| name | bet |
| description | Use when generating fresh recommendations or viewing the current ones. Runs the decide layer (Kelly + portfolio + calibration) and shows the latest recommendations. |
| argument-hint | [--league LEAGUE] [show] |
/bet — Generate or view betting recommendations
Two modes: show the current recommendations, or run the decide layer
to generate fresh ones. Bet placement lives in /place-bets; this skill
stops short of opening a browser.
Mode 1: Show recommendations (default, or show)
Display the latest recommendations from data/decisions/recommendations/.
The Parquet store is partitioned by sport, country, and run_date —
the most recent run_date per league is what the placer would consume.
cd /Users/brynjolfurjonsson/sports && Rscript -e '
suppressPackageStartupMessages(devtools::load_all())
sports::rebuild_duckdb()
con <- DBI::dbConnect(duckdb::duckdb(), "sports.duckdb", read_only = TRUE)
print(DBI::dbGetQuery(con, "
SELECT sport, country, sex, match_date, home_team, away_team,
market, outcome, line, p, odds, ev, kelly, bet_amount
FROM recommendations
WHERE run_date = (SELECT MAX(run_date) FROM recommendations)
ORDER BY ev DESC
"))
'
Equivalently, the placer's preview script shows the same set, deduped against
the ledger (i.e. what would actually be placed):
Rscript scripts/preview_bets.R
Present the table to the user. Then suggest:
- "Want me to refresh recommendations? (
/bet --all or /sports-update)"
- "Want me to actually place these? (
/place-bets)"
If recommendations/ is empty or stale, run the decide layer first
(Mode 2 below).
Mode 2: Run the decide layer
Generate fresh recommendations. The decide layer reads the latest
beliefs/latest/ snapshot for each (sport, country, sex), prepares odds,
runs joint Kelly + portfolio optimisation + calibration, and writes
candidates + recommendations Parquet.
Decide is fast (~seconds) and inline-safe — no detached launch needed.
Rscript scripts/04_decide.R
Or for a single league / sex:
Rscript scripts/04_decide.R --league football_iceland
Rscript scripts/04_decide.R --league football_iceland --sex male
After completion, drop into Mode 1 to display the new recommendations.
When to use which entrypoint
| User intent | Entrypoint |
|---|
| "Recompute recommendations against fresh odds" | scripts/04_decide.R (assumes fits exist) |
| "Recompute everything end-to-end" | /sports-update |
| "Just refresh the model fits" | /sports-update fit |
| "Place bets I see in the table" | /place-bets |
If the latest fit is older than betting.max_age_hours (config/leagues.yml),
decide will warn — refit first via /sports-update fit.
Error handling
- "No recommendations rows for run_date X" → the decide layer hasn't run
for any league yet, or all runs are filtered out by EV threshold. Run
Mode 2.
- "No posterior found" → fit is missing for that league × sex. Run
/sports-update fit --league {key}.
- "Stale posterior warning" → fit is older than
max_age_hours. Refit
first.
- "No odds rows for date range" → the odds scrape hasn't run recently.
Run
Rscript scripts/02_scrape_odds.R.
Reference
- Decide entrypoint:
scripts/04_decide.R → R/decide-pipeline.R::decide_league()
- Sub-stages:
R/decide-{odds,kelly,portfolio,calibration}.R
- Recommendations Parquet:
data/decisions/recommendations/sport=*/country=*/run_date=*/
- Candidates Parquet (with stage column for debugging):
data/decisions/candidates/
- Per-league betting config:
config/leagues.yml::{league}.betting
- Global Kelly cap + budget:
config/bankroll.yml
- Placement (separate skill):
/place-bets
