LAIV Edge · NFL research data

Every prop line we hold, and what happened

Four seasons of NFL player-prop odds joined to the actual outcome, plus the picks LAIV published. Click any market to explore it.

Ask AI

Ask anything about this data

Claude can query the 1,997 lines from 2026 and the four-season aggregates directly. It answers from the data, not from memory. Runs on LAIV's own key against the full database, so it costs you nothing.

Set once, remembered on this device.

Pick a market

19 markets, 2023 to 2026

The bar shows how much of each market we can grade. Grey means we hold the odds but not the outcome yet.

Market

Hit rate by season

By position

Hit rate by line, low to high

What price shopping is worth here

Every line, 2026

Week by week, player by player

LAIV picks only

What LAIV published in 2026

Three weeks of a live product. Small samples, shown for shape rather than conclusions.

Four traps that will produce confident nonsense

Alternate lines are in the data. One player and market can carry six lines in a week. Filter is_main_line. Deduplicating by best price silently selects long-shot alts and yields impossible backtests.
Week 18 is a rest week. Playoff-bound teams sit starters. Any trailing window reaching into it reads a rest day as a role change.
Six flexed games from 2023-25 were priced mid-game. Those are not pre-game odds. Exclude them.
Anytime TD composition shifts by year. Median Yes price was +2000 in 2023 and +493 in 2025, so the rising scoring rate is the cache getting less longshot-heavy, not scoring changing.

The Rams were absent from 2023-25 odds until recently and are now backfilled. Any LAIV result dated before 27 September 2026 was computed without them.

Going deeper than this page

Everything above is pre-aggregated. The full database is one DuckDB file, 40 MB, no server and no credentials. Ask Vivaan for laiv_football.duckdb, then:

pip install duckdb

import duckdb
con = duckdb.connect("laiv_football.duckdb")
con.execute("SHOW TABLES").df()

# grade any market yourself
con.execute("""
  SELECT season, count(*) n,
         round(100.0*avg(CASE WHEN actual > line THEN 1 ELSE 0 END),1) AS over_pct
  FROM graded
  WHERE market='player_rush_yds' AND is_main_line AND actual IS NOT NULL
  GROUP BY 1 ORDER BY 1
""").df()

quotes 1,335,600 book-level rows · consensus 295,695 with median, best and worst price · graded the view above · player_stats 150 nflverse columns · defense_games snap share and book tackles · games closing spread, total and wind · laiv_picks our published board. Works from Python, R or the command line.

Built 27 September 2026 from the production odds cache and nflverse.