CageSide
DashboardEventsOur PicksCommunityFightersFight PredictorSign in

Disclaimer: CageSide is a research tool, not a sportsbook. Picks and "BETs" are statistical model outputs for educational purposes — not gambling advice and not a guarantee of any outcome. Past model performance does not predict future results. Sports betting carries a substantial risk of loss; only wager what you can afford to lose. You must be 18+ (21+ in some U.S. states) to use this site.

Problem with gambling? Call the U.S. National Council on Problem Gambling: 1-800-GAMBLER (24/7, confidential, free).

© 2026 CageSideAboutTermsPrivacyContact@CageSideApp
  • Home
  • Picks
  • Events
  • Feed
  • You

About CageSide

What this is

CageSide is a UFC research tool. We run a statistical model on every upcoming UFC fight and publish its picks across four betting markets — moneyline, fight goes to decision, over 1½ rounds, and over 2½ rounds. We do not take bets. Nothing here is gambling advice. The picks are research, paper-traded so we (and you) can see whether the model actually finds edges over time.

The thesis

Sportsbooks set sharp lines on main events because that's where the money is. They set less sharp lines on prelims and undercards — especially for fighters making their UFC debut or recently signed from Dana White's Contender Series, LFA, Cage Warriors, OKTAGON, ACA, and other regional promotions. Information asymmetry. Our model is trained on every UFC fight back to 2010, plus pre-UFC career data scraped from Sherdog, with the goal of pricing those underexposed fights more accurately than the books do.

What the model does

  • Five separate models, one per market. Each is gradient-boosted with isotonic recalibration. Trained on time-respecting walk-forward splits to avoid leak.
  • Feature set: per-fight rolling stats, fight outcomes, Elo-style ratings, age curves, layoff penalties, opponent style archetypes, pre-UFC promotion history, recent finish rate, "war" indicators (back-to-back five-round decisions), and per-round cardio signals from UFCStats.
  • Empirical-Bayes shrinkage on every rate stat so fighters with 2 UFC bouts don't get scored as if they have 20.
  • BET rules per market — minimum edge thresholds, model confidence caps, and dead-division exclusions — are all derived from backtested ROI, not gut feel. When a rule changed because the backtest said so, we noted it in the code comments.

What the model does NOT do

  • It does not predict the future. It estimates probabilities. Variance does the rest.
  • It does not know about overnight news — injury reports, weight cut issues, corner changes.
  • It does not place bets, manage your bankroll, or judge whether sports betting is right for you. That's on you.
  • It does not always agree with the books. When it does agree, the edge is small. When it disagrees, the edge is larger but the model is more often wrong on those same fights. That's the trade.

The community side

Sign in, drop your own picks on any fight, comment on matchups, post takes. The leaderboard tracks confidence-weighted scoring across everyone who plays. The community feed shows the most recent picks + comments + posts site-wide.

Built by one person

CageSide is a one-developer project. Bug reports, feature requests, and feedback welcome at hello@cagesidepicks.com.

Sources of truth

  • Fight data: UFCStats
  • Pre-UFC records: Sherdog
  • Betting lines: BestFightOdds + The Odds API
Sign up free

18+ only (21+ in some U.S. states). See Terms + Privacy.