about
why this exists
we built this because vibes were losing us leagues.
not metaphorically. literally sitting in draft rooms in 2023 taking guys because they "just look like a WR1" while the production data said WR3. taking aging veterans at peak-career prices. ignoring injury return uncertainty because "he looked good in practice." we were doing this every year and wondering why our win rates were mediocre.
the model is the correction.
what it is
regression incoming is a quant nfl fantasy model. it runs on eight seasons of historical data, 99 features per player, and a walk-forward cross-validation harness that stops us from lying to ourselves about how good the projections are.
the outputs are confidence-scored: every projection comes with a number between 0 and 1 that tells you how much to trust it. a guy with a 0.30 confidence score is a range estimate. a guy with a 0.92 confidence score is a point estimate. we built it this way because false precision is how you finish last.
the injury-return pathway adds 58 players to the 2026 rankings who would otherwise be mispriced. the rookie model runs in two stages: draft capital prediction before the draft, landing-spot-adjusted fantasy projection after it.
we log every prediction. we check the results in January. the model's track record is the only thing that matters.
what it is not
it is not a guarantee. it is not financial advice (it is fantasy football). it is not going to tell you who to start week 6 based on "momentum."
the model is a tool that replaces vibes with numbers. you still have to make decisions. the numbers just mean you're making decisions with better priors.
the x402 api
we put the rankings behind an x402 API because the access model matches how the model works: pay for what you use, no subscription, no friction. $0.10 per call. $3 for draft week unlimited. AI agents can pay and pull without registering an account.
if you're an agent: discovery manifest, OpenAPI spec.
if you're a human: the API docs have everything you need.
follow the updates
- X / Twitter: weekly model takes, rankings debates, the calls we got wrong — @regression_inc
- Substack: long-form methodology and analysis — regincoming.substack.com
come argue with our rankings in the replies. we log every W and L in January.