Modeling · Applied statistics · Machine learning

Measure uncertainty. Manage risk.

Build and employ apps and methodologies to deliver solutions and solve problems.

Explore the products Sample output

Products

Live beta v1

FanFav

Weekly calibrated NFL projections, conformal intervals, and lineup optimization.

By invite only (for now)

Open app See sample output

Live beta v1

Model Ledger

In-season modeling and backtesting surfaces for NFL models.

By invite only (for now)

Open app See sample output

Research

Research in progress

Inference workbench

Regression and Bayesian methods.

Research future

Cliodynamics

Quantitative modeling in theoretical history.

Sample output

Real output from the models, 2025 NFL season. Static snapshot.

Weekly player projections

2025 · Week 20 · DK

Player	Pos	Team	Opp	Salary	Pred	70% interval	Value
Puka Nacua	WR	LAR	CHI	$9,000	23.1	13.6 – 32.7	2.6
Christian McCaffrey	RB	SF	SEA	$8,800	22.6	12.8 – 32.3	2.6
Drake Maye	QB	NE	HOU	$6,300	20.3	9.8 – 30.8	3.2
Matthew Stafford	QB	LAR	CHI	$6,600	19.3	11.4 – 27.2	2.9
Jaxon Smith-Njigba	WR	SEA	SF	$8,500	19.3	9.7 – 28.8	2.3

Every projection carries a 70% interval. Value is points per thousand of salary.

Interval honesty

holding

A 70% interval should contain the real result 70% of the time. Across the 2025 mid-season weeks it averaged 0.60.

Reported openly — 13 mid-season weeks, averaging about 10 points below the promise.

Where the model leans

watch

Season-average gap between prediction and result, per player. Negative means the model under-predicted — these five it was most cautious about.

Puka Nacua WR · LAR−4.8
DJ Moore WR · CHI−3.3
Colston Loveland TE · CHI−2.9
Drake Maye QB · NE−2.6
Christian McCaffrey RB · SF−2.6

545 players tracked · 64 lean-in · 215 fade. The model over-predicts more often than not, which is what the watch flag tracks.

The approach

Applied statistics and machine learning methods for problem-solving. Distribution sampling, conformal prediction, regression, and various other statistical techniques.