Gridline — probability-first NFL analytics

A fully automated NFL model that turns play-by-play, usage, game environments and weather into calibrated outcome probabilities — then ranks every prop, moneyline lean and DFS play by the probability itself, not by a marketing score.

Launch App → Request access

Access is per-email. The app is private; this page is public.

PROBABILITY LEADERBOARDillustrative layout

Player A · REC 4.5 · ▲ OVER0.71

Player B · REC YDS 62.5 · ▼ UNDER0.66

Player C · REC 5.5 · ▲ OVER0.63

Player D · REC YDS 48.5 · ▼ UNDER0.61

Every field carries its provenance: measured · derived · unavailable. Missing inputs make a prop non-bettable — they are never filled in.

How it works

Rigorous where it counts, honest about what it is: a model, validated against real outcomes on an ongoing basis.

1 · Automated data pipeline. Every refresh pulls play-by-play, weekly usage and snap counts, depth charts and injury reports, schedules, Vegas-derived game environments (spreads, totals, implied team totals decomposed into projected plays and pass attempts) and stadium-keyed weather. Every feed is time-stamped; stale data is flagged, never silently reused.

2 · Opportunity-first projection engine. A shared opportunity model (targets, routes, snap share, role confirmation, blended prior-season and current-season form with partial-game exclusion) feeds each market. Environment, opponent and weather enter as elasticities on the opportunity anchor — no hand-tuned "score" arithmetic.

3 · Outcome distributions, not point estimates. Receptions follow a negative-binomial distribution and receiving yards a gamma distribution, fitted with dispersion parameters. Each posted line becomes a probability of OVER and of UNDER from the full distribution.

4 · A probability leaderboard. Props are ranked by max(P(over), P(under)). Posted odds never gate a pick; closing lines are logged passively for later review. Moneyline leans use an external, clearly labelled model and are never blended into the engine.

5 · Provenance on every field. Each input is tagged measured, derived or unavailable. A prop is bettable only when its core inputs are all measured; otherwise it is shown as non-bettable with the reasons listed.

6 · Calibration-first validation. Every projection is logged before kickoff and settled against real box scores. Claimed probabilities are compared with realized hit rates by band; bands without enough graded picks are marked insufficient rather than reported.

Probabilities are validated against real outcomes on an ongoing basis. Until enough graded results accrue, the app labels its output as unvalidated. Gridline makes no performance claims and promises no profit.

Features

One private command center, phone-friendly, always on.

Probability board

Receptions and receiving-yards props ranked by probability, with the pick direction, the tier, the flags and a WHY view showing every contributing factor and its provenance.

Moneyline leans

Per-game win probabilities from nfelo (an external QB-adjusted Elo model, labelled as such), ranked by lean strength, alongside market lines, projected pass volume and weather.

DFS projections

DraftKings-points projections with floor and ceiling from Monte-Carlo simulation, a value board, one dominant tag per player and stack cards with a Core-4 summary.

Research-tuned 50-lineup optimizer

Correlated sampling from the model's own distributions, research-backed construction defaults (salary dispersion, stacking, bring-backs, exposure caps) and a readout of every target actually hit.

One-click DraftKings export

Lineups export straight to DraftKings' bulk-upload CSV, with a readable companion file. Files are never overwritten.

Live inactives checks

A game-day poller watches inactives and roof status before kickoff, alerts on anything that touches your board or lineups, and rebuilds a corrected set on request.

Request access

The app is private and login-protected. Tell us who you are and we will add your email.

Access is granted per-email by the operator.