Replay · NFL

Test a rule against real closing prices.

Each rule here was replayed against every settled line it would have bet, at the closing price, with how sure the result is. Choose one to read it.

Rules replayed

4 more listed that the engine cannot run

Seasons replayed

1999–2024 · 2025 held out

Priced at

all 4 replayed rules · consensus close · a choice in the builder

00Build a rule

Six groups, the common choice open and the rest one click away. Every control is a field the engine takes; nothing on this page runs a rule.

1What you’re bettingProp · under · main line

2When it’s pricedat close · consensus close

3How muchflat 1u

Methodsizing changes what bets are worth, not which are made

Flat% of bankrollFractional Kelly

Unit size

units a bet

4Limits and stopsplayer-week ≤1

Max bets per game

Max per player per weekthe engine's default is 1

Max bets per week

No stops: a drawdown stop, a losing-weeks stop, a per-season cap, one leg per game and a cap on live exposure are not fields of lab.strategy/1, so the engine cannot apply them and none is offered.

5What must be true2023–2024 · 0 conditions

Seasons2025 is held out and cannot be picked

to2025 · held out

Seasonregular season or postseason

RegularPostseason

Weeks

Venuethe catalogue's game.home

AnyHomeAway

Roofthe catalogue's game.dome

AnyOpen airDome

Conditionseach a figure stored as of before kickoff

The feature picker: coming once the Lab export publishes its catalogue, lab/nfl/catalogue.json. Until then a condition is typed as its feature key.

6How it’s judged2,000 resamples · 1,000 draws

Bootstrapwhat the ROI interval resamples

Whole weeks · the engine’s one method

Bootstrap resamples100 to 2,000; more steadies the interval and does not move the result

5001,0002,000

Random draws for the luck check100 to 1,000 random rules of the same size

1005001,000

Settlement

A push is refunded · no snap voids the bet · the engine’s only rules

Reset to the rule shown

Your rule, read back

Bet the under on every player prop market, the main line, priced at the close at the consensus of draftkings, fanduel, betmgm, de-vigged multiplicatively, paid at each book's posted juice, −110 where none is posted, over 2023–2024, regular-season weeks 1–18; 2025 is held out.

Staked 1 unit a bet, flat. Limits: 1 a player a week. A push is refunded; a player with no snap voids the bet. Judged on 2,000 resamples of whole weeks and 1,000 random rules of the same size.

The spec, as the engine would be handed it

{
 "schema": "lab.strategy/1",
 "sport": "nfl",
 "bet_type": "prop",
 "markets": [
  "receptions",
  "receiving_yards",
  "rush_attempts",
  "tackles_assists",
  "sacks"
 ],
 "seasons": {
  "from": 2023,
  "to": 2024,
  "season_type": "REG",
  "weeks": {
   "from": 1,
   "to": 18
  }
 },
 "side": "under",
 "conditions": [],
 "line_choice": "main",
 "price": {
  "source": "consensus_close",
  "books": [
   "draftkings",
   "fanduel",
   "betmgm"
  ],
  "devig": "multiplicative",
  "use_posted_juice": true,
  "assume_juice_if_missing": -110
 },
 "timing": "close",
 "staking": {
  "method": "flat",
  "unit": 1
 },
 "limits": {
  "per_game": null,
  "per_player_week": 1,
  "per_week": null
 },
 "settlement": {
  "push": "refund",
  "no_snap": "void"
 },
 "evaluation": {
  "bootstrap": "by_week",
  "resamples": 2000,
  "null_draws": 1000
 },
 "holdout": {
  "season": 2025,
  "revealed": false
 }
}

Before anything runs

13,680 bets across 36 weeks

This is “Fade every over”, already replayed: the count is the file’s, not an estimate.

Read its result →

This rule’s address

Every control is in the address, so the link is the rule. Nothing is saved on this site.

Copy link

Describe a rule in words Not connected yet

When it is connected, it will turn a sentence into a rule you can read and edit before anything runs. Today nothing typed here is read: build the rule with the controls instead.

01Rules

4 rules replayed · 4 more listed that the engine cannot run · Film Room

Loses01Fade every over13,680 bets · −3.3% ROI · open in the builder
Unclear02Rush attempts unders2,256 bets · −1.8% ROI · open in the builder
Loses03Chase the streak1,150 bets · −13.2% ROI · open in the builder
Cannot run04Follow the model's leansWaiting for walk-forward model predictions persisted for the historical seasons.
Cannot run05Buy the middleWaiting for a rule language with two legs: lab.strategy/1 bets one side at one price.
Cannot run06Arbitrage between booksWaiting for a rule language with two legs: lab.strategy/1 bets one side at one price.
Loses07Home underdogs of 3 to 71,400 bets · −5.6% ROI · open in the builder
Cannot run08Unders in the windWaiting for forecast weather joined to the replayed games.

4 rules replayed on this site so far. The more rules are tried, the more likely one clears zero by chance alone: with 4 tried, a 95% interval that allows for it spans about ±2.50 standard errors (Bonferroni), not ±1.96. The intervals on this page are the plain 95%.

Arbitrage between books

Listed · not replayed

Waiting for a rule language with two legs: lab.strategy/1 bets one side at one price.

The engine cannot run this rule, and says why:

not expressible in lab.strategy/1: the same two-leg shape as the middle

Measured elsewhere: R16: receptions arbitrage mean size +1.09pp.

Film Room lists every rule, replayed or not.

Hypothetical results · nothing here was bet · not advice · lab index written Sun, Sep 27, 2:14 AM ET