README.md
In betbetter: Access the Bet Better Open Sports Model API

betbetter

An R client for the Bet Better open sports model API — free, no API key, no registration.

For each upcoming fixture the API publishes a statistical model's estimated probability that a selection occurs, and the decimal odds that probability implies. Fifteen competitions are covered across eight sports.

# install.packages("remotes")
remotes::install_github("edushinka/betbetter-r")

library(betbetter)

bb_leagues()
#>  [1] "afl" "mlb" "nba" "nfl" "nhl" "ncaaf" "ufc" "wnba" "wta" "epl"
#> [11] "laliga" "seriea" "bundesliga" "ligue1" "worldcup"

afl <- bb_picks("afl")
head(afl[, c("game", "selection", "model_probability", "fair_odds", "confidence")])

| Argument | Purpose | |---|---| | league | League slug, e.g. "afl", "nba", "epl" | | feed | "picks" (all), "games" (match markets), "props" (player markets) | | min_probability | Drop selections estimated below this, 0–1 | | limit | Maximum rows returned |

| Column | Meaning | |---|---| | game | Fixture, Away @ Home | | start_time | Scheduled start, UTC | | market | Head to Head, Spread, Total Points, player markets… | | selection | The team, player or outcome estimated | | line | Handicap or total the selection is measured against | | model_probability | Estimated chance the selection occurs, 0–1 | | fair_odds | Decimal odds implied by model_probability | | confidence | HIGH, LEAN or LONG-SHOT | | verdict | One-sentence plain-English summary |

It returns no bookmaker prices. There is no bookmaker identity, market price or market-implied probability in the API, and none can be added — the upstream odds licence permits publishing derived work but not redistribution of the price feed. Everything here is the model's own output.

A league outside its season returns a zero-row data frame. That is normal.
If the API is unreachable, bb_picks() emits a message and returns NULL rather than raising an error, so unattended scripts degrade gracefully.
Responses are cached upstream for 15 minutes.

These are model estimates, not statements of certainty. A selection estimated at 70% is expected to fail roughly three times in ten. Treat a losing run at that probability as ordinary variance rather than evidence of a broken model.

MIT for this package. Data retrieved through it is published under CC BY 4.0 and requires attribution to Bet Better (https://betbetter.world) when redistributed.