Get the data

Every number on this site is released as data: plays, teams, players and games, updated each week of the season.

Where

The release lives on GitHub: github.com/davisj2007/nfl_case_data . Every season from 2006 to the present has its own folder, in gzipped CSV and in Parquet — the same contents. Completed seasons are scored with the current model; the current season is updated weekly.

What is in it

file	one row per
`case_plays`	play: CASE, its two parts, win probability, leverage
`team_game, team_season`	team and game, or team and season, with EPA, yards and points
`team_ledger_season, team_ledger_game`	team: produced, allowed and net
`team_ledger_by_unit`	team and unit: passing, rushing, special teams, penalty, clock
`player_season, player_game`	player view (passing, rushing, receiving)
`game_summary`	game: excitement, tension, lowest win probability, lead changes
`game_timeline`	play, in each game's path
`implied_wins`	team and second of game time

Every column is documented in DICTIONARY.md , generated from the data itself.

Load it

In R:

      library(readr)
plays <- read_csv(
  "https://raw.githubusercontent.com/davisj2007/nfl_case_data/main/data/2026/case_plays_2026.csv.gz")

In Python:

      import pandas as pd
plays = pd.read_csv(
    "https://raw.githubusercontent.com/davisj2007/nfl_case_data/main/data/2026/case_plays_2026.csv.gz")

Before you publish

Read CAVEATS.md first. Three matter most:

Player views are tabulations, not allocations: they do not add up to the team.
Every figure carries an as_of stamp. Early in a season, a season table is a few games.
CASE describes what happened, in context. It is not a forecast.

License and credit

Released under Creative Commons Attribution 4.0 : share and adapt it, including commercially, with credit. CASE is built on play-by-play data from nflverse , also CC BY 4.0 — please credit both.

Cite as: Davis, J. ( 2026 ). CASE: Context Aware Scoring Expectation . Jake Davis Analytics. case.jakedavisanalytics.com