Frequently Asked Questions

75.4% walk-forward accuracy across 3,148 out-of-sample fights over 7 annual windows. That means publishable walk-forward windows train on strictly prior data and test on fights the model had never seen. No cherry-picking, no data leakage. You can verify all of this on our Accuracy and Calibration pages.

Our proprietary rule + Elo predictor combines tuned heuristics informed by 6,871 historical fights with head-to-head Elo ratings that weight recent fights more heavily. The predictor scores striking accuracy, takedown rates, submission attempts, fight pace, reach differentials, style-matchup adjustments, and method probabilities across 149 input signals. Each fight is also simulated 15,000 times via Monte Carlo to generate method-of-victory probabilities and round-by-round projections. The final pick is passed through a walk-forward isotonic calibration layer tracked across 3,148 out-of-sample fights and capped at 0.85 — we never claim 95%+ lock confidence in a sport with real variance.

The projected stats (strikes, takedowns, control time) come from a 15,000-fight Monte Carlo simulation. The winner pick is the calibrated rule + Elo side after data-quality and risk checks. Market odds, expert consensus, and community signals are shown as alignment/risk context rather than being generally blended into the displayed probability. The simulator models volume, positioning, and grappling statistically; the rule + Elo predictor has encoded the non-linear realities of MMA — KO power against a damaged chin, age and form effects, stylistic matchups — that don't show up in a round-by-round stat simulator. When a fighter lands more strikes in the sim but loses the pick, it usually means they absorb more damage from a harder-hitting opponent, or they're an aging fighter on a losing streak against a live dog — both patterns the predictor has thousands of examples from. We show both numbers because prop bettors price takedowns, strikes, and rounds directly off the sim, while moneyline bettors care about the final pick. When the two diverge, the Win CI widens to reflect the honest uncertainty.

The Win CI combines two sources of uncertainty. First, Monte Carlo sampling noise (how much the 15,000-sim win rate could shift with more sims). Second, disagreement between the simulator and the calibrated model — when the sim says 50/50 and the model's other signals say 73/27, the real probability is somewhere in a range, not a pinpoint. The CI reports that range honestly. Tight CIs (a few percentage points) mean every signal agrees. Wide CIs (15–25 pp) mean the model sees conflicting evidence and you should size bets accordingly.

Walk-forward validation tests the model on fights it has never seen during training. We train on all fights before a certain date, then predict fights after that date. This prevents the model from 'memorizing' results and gives an honest accuracy number. Many competitors use backtesting which inflates accuracy.

The rule + Elo predictor isn't a trained ML model — it's a deterministic system with continuously-updating per-fighter state. Elo ratings recompute automatically every time new fight results land in our database (typically within hours of a UFC event). Fighter profile stats refresh live on every prediction. The walk-forward isotonic calibration table refits automatically on the first Sunday of each month if enough new fights have accumulated — but only if the new fit actually beats the old on held-out Brier score. Data pipelines (odds, expert picks, sentiment, rankings) refresh on independent schedules (every 2-24h depending on fight week status).

Lock or Fade is our free community prediction game. Pick a winner for every fight on the upcoming card, and your picks are automatically scored after the event. Compete on the all-time leaderboard and track your accuracy, streak, and pick history. No subscription needed.

The Value Finder compares our model's win probabilities against live sportsbook odds from 7 major books. When the model disagrees with the market (e.g., model says 60% but odds imply 45%), that's a potential value bet. Pro feature.

Each fight is simulated 15,000 times round-by-round, modeling striking exchanges, takedowns, submissions, fatigue, and damage. This produces method-of-victory probabilities (KO/Sub/Decision), round finish distributions, and stat projections. Much richer than a simple win/loss prediction.

The Parlay Builder lets you combine multiple picks into a parlay and calculates combined odds, expected value, and correlated risk. It flags parlays where outcomes are correlated (e.g., two fighters on the same card) and shows you the edge.

A Pro feature that lets you adjust predictions for real-world variables. Select weight miss, short notice replacement, injury (legs/hands/cardio), venue altitude, or referee assignment — and instantly see how the win probability and method breakdown shift. No other UFC prediction tool offers this.

We track two distinct metrics on the Accuracy page: True CLV (Closing Line Value) measures whether the market moved TOWARD our pick after we made it — positive CLV means you got a better price than the close, the sharp-betting gold standard. Model Edge vs Close measures how much our model's confidence disagreed with the market's closing probability on the picked side — positive means we saw value the market didn't. Both are shown separately so you can evaluate the model as a bettor's tool (CLV) and as a forecasting tool (Edge) independently.

Fade signals flag market favorites that the model disagrees with. When a fighter is a heavy favorite (e.g., -200) but the model gives them a significantly lower probability, FightEdge flags it as a fade opportunity. Found on the Value Finder page.

It compares fighters across 11 statistical dimensions (striking volume, accuracy, defense, takedowns, submissions, and more) using cosine similarity. On any fighter's profile page, you'll see the top 5 most similar fighters — useful for understanding how a new matchup might play out based on comparable historical fights.

Yes. New subscribers get 7 days of full Pro access free. Your card is required to start the trial but won't be charged until it ends. Cancel anytime during the trial and you'll never be charged.

Yes. Cancel through the Stripe billing portal from your Profile page. You keep Pro access until the end of your current billing period.

All major credit and debit cards via Stripe. Your card details are never stored on our servers — Stripe handles all payment security (PCI-DSS Level 1 certified).

We offer a 7-day money-back guarantee on your first subscription payment. Contact us via the contact form within 7 days of your charge.

You revert to the Free tier. All your data (bet history, Lock or Fade picks, preferences) is preserved. You can resubscribe anytime to regain Pro access.

Passwords are hashed with PBKDF2-SHA256 (1 million iterations). All traffic is encrypted via HTTPS. Your data is stored on AES-256 encrypted AWS infrastructure. We never store credit card numbers. See our Privacy Policy for full details.

Yes. Go to your Profile page and click 'Delete My Account' in the Danger Zone. This permanently deletes all your data, cancels any active subscription, and cannot be undone.

No. We never sell, rent, or share your personal data with third parties. Period.

Every UFC event — numbered PPVs, Fight Nights, and full cards including prelims. Our model runs predictions on every scheduled bout.

No. FightEdge is not affiliated with, endorsed by, or sponsored by UFC or Zuffa LLC. Fighter names and event data are used for informational purposes under nominative fair use.

No. FightEdge provides statistical analysis for informational and entertainment purposes only. We are not a sportsbook and do not accept or facilitate bets. Always gamble responsibly. See our Disclaimer.

FightEdge provides statistical analysis and data-driven predictions — not guaranteed outcomes. Public benchmark status: 75.4% walk-forward accuracy across 3,148 out-of-sample fights over 7 annual windows. CLV (Closing Line Value) is tracked separately as an edge signal over time. However, sports outcomes are inherently uncertain, individual results vary, and past performance does not guarantee future returns. Always bet responsibly and within your means.

Not yet — FightEdge is currently a web application optimized for both desktop and mobile browsers. A dedicated mobile app is on our roadmap. The web version works great on phones with full responsive design, including all analysis, parlay builder, and community features.

The model performs best on fighters with 5+ UFC fights (more data = better predictions). For debuting fighters or those with fewer than 3 fights, predictions are regressed toward division averages and carry higher uncertainty. The model also doesn't account for undisclosed injuries, mental state, or camp-specific preparation — which is why we built the What-If Scenario Engine for manual adjustments.

The model is most confident in matchups between established fighters with clear stylistic differences (striker vs grappler, power vs volume). Coin-flip matchups between similar fighters naturally have lower confidence. The Calibration page shows exactly how the model performs at each confidence level — the pattern is consistent under-confidence, meaning fighters win even more often than the model predicts at every bucket.