Free — no account
Every prop. Every number. Nothing hidden.
We read all 26,395 props on the PrizePicks board, test each one we can model against the player’s game log, and publish the few that survive with the evidence beside them. No credits, no unlocks, no email.
- CFB
- NBA
- WNBA
Today
refreshes every ~15 min
Biggest line moves
All line moves- Jesse LegreeCFB · Receiving Yards, now 106.5, down from 125.5 at today's open
- Matthew StaffordNFL · Pass Yards, now 248.5, down from 265.5 at today's open
- Brad JacksonCFB · Pass Yards, now 219.5, up from 204.5 at today's open
- Steve AngeliCFB · Pass Yards, now 222.5, down from 236.5 at today's open
- AJ NewberryCFB · Rush+Rec Yds, now 88.5, up from 75.5 at today's open
Top of today’s board
The best-evidenced picks right now
- Cleared
- 13/15
- Projection
- 13.3
- L15 avg
- 13.3
- Market line
- 11.5
- Books
- 11.5
- Other apps
- 10.5–
11.5
19–10 · 66% hit · break-even 44.7%
Likely true rate 50–78%, clear of break-even.
- Cleared
- 6/7
- Projection
- 1.7
- L7 avg
- 1.7
- Market
- no price
15–24 · 38% hit · break-even 44.7%
Likely true rate 27–52%. Too close to break-even to call yet.
ML ensemble 56% · experimental
- Cleared
- 13/15
- Projection
- 4.5
- L15 avg
- 4.1
- Market line
- 4.5
- Books
- 4.5
- Other apps
- 4.5
6–6 · 50% hit · break-even 66.7%
Likely true rate 29–71%. Too close to break-even to call yet.
Final
10graded from the box score · demons count on their own line, apart from the record
7 more in play, awaiting a grade or graded today on the board
This refresh
26,395 props in.
3 out.
Every refresh we read the entire PrizePicks board, keep only what we can actually model, project it against each player’s game log, and publish what survives. Most of it doesn’t.
-
scanned26,395
Every projection on the PrizePicks board this refresh, across all 28 leagues.
-
modelable14,596
Leagues with a per-player game log, minus period- and half-scoped boards. No history, no projection.
-
projected12,103
Props with enough sample to build a distribution and compare it to the line.
-
published3
What survived the sample, injury, role and concentration filters.
How to read a pick
Read the tape
Every pick on the board carries up to the player’s last 15 games, hung off the prop line. It is the evidence the percentage was built from, drawn so you can check it in a glance.
NBAToumani CamaraPoints OVER 6.5the top pick’s real game log
The line is the baseline
The rule across the middle is the prop line — 6.5 here. One bar per game: 15 games ago on the left, the last game on the right.
Up means it cleared
Lit bars rose past the line; dim bars fell short — 13 of 15 here. On an UNDER, clearing means finishing below the line, so up always means the pick won that game.
Height is the margin
A run of narrow escapes looks different from a run of blowouts — exactly what a single hit-rate percentage hides.
Method
How the model works
What happens between the PrizePicks board and a published pick — including what the model refuses to price and how far off it has measured itself to be.
Every refresh, in four phases
-
01
Fetch the board
Every PrizePicks prop across 28 leagues — 26,395 projections this refresh.
-
02
Find the history
Players are resolved to their game logs, through official player IDs for the US leagues. A league qualifies only when a per-player history exists and its stat columns have been verified by hand. Tennis, esports, darts and motorsport have none, so they are skipped: a projection there would be a guess wearing a percentage sign.
-
03
Price it
Each surviving prop gets matchup context, a model probability and, where a sportsbook prices it, a market price.
- Market price — devigged sportsbook odds at the same line as the prop (a book on another line gives no price); the publish check averages every book that prices it
- Matchup engine — pace, rest days, home/away, defense, back-to-backs, and teammates ruled out (NBA)
- Edge Engine — the Tier-2 blend below turns it into one probability
-
04
Publish and grade
Filters cut most of what is left: minimum sample size, minutes stability, injury and DNP guards, and per-player and per-team caps so one bad night cannot sink a slate. Every published pick is then paper-traded and graded automatically from the box score.
See the graded record
The probability
The Edge Engine’s Tier-2 probability blends three estimates of the same question: does this player clear this line?
- Empirical CDF
- 35%
- The player’s own distribution of results, read at the line.
- Hit rate
- 30%
- How often the line was cleared across the game log.
- Possession bootstrap
- 35%
- 5,000 games resampled from the player’s log, adjusted for the minutes trend, pace (NBA and NHL) and teammates ruled out (NBA).
Combo stats such as Pts+Rebs+Asts shift weight toward hit rate.
Adjustments on top
What the game log alone cannot see, and the checks a pick passes on its way out.
- NBA paceteam pace ratings
- NFL weatherstadium forecasts, 32 stadiums
- MLB weatherstadium forecasts
- MLB contact-quality regressionexpected stats
- Game total and spreadbetting-market lines
- Home/away splits
- Back-to-back detection
- Defense multiplier
- Minutes trend
- Lineup boostNBA only
- Calibrationisotonic regression per prop type; Platt scaling as the fallback
- Kelly sizing
- Market vetosportsbook price check: the average of the books that price the prop
Where the numbers come from
Everything the pipeline reads from, from the board itself to the weather at the stadium.
- PrizePicks board
- Official game logs and box scores
- Injury reports
- Sportsbook player-prop lines
- Game totals and spreads
- Stadium weather
- Pace, tracking and advanced stats
- Headlines from major sports outlets
Traps we refuse
Period-, quarter- and half-scoped boards — NFL1H, CFB1H, NHL1P and their kind — are excluded outright. The history is full-game, and a full-game average against a first-period line measures two different things.
- Goalie’s average
- 25saves a game
- First-period line
- 7.5saves
- Model said
99%a lock- The truth
- ≈50%a coin flip
Learned the expensive way. Nothing period-scoped comes back until a period-level game log exists to price it.
Calibration, printed
Raw model output runs hotter than reality — published picks have historically landed below their modelled confidence. A curve fitted on resolved picks corrects the raw output before anything is published, and the gap it corrects is measured and printed here rather than applied quietly.
Measured on 852 graded predictions: raw model output ran 12.6 percentage points hotter than reality. It is printed because a model that hides its own overconfidence is worth less than no model. Every graded pick is on the board’s track record.
What this is not
A model is an estimate, not a prediction. This is model output, not advice and not a guarantee. Treat any win rate with the scepticism its sample size deserves, and bet only what you can afford to lose. 21+.
Why it’s free
Locking a pick behind credits does not make it better — it just stops you checking it. Everything the model produces is on the board with the evidence attached, so you can disagree with it.
Today’s board
Check the evidence yourself
3 picks cleared every filter this refresh, each with its tape and projection, and the sportsbook price wherever one exists. Nothing is locked.