£1-Stake Bankroll MODEL

Pick Accuracy MODEL

Hit-rate by confidence tier

Event-by-event MODEL

No settled events yet — the bankroll starts the first time a scored card gets a final result.

How to read the source tags

Every section of the dashboard is tagged with one of four labels so you always know whether you're looking at raw data, an AI prediction, or maths derived from inputs.

How the model works MODEL

It's a pattern-matcher (an XGBoost model) trained on 8,536 past UFC fights — not an opinion, and it doesn't watch tape. It learned which stat differences tend to go with a win, then applies that to this week's fighters to give each a win probability.

• What it sees: each fighter's career W/L, finish-rate split (KO/SUB/DEC), striking volume & accuracy, takedowns per 15 min, control time, reach, height, age and UFC experience — plus the gap between the two fighters on each.
• How accurate: validation AUC 0.700 — on a fight it hasn't seen, it ranks the real winner above the loser 70% of the time. A coin-flip is 0.50; in an upset-heavy sport anything above 0.65 is genuinely predictive. Useful, not certain — roughly 3 in 10 confident-looking picks still lose.
• Missing stats: debutants and fighters with gaps still get a number, but it's a guess on thin information — that's exactly what the grey confidence band is warning you about.

What it can't see MODEL

• Style match-ups — it only sees each fighter's own averages, never how they interact, so "styles make fights" is invisible to it.
• Context — weight cuts, injuries, camp changes, short-notice fights, layoffs and motivation are all unseen.
• Debut / grey fighters — thin history means a low-confidence guess. Treat grey picks as "no read".
• Stale-odds traps — a huge "value" edge (40%+) usually means the odds are wrong or the sample is tiny, not free money. Trust a green +5–15% edge over a grey +80% edge.

Confidence Bands HEURISTIC

Heuristic rules layered ON TOP of the model's raw probability to flag how much you should trust it.

• Green: model prob ≥ 60% AND both fighters have ≥5 UFC fights AND no missing stats. Trust the read.
• Amber: anywhere between green and red thresholds. Most fights land here — there's a view but it's not screaming. Look at the drivers.
• Red: 45-55% toss-up. Pick another fight.
• Grey: debutant fighter or critical stats missing. Model can't see this one — skip.

Top Drivers MODEL

The three diff features with the largest absolute value in this matchup. The model uses many more features internally, but these are the simplest to read at a glance. Red/blue border on each chip indicates which fighter the driver favors.

• Reach (inches diff) — longer reach generally favors the rangier fighter, especially in striking-heavy matchups.
• Age (years diff) — younger fighter has the advantage in most cases; reversed for fighters in their athletic peak (28-32).
• Striking output (SLpM) — significant strikes landed per minute. Higher volume = more activity but not necessarily more accuracy.
• Striking accuracy % — strikes landed / strikes attempted. A precision signal.
• Takedowns per 15min — grappling volume. Strong predictor for control-time fighters.
• UFC experience (fights diff) — adapting to UFC-level competition matters.

Method of Victory HEURISTIC

Not ML — a simple split of the win probability across finish types. If A is a KO artist and B has been knocked out before, A's KO share ticks up; same logic for submissions and decisions. The six bars (three per fighter) add up to 100%.

Round Props (Over X.5 Rounds) HEURISTIC

Built from both fighters' historical average fight length and finish rates. A fight between two decision-heavy grinders will show high Over 2.5 / Over 4.5 probabilities; two finishers in the same cage will show high Under.

• Combined avg fight minutes derived from both fighters' last-5 control time + finish round history.
• Adjusted by finish-rate weighting (high finishers shift the distribution earlier).
• Over 4.5 only shown for 5-round fights (main events + title fights).

Suggested Bets HEURISTIC

Rule-based, not AI-written. A suggestion only appears when the stats line up for it — up to three per fight.

• KO/TKO — the model gives a strong KO chance and the opponent has been finished by strikes recently.
• Over rounds — both fighters tend to go long and a finish looks unlikely.
• Moneyline value — a high-confidence pick clearly disagrees with the bookmaker's price.
• Border colour shows confidence — green high, amber moderate, red low.

Edge CALC

The expected return of betting at the displayed odds, assuming the model's probability is correct. Live recalculated whenever odds change.

Edge = (model probability × decimal odds) − 1

• Green (≥ 5%): notable value — book is offering more than the model thinks the chance is worth.
• Amber (0–5%): marginal positive expected value. Not enough to override gut.
• Grey (< 0%): negative edge. Book either has it right or favors the other side.
• Big edges (40%+) are usually noise, not signal: amber/grey confidence + small fighter sample = inflated edges. Trust green-confidence + 5-15% edge over grey-confidence + 80% edge.

Bet Auto-Settle DATA

When a fight's result is published, open moneyline bets settle automatically on your next visit — a straight winner check, no AI. Method, round and custom bets stay manual (too many edge cases).

Opinion Auto-Grade DATA

Once a fight settles, your pick is graded on three things separately — winner, method and round — so you can nail the winner but miss the method on the same call.

• PERFECT — winner, method and round all right.
• WINNER — you got the winner.
• MISS — winner wrong.

Live Odds DATA

Pulled from a live UK-books odds feed, with one preferred book picked per fight. Obviously bad quotes are rejected, and you can type your own price into any odds box to override.

Accounts & privacy DATA

• Shared card, shared community — everyone sees the same fights, model picks and live odds, plus the same Opinion Log and Leaderboard.
• Your bets stay private — the Bet Tracker is tied to your account and yours alone.
• Guest mode — browse without an account; your picks just won't be saved.

Not yet

• No in-fight odds refresh — odds are set when the card is built; type your own price to override.
• No written fight breakdown — the model gives numbers, not paragraphs (auto-grading still tells you Won/Lost).
• No best-price shopping — one preferred book per fight, not a comparison across books.

Effective 26 May 2026 · Last updated 26 May 2026

Privacy Policy

Who runs this site

MMA Picks is a free amateur dashboard operated as part of The Mono Hub. Contact: contact@themonohub.com. We are not a registered company and not regulated as a betting operator — we do not take real-money bets.

What we collect

• Account data: email address, username, hashed password. Stored by Supabase (EU North region, Stockholm).
• Picks & bets you log: the fighters, methods, rounds, notes, stakes, odds you enter. Opinions are public to other logged-in users; bets are private to you.
• Odds overrides: any bookmaker odds you type in. Shared across all signed-in users (it's a community resource).
• Browser localStorage: your latest data is cached locally for offline reads. Cleared on sign-out.
• Server logs: Netlify and Supabase keep standard request logs (IP, timestamp, URL) for security and uptime. We don't read these except to debug.

What we don't collect

• No third-party analytics or tracking pixels at this time.
• No advertising cookies. No marketing trackers.
• No payment information — we don't sell anything.
• We don't share your data with anyone other than the hosting providers above.

Your rights (UK GDPR)

• Access: you can see all your picks and bets directly in the app.
• Export: Bet Tracker and Opinion Log both have CSV export buttons.
• Delete: email us and we'll delete your account + all rows within 7 days. Self-serve delete coming in a future release.
• Withdraw consent: sign out and stop using the site.

Cookies

We use only essential authentication cookies set by Supabase to keep you logged in. No cookie consent banner shown because no non-essential cookies are set.

Terms of Service

What this is

MMA Picks is a free dashboard for tracking pre-fight predictions and bets you place elsewhere. We are not a betting platform — no real money flows through this site. Bets are recorded for your own tracking only.

Your account

• You're responsible for keeping your password secure. Don't share your account.
• One account per person. Multiple accounts to game leaderboards may be removed.
• You must be 18+ to use the site (the activity it relates to — sports betting — is age-restricted in most jurisdictions).
• We can suspend or delete accounts that breach these terms.

Acceptable use

• No hateful, threatening, discriminatory, or illegal content in usernames or notes.
• No impersonation of other users, fighters, or organisations.
• No automated scraping, spamming, or attempting to bypass authentication.

Disclaimers

• Not affiliated with UFC. This is an independent fan project. "UFC", "Ultimate Fighting Championship", and related marks are trademarks of Zuffa LLC / TKO Group Holdings. We do not represent or speak for them.
• Not financial or betting advice. Model probabilities, suggested bets, and consensus picks are informational only. Use of any number here for real-money betting is at your own risk. Gambling can be addictive — if it's a problem call BeGambleAware (UK) or your local equivalent.
• Data provided "as is". Fighter stats are scraped from UFCStats.com. Model predictions are statistical estimates. We don't guarantee accuracy or completeness.

Liability

To the maximum extent allowed by law, MMA Picks and its operator are not liable for losses (financial or otherwise) arising from your use of the site, including bets you place elsewhere based on information here.

Changes

We may update these terms. Material changes will be flagged at the top of this page and noted in the change log. Continued use after updates means you accept the new terms.

Governing law

England & Wales. Any disputes go to the courts of England and Wales.