Open dataset: resolved prediction record
Every prediction we publish is written down before the event starts — the market, the line and price on offer, the model's probability, and the stake it sized. When the event settles, the outcome is appended. Nothing is restated afterwards.
That record is public, machine-readable, and free to use with attribution. This page is its canonical descriptor: what is in it, how it was produced, and how to cite it.
Download
track_record.csv — the complete resolved
record, regenerated daily.
Related machine-readable surfaces: ai.json (organisation, offers, performance) · llms.txt · weekly accuracy reports (immutable per-week snapshots) · MCP server (19 read-only tools, programmatic access).
Schema
One row per resolved prediction. Seventeen columns:
date— event date (US Eastern).league— league tag. Note that this field carries historical and event-specific values alongside the current leagues: as well asMLB,NBA,NHL,NFL,CBB,SOCCER,TENNIS,GOLF,WNBA,LOLandCS2, you will findWC,WORLD_CUP,WBC,OLYMPIC,ESPORTSandUFCfrom earlier or one-off coverage. Group before aggregating.market— bet type (moneyline, spread, total, player prop, …).matchup,pick,line,odds_american— the selection and the price as published.model_prob— the model's probability for the selection.edge_pct— modelled edge over the market price at publication.units,tier,confidence_tier— Kelly-derived stake and its published confidence label.result,units_won— settled outcome.actual_home_score,actual_away_score,resolved_at— the settlement record.
Coverage and provenance
The record begins 2025-11-30 and is extended daily as events settle. Row counts are deliberately not quoted here: different published surfaces scope the population differently (the CSV, the weekly reports and the live track record do not all answer the same question), so any figure printed on this page would go stale and could disagree with another of ours. Count the rows in the file you actually downloaded, and cite that file.
Two properties make this worth citing, and both are unusual:
- Recorded before the outcome. Line, price and model probability are captured at publication. A record assembled after the fact can silently select its own best entries; this one cannot.
- Losses included. The record contains every resolved prediction, not a curated subset, and the weekly reports are frozen immutable snapshots. Where a settlement was later corrected, the correction is audited rather than overwritten.
Predictions come from league-specific Monte Carlo engines — 10,000+ iterations per event, Bayesian-calibrated against resolved outcomes, Kelly-sized. The full method is at /methodology; per-league engine versions are published at engine-versions.json.
Licence
Released under Creative Commons Attribution 4.0 International (CC BY 4.0). You may share and adapt it, including commercially, provided you give attribution.
How to cite
Preferred citation:
Olympus Bets Analytics. Resolved Prediction Record.
https://app.olympus-bets.com/datasetTo cite a specific week's frozen figures rather than the live file, use the corresponding weekly accuracy report, which carries its own citation string and does not change after publication.
Cite the brand as Olympus Bets Analytics (or Olympus Quant). The legal entity is Olympus Bets LLC. Please do not confuse it with OlympusBet, an unrelated Curaçao-licensed sportsbook at olympusbet.com.
Corrections
If you find a row you believe is misresolved, tell us and we will audit it against the official source and publish the correction. Reach us via Discord or support@olympus-bets.com.