Which NFL player props have the best chance this week?
How this board is built, and how well it has done
How this is built. Five markets, each with a ladder of posted-style bands — anytime touchdown,
rushing yards (25+ to 100+), receiving yards (20+ to 100+), receptions (2+ to 7+) and passing yards
(175+ to 300+). For every player in a team's last three games, sixteen signals are read from games played
before the one being described: his own rate of clearing that exact band, where the band sits in
his own distribution, his share of the team's carries, targets or attempts, how that share has moved over
three games, what the opposing defence concedes to his position, and the market's implied points total
for his team. One penalised logistic regression per market prices every rung of its ladder, so the
probabilities down a player's ladder stay consistent with each other. Everything is scored by replaying the
seasons week by week, never letting a model see the week it is predicting.
How well does each market score?
Walk-forward: at every week each model had been fitted only on earlier weeks. Lower log loss is
better. The baseline is the player's own rate of clearing that same band, over exactly the games being graded.
Market
Band-games graded
Log loss
Own-rate baseline
AUC
Beats baseline
What happened to the picks the board actually showed?
The board only shows bands priced above 50%. Selecting those naturally picks the
easier bands, so a high hit rate there is mostly arithmetic. The column that matters is the last one.
Market
Bands shown
Landed
Model predicted
Own base rate already implied
Attributable to the model
Which bands does the model handle, and which does it not?
Per rung of each ladder. A band where the model does not beat the player's own rate is a band
worth distrusting, and it is shown rather than quietly dropped.
Band
Graded
Base rate
Log loss
Own-rate
Shown on board
Of those, landed
Are the probabilities calibrated?
A probability only means something if it comes true that often. Every band is included here,
including the ones below 50% that never reach the board — filtering the evidence to the cases that look good
is how a backtest comes to mean nothing.
Market
Predicted band
Band-games
Average predicted
Actually landed
Gap
Player box scores, rosters, venues and kickoffs from ESPN's public site API; game lines, totals
and officials from SportsDatabase. Bands are round numbers a bookmaker typically posts, not prices we hold:
historical prop prices are not stored, so no return, ROI or edge against a price is claimed anywhere on this page.
A probability above 50% is not value — the break-even at standard −110 pricing is 52.4%, and prop markets are
priced wider than that. Informational only, not betting advice · 18+ · Gamble responsibly.
Do these NFL prop models beat each player's own base rate?
In 5 of 5 markets, measured over 114,154 graded band-games and
replayed walk-forward so no model saw the week it was predicting. The baseline is each player's own rate of
clearing that exact band, over exactly the games being graded — not 50%, and not a position average.
NFL player prop markets: each fitted model against the player's own rate of clearing the same band, scored walk-forward. Lower log loss is better.
Market
Band-games graded
Log loss
Own-rate baseline
AUC
Beats baseline
Anytime touchdown
13,890
0.3271
0.3485
0.787
yes
Rushing yards
15,948
0.3495
0.3890
0.877
yes
Receiving yards
42,700
0.3749
0.4068
0.855
yes
Receptions
36,600
0.3948
0.4353
0.866
yes
Passing yards
5,016
0.5601
0.5741
0.770
yes
What bands are priced?
Round numbers a bookmaker typically posts, chosen to straddle the measured distribution:
What happened to the props the board actually showed?
The board shows only bands priced above 50%. Selecting those naturally picks the easier bands — a lead
back clearing 25 rushing yards is close to a formality — so a high hit rate in that slice is mostly
arithmetic rather than skill. The honest figure is the last column: what those selections landed, minus what the
players' own base rates already implied for them.
Performance of the bands the board displayed, against what the same selections' own base rates implied.
Market
Bands shown
Landed
Model predicted
Own base rate implied
Attributable to the model
Anytime touchdown
374
52.1%
61.4%
57.6%
-5.5 pts
Rushing yards
2,706
69.8%
70.8%
68.2%
+1.7 pts
Receiving yards
6,953
66.8%
71.7%
68%
-1.2 pts
Receptions
8,516
69%
72.5%
69.1%
-0.1 pts
Passing yards
1,889
63.6%
67.1%
63.2%
+0.4 pts
One result worth stating plainly: on the anytime touchdown and receiving yards and receptions board slice the
model did worse than the base rates of the selections it made. Above 50% is exactly where that market is
weakest, and it is reported here rather than left for a reader to discover.
For the current week, 9,853 bands were priced and
3,195 came out above 50%, of which the
236 strongest reach the board. The filter applies to what the board
displays and never to the measurement above, which counts every band — a backtest that only counts the
cases that looked promising is not a backtest.
Is a prop above 50% a good bet?
No, and this page does not claim one. The break-even strike rate at standard −110 pricing is
52.4% and player prop markets are priced wider than the main lines. No band here is graded against a
price at all: historical prop prices are not stored, so there is no return, ROI or edge figure anywhere on this
page. What is claimed is narrower and checkable — the ladder is better ordered than a base-rate ranking of
the same players.
The weekly board itself — every band above 50% across all five markets, with the volume share, implied
team total and defensive matchup behind each number — is part of Premium. This measurement is not, and
will not be.
NFL player props — touchdowns, rushing, receiving, receptions and passing yards
Five markets, each priced as a ladder of bands rather than a single line. Anytime touchdown is the
flagship, but the yardage and reception markets are where most prop volume sits, and they behave quite
differently: a touchdown is a rare binary event driven by goal-line role, while receptions are a near-linear
function of targets.
Why bands rather than one line?
A bookmaker posts a line and prices both sides of it. Without stored historical prices there is no way to
grade against that line, so this page prices a ladder instead — 25+, 40+, 50+, 60+, 75+ and 100+ rushing
yards, and the equivalent for each other market. The thresholds straddle the measured distribution: across
1,785 running-back games with six or more carries the median rushing total is 50 yards and the 90th percentile
is 105, so the ladder covers the realistic range without a rung that almost never resolves.
Why one model per market instead of one per band?
Treating "60+ rushing yards" as its own market and fitting it separately throws away most of the data and
leaves the extreme rungs fitted on a few hundred rows. Instead each market gets a single fit in which the band
itself is an input: how far the threshold sits above that player's own recent median, scaled by his own spread.
One model then prices every rung, every rung informs the fit, and the probabilities down a player's ladder stay
monotonic — which independent fits do not guarantee and a reader would rightly expect.
What does "only bands above 50%" actually filter?
Every band is priced and every band is scored. The display shows only those the model puts above an
even chance, which on a typical week is around a quarter of the roughly four thousand bands priced. The filter
is never applied to the measurement: the accuracy tables above include every rung, because a backtest that
only counts the cases that looked promising is not a backtest.
Is a band above 50% a good bet?
Not on its own, and this page does not say so. The break-even strike rate at standard −110 pricing is
52.4%, and player prop markets are priced wider than the main lines. More importantly, selecting bands
above 50% selects the easy bands — a lead back clearing 25 rushing yards is close to a formality — so
the high hit rate in that slice is mostly arithmetic. The honest figure is the gap between what those
selections landed and what the players' own base rates already implied, and that gap is small. It is reported
above, per market, including the market where it is negative.
Frequently asked questions
Which NFL player props have the best chance of hitting this week?
The board prices every band in five markets — anytime touchdown, rushing yards, receiving yards, receptions and passing yards — and shows only those a fitted model puts above a 50% chance. On a typical week about a quarter of the roughly four thousand bands priced clear that bar.
What bands are used for NFL yardage props?
Rushing yards at 25+, 40+, 50+, 60+, 75+ and 100+; receiving yards at 20+ through 100+; receptions at 2+ through 7+; passing yards at 175+ through 300+. The thresholds straddle the measured distribution — across 1,785 running-back games with six or more carries the median rushing total is 50 yards and the 90th percentile is 105.
Does a prop above 50% mean it is a good bet?
No. The break-even strike rate at standard -110 pricing is 52.4%, and player prop markets are priced wider than main lines. Selecting bands above 50% also selects the easier bands, so the high hit rate in that slice is mostly arithmetic. The meaningful figure is the gap between what those selections landed and what the players' own base rates already implied, which is small and is reported per market.
How is an NFL prop model measured honestly?
Walk-forward, so the model only ever reads games played before the one it is predicting, and against a matched baseline — the player's own rate of clearing that exact band over exactly the games being graded, rather than 50% or a position average. The calibration table includes every band, not only the ones displayed.