Here’s the test that matters: when we say a number, does it actually hit that often? Every pick is graded against its final pre-match number, and every one of them is on this page.
These buckets cover our ATP and WTA main-tour picks, and since July 8 challenger events too — challenger picks count inside the same ATP/WTA buckets below, and they’re roughly a tenth of the volume so far. Upset calls are the one thing kept out of them: backing a player the market makes a clear underdog is a pick we expect to lose more often than we win, so averaging it in here would describe neither half. They get their own section further down, graded the same way. UFC, cricket, college football, the NFL and table tennis each have their own graded section further down.
Across 2,209 graded tennis picks, our pick won 65.1%; in UFC we're 46-24 over 70 graded picks; in cricket we're 83-37 over 120 graded picks; in college football we're 186-35 over 221 graded picks; in the NFL we're 33-21 over 54 graded picks; in table tennis we're 2,503-1,499 over 4,002 graded picks. The tables below show how that holds up at each confidence level — including where we’ve run overconfident. Nothing’s hidden.
6,676
graded picks, all sports
65.1%
tennis pick win rate
0.226
tennis accuracy score
Tennis
Our public tennis record — how often our picks won, an accuracy score, and every graded pick broken out by confidence level.
Graded picks
2,209
the number we published before each match started, all tours
How often our pick won
65.1%
across every graded pick; upset calls have their own record below
Accuracy score
0.226
lower is better; 0.25 = a coin flip
Does the number hold up?
How our picks did, by confidence level
| We said | Matches | Our pick won | Actual rate | Likely range | Accuracy score |
|---|---|---|---|---|---|
| 50-55% | 532 | 295 | 55.5% | 51.2–59.6% | 0.248 |
| 55-60% | 503 | 328 | 65.2% | 60.9–69.2% | 0.234 |
| 60-65% | 441 | 282 | 63.9% | 59.4–68.3% | 0.231 |
| 65-70% | 315 | 227 | 72.1% | 66.9–76.7% | 0.204 |
| 70-75% | 201 | 141 | 70.1% | 63.5–76.0% | 0.208 |
| 75-80% | 98 | 73 | 74.5% | 65.0–82.1% | 0.191 |
| 80-85% | 63 | 50 | 79.4% | 67.8–87.5% | 0.165 |
| 85-90% | 33 | 26 | 78.8% | 62.2–89.3% | 0.174 |
| 90-95% | 16 | 12 | 75.0% | 50.5–89.8% | 0.214 |
| 95-100% | 7 | 4 | 57.1% | 25.0–84.2% | 0.402 |
Grading uses only the number we published before each match started — locked in before first serve. Retirements and walkovers are excluded.
Set-score accuracy
Every match page shows a "most likely score" — this is how often that exact score actually happened. Best-of-3 and best-of-5 are not directly comparable: bo3 has 4 possible scores and bo5 has 6. They are also graded on very different numbers of matches, so the range under each rate says more than the rate does.
Best-of-3
40.9%
exact score correct, 767 of 1,876 · 95% range 38.7–43.1%
Best-of-5
35.7%
exact score correct, 50 of 140 · 95% range 28.3–43.9%
Set-count calls
Since July 19 every match page carries a set-count call — how many sets we think it goes. Most matches genuinely end in straight sets, so the honest baseline is high: always predicting straight sets would be right about two-thirds of the time in best-of-3. The rows below split our record by what we called, so the bold three-set calls are graded in the open, separately from the safe ones.
Best-of-3 · Straight sets (2)
64.0%
called it, 573 of 896
Best-of-3 · Three sets (3)
36.3%
called it, 262 of 722
Best-of-5 · Three sets (3)
41.2%
called it, 7 of 17
Best-of-5 · Goes long (4+)
58.1%
called it, 61 of 105
Upset calls
About 1 in 12 of our picks back a player the market makes a clear underdog — the betting favorite is the other side. We expect to lose these more often than we win them; the honest comparison is against underdogs in general, who win roughly a third of the time. These are graded here rather than in the confidence buckets above, so neither number is flattered by the other.
Upset-call record
81-123
39.7% of 204 graded calls
If we’d sided with the market
60.3%
how often the market favorite would have won these
How often any underdog wins
~35%
roughly, across underdogs in general
ATP
63.9% of picks won over 1,386 picks · accuracy score 0.231
WTA
67.2% of picks won over 823 picks · accuracy score 0.216
UFC
Our UFC record: every fight pick we’ve graded, by confidence band. It’s early days: 70 picks graded so far, so read these numbers as a starting point, not a verdict.
Picks graded
70
the number we published before each fight started
How often our pick won
65.7%
46-24 across every graded fight
Accuracy score
0.221
lower is better; 0.25 = a coin flip
Does the UFC number hold up?
How our UFC picks did, by confidence level
| We said | Fights | Our pick won | Actual rate | Likely range | Accuracy score |
|---|---|---|---|---|---|
| 50-55% | 22 | 13 | 59.1% | 38.7–76.7% | 0.247 |
| 55-60% | 15 | 7 | 46.7% | 24.8–69.9% | 0.254 |
| 60-65% | 11 | 10 | 90.9% | 62.3–98.4% | 0.169 |
| 65-70% | 10 | 7 | 70.0% | 39.7–89.2% | 0.215 |
| 70-75% | 12 | 9 | 75.0% | 46.8–91.1% | 0.184 |
Grading uses only the number we published before each fight started — locked in before the walkouts. Cancelled bouts are excluded.
Cricket
Our cricket record: every pre-toss lean we’ve graded, by confidence band. It’s early days: 120 picks graded so far, so read these numbers as a starting point, not a verdict.
Picks graded
120
the number we published before each match started
How often our lean won
69.2%
83-37 across every graded match
Accuracy score
0.196
lower is better; 0.25 = a coin flip
Does the cricket number hold up?
How our cricket leans did, by confidence level
| We said | Matches | Our lean won | Actual rate | Likely range | Accuracy score |
|---|---|---|---|---|---|
| 50-55% | 31 | 17 | 54.8% | 37.8–70.8% | 0.245 |
| 55-60% | 32 | 16 | 50.0% | 33.6–66.4% | 0.250 |
| 60-70% | 31 | 26 | 83.9% | 67.4–92.9% | 0.182 |
| 70-80% | 21 | 19 | 90.5% | 71.1–97.3% | 0.107 |
| 80-90% | 4 | 4 | 100.0% | 51.0–100.0% | 0.023 |
| 90-100% | 1 | 1 | 100.0% | 20.7–100.0% | 0.005 |
Grading uses only the number we published before each match started — locked in before the toss. Cancelled fixtures, ties, and no-results are excluded.
College Football
Our college football record: every published win probability we’ve graded, by confidence band. The probability is the pick, so the test is simple. A stated 68% should have landed about 68% of the time, and this table is where that shows.
Picks graded
221
the number we published before each kickoff
How often our pick won
84.2%
186-35 across every graded game
Accuracy score
0.128
lower is better; 0.25 = a coin flip
Does the college football number hold up?
How our college football picks did, by confidence level
| We said | Games | Our pick won | Actual rate | Likely range | Accuracy score |
|---|---|---|---|---|---|
| 55-60% | 28 | 18 | 64.3% | 45.8–79.3% | 0.234 |
| 60-65% | 28 | 19 | 67.9% | 49.3–82.1% | 0.221 |
| 65-70% | 23 | 20 | 87.0% | 67.9–95.5% | 0.155 |
| 70-75% | 22 | 18 | 81.8% | 61.5–92.7% | 0.156 |
| 75-80% | 18 | 16 | 88.9% | 67.2–96.9% | 0.112 |
| 80-85% | 31 | 27 | 87.1% | 71.1–94.9% | 0.116 |
| 85-90% | 20 | 19 | 95.0% | 76.4–99.1% | 0.052 |
| 90-95% | 26 | 25 | 96.2% | 81.1–99.3% | 0.038 |
| 95-100% | 25 | 24 | 96.0% | 80.5–99.3% | 0.039 |
Grading uses only the number we published before kickoff. A typical week publishes fewer games than it plays: anything under our 55% confidence floor — and the handful of games where our number sits so far from the consensus that history says we’re probably the ones wrong — never gets published at all. The games we leave out are part of the honesty, not a gap in it.
College Football Futures
Our 2026 national championship board — one favorite, one live dog, one longshot, published in the preseason and settled when the title is decided. 3 still open.
- Top pickOhio State Buckeyes16%Open
- ContenderIndiana Hoosiers5%Open
- LongshotTennessee Volunteers2%Open
Futures are reported as a plain board, not a calibration curve — the sample is three picks a season.
NFL
Our NFL record: every published win probability we’ve graded, by confidence band. Every game on the slate gets a call, so these buckets cover the coin flips too, right down to where a stated 55% should have landed about 55% of the time. It's early days: 54 picks graded so far, so read these numbers as a starting point, not a verdict.
Picks graded
54
the number we published before each kickoff
How often our pick won
61.1%
33-21 across every graded game
Accuracy score
0.214
lower is better; 0.25 = a coin flip
Does the NFL number hold up?
How our NFL picks did, by confidence level
| We said | Games | Our pick won | Actual rate | Likely range | Accuracy score |
|---|---|---|---|---|---|
| 50-55% | 9 | 4 | 44.4% | 18.9–73.3% | 0.255 |
| 55-60% | 8 | 3 | 37.5% | 13.7–69.4% | 0.267 |
| 60-65% | 11 | 6 | 54.5% | 28.0–78.7% | 0.259 |
| 65-70% | 10 | 6 | 60.0% | 31.3–83.2% | 0.255 |
| 70-75% | 6 | 4 | 66.7% | 30.0–90.3% | 0.224 |
| 75-80% | 5 | 5 | 100.0% | 56.6–100.0% | 0.045 |
| 80-85% | 5 | 5 | 100.0% | 56.6–100.0% | 0.035 |
Grading uses only the number we published before kickoff. Every game on the slate is published — there is always a pick side, and our confidence is capped at 90%, the edge of the calibration table the model was validated on. A tie is void: it is excluded from the record and from every bucket above, and it never counts as a miss.
NFL futures
Our 2026 Super Bowl board — a team from our top three by simulated title odds, a contender, and a longshot, published before the season and settled when the game is played. 3 still open.
- Top pickSeattle Seahawks23%Open
- ContenderHouston Texans8%Open
- LongshotSan Francisco 49ers3%Open
Futures are reported as a plain board, not a calibration curve — the sample is three picks a season. The numbers come from our own season simulation, and two known limits of it push the top of the board high; the futures page says which.
Table Tennis
Our table tennis record: every TT Elite match we’ve graded, by confidence band. We call around 300 matches a day, so this sample grows fast; 4,002 picks are graded so far.
Picks graded
4,002
the number we published before each match started
How often our pick won
62.5%
2,503-1,499 across every graded match
Accuracy score
0.229
lower is better; 0.25 = a coin flip
Does the table tennis number hold up?
How our table tennis picks did, by confidence level
| We said | Matches | Our pick won | Actual rate | Likely range | Accuracy score |
|---|---|---|---|---|---|
| 50-55% | 1,164 | 639 | 54.9% | 52.0–57.7% | 0.247 |
| 55-60% | 912 | 535 | 58.7% | 55.4–61.8% | 0.244 |
| 60-65% | 720 | 445 | 61.8% | 58.2–65.3% | 0.235 |
| 65-70% | 497 | 348 | 70.0% | 65.9–73.9% | 0.211 |
| 70-75% | 320 | 244 | 76.3% | 71.3–80.6% | 0.182 |
| 75-80% | 200 | 148 | 74.0% | 67.5–79.6% | 0.192 |
| 80-85% | 112 | 83 | 74.1% | 65.3–81.3% | 0.198 |
| 85-90% | 57 | 44 | 77.2% | 64.8–86.2% | 0.188 |
| 90-95% | 20 | 17 | 85.0% | 64.0–94.8% | 0.130 |
Grading uses only the number we published before each match started, frozen when it first went up. Retired and walkover matches are excluded.
NFL fantasy projections
The Lineup Desk publishes weekly point projections for every startable player, free in full, and grades them against actual fantasy points the following Tuesday. Across 4 graded weeks, our projections put same-position pairs in the right order 65.1% of the time, and the average projection missed by 5.73 half-PPR points.
Get the free ebook + tomorrow's picks
Sign up for the daily desk email and our nine-chapter beginner's guide to trading sports prediction markets comes free as a PDF. One email a day with the day's free picks in full, unsubscribe anytime.