33% |
| 1,225 | 54.6% | +10.58% | |
| YouTube | 752 | 51.4% | −1.10% |
| X / Twitter | 69 | 52.2% | +1.48% |
Reddit's +10.58% is the outlier in the table and we want to flag it rather than sell it. Reddit posts in our collector skew toward longer prices than the pick-of-the-day feed does, and section 5 already showed that long prices carried the positive returns in this window. We read the Reddit number as largely a price-mix effect, not evidence that Reddit handicappers are better. Isolating that properly needs a price-matched comparison, which is on the list for the next edition.
9,893 picks reached a final result. We analysed 6,906. Here is every filter, in plain terms, because a number you cannot audit is not worth citing.
| FILTER | WHY |
|---|---|
| Odds outside ±1000 | Multi-leg parlays are stored with combined odds. One captured pick carried +13,535 (“Dodgers and Yankees under 10.5, Pirates ML, Cubs ML, Astros…”). A handful of hit parlays dominates any ROI figure they appear in — with them included, the longest-price bucket showed +178.90%, which is an artifact, not a result. |
| Odds between −99 and +99 | Not valid American odds. These are parse failures, overwhelmingly 0, from posts where the price was written in decimal or European format (@2,00) and misread. |
| Stake outside 0.1–10 units | Same root cause. Where a decimal price was misread as a stake we saw “units” of 75–85, which would silently dominate any stake-weighted average. |
| Null author | Cannot be attributed to a handicapper, so cannot support any per-capper analysis. |
| Unsettled / pending | No final result yet. |
Both artifacts above would have made this report better reading. A “+178% ROI on longshots” headline is far more clickable than −0.58%. We are publishing the exclusions so that anyone can check we did not choose the filters after seeing which way they moved the answer. The filter set is defined once, in SQL, and every table above is generated from it — see section 11.
This report and the accompanying dataset are free to use, including commercially, under CC BY 4.0. You do not need to ask permission. Attribution with a link is all we ask.
<a href="https://captracker.markets/state-of-sports-handicapping">The State of Sports Handicapping 2026</a>, CAPTRACKER
Journalists: we will run custom cuts of this data on request — by sport, by platform, by date range, or for a named handicapper — at no cost and with no conditions attached. Contact details and a full media kit are at captracker.markets/press.
Every figure in this report comes from one annotated SQL file run against the collection database. The aggregate tables are published as CSV.
The CSV contains every aggregate table in this report as tidy rows. Individual handicappers are not named in the dataset: this report is about the population, and we are not interested in publishing a list of people to point at. Per-capper records for handicappers we track are visible on the site itself at /leaderboard.
How the collection and settlement pipeline works — and how our own betting record is kept separately from it — is documented at /methodology. If you find an error in this report, tell us and we will correct it in place and note the correction below.