Octametrics

The MMA stats most places give you are landed and attempted. That's it. OctaMetrics: weight-class percentile rankings, style and performance archetypes, ELO with projection bands, an ML prediction model that shows its work. octametrics.net/?utm_source=x&…

Allen doesn't shoot often — 13th percentile attempt rate in FW. But when he does, he converts at 74th pct accuracy. One takedown into Costa's 24th pct sub defence is a finish waiting to happen. Model's 17% submission probability looks underpriced if this hits the mat.

Allen's path: Never KD'd in 14 UFC fights (100th pct chin). Sub success rate 86th pct — Costa's sub defence 24th. If Allen converts a takedown, the submission is there. Model calls KO/TKO (42%) but submission probability is 17%.

Allen vs Costa — Saturday's FW main event. The dog case is out there. Here's what the data behind it looks like, and what it misses. 🧵 octametrics.net/?utm_source=tw…

Fair challenge. There's a disclaimer on the Robberies page itself - xF measures metrics such as striking damage, knockdown risk and submission pressure, so grapple / control heavy wins can surface as false positives. It's a discovery tool, not a final verdict. Our models 75.7% accuracy on decisions sits nearly 20 percentage points behind the 94%+ on finishes. Some of that gap is judging noise, some is model limitations as mentioned above - we're always working on improving.

@OctaMetricsHQ @BlindBearMedia I had a look on your site and there are a ton of 1 sided beatings that your model claim were robberies. Ige vs Landwehr, Jones vs OSP, Merab vs Yan 2, Garbrandt vs Jones, JDM vs Belal, USMAN VS MEEK???? Your model needs tweaking.

Arman’s resume is kinda awful besides his win over Charles Oliveira It can be argued he didn’t even beat Charles too

xF (Expected Finish) is our per-round ML model trained on 8,700+ UFC fights, it calculates each fighter's probability of finishing that round based on actual striking output, takedowns, and fight control. Judges watch; xF measures. Our Robberies panel stacks those round-by-round xF scores to assign a "true" winner for every round, independent of the judges. When the scorecards flip the result, it flags it as a potential robbery. Charles edged Arman on xF in R1 and R3, hence 2-1. On accuracy: xF calls the winner correctly in 85.4% of all fights (8,500+ bouts). Where it's most confident is finishes — 94.3% on KO/TKOs and 94.4% on Submissions. Decisions are the noisy ones at 75.7%, which is exactly where robberies hide.

@BlindBearMedia Our 'Robberies' panel has Charles winning that fight against Arman 2 rounds to 1..

What we're investigating: – Weight Late Round Output higher in 5-round fights – Surface 90th-pct+ defensive percentiles in top 3 regardless of magnitude – Re-balance Finishing Ability vs Late Round Output when finishing rates diverge Updates after the next retrain.

3/ Miller–Gordon (Miller by submission, R1). This one's different. The model DID surface Miller's Finishing Ability edge (+0.041) — but still picked Gordon 67% because Late Round Output favoured the younger fighter. A veteran finisher inside R1 shouldn't be a tail event.

Three things UFC 328 taught us about what our model needs to fix. Each one names a specific factor we're going to weight differently. 🧵 octametrics.net/?utm_source=tw…

Percentile data on high level metrics shows they map quite closely to one another. Dig deeper and you'll see Khabib landed ~2x the ground strikes of Khamzat. No doubt both showed dominance but Khabib wins here too 1.34 to Khamzat's 0.9 on our xF model. Interestingly Khamzat's xF has dropped since facing better opposition.

the difference between Khamzat Chimaev and Khabib’s GNP is STAGGERING

@HueyPicksMMA The volume story holds under opponent adjustment. Career xSS (strikes above our model expectation, adjusted for pace and opposition): — Costa +24 (+3/fight) — Allen −5 (−0.4/fight) The gap is real — not an artefact of easy opposition.

@OpenNoteGrapple Just one look at his pizza chart on our site shows Ilia is anything but 'just a striker'. Elite grappling offensively and defensively with heavy hands if you dare to strike.

Ilia’s most common method of victory is submission lol

@finz_the Our percentile data agrees. Pressure percentile 88th and Activity Level 93rd! Takedown Volume coming in at 81st percentile and Take Defense at 75th. Whilst 73rd percentile KO Power doesn't sound too high the Knockdowns/15mins at 79th show the volume of strikes at 85th do work!

[-165] Daniel Santos might be the most SLEPT ON fighter in the UFC: 1. Chute Box Guy a. INSANE pressure b. BIG knees c. BIG elbows 2. Good Grappling/Anti-Grappler a. 12 TDs landed in 5 fights b. 73% TD Defense 3. BIG Power a. Slept Joo Sang Yoo with 1 punch #UFCVegas117

@DovySimuMMA His numbers are impressive whichever way you look at them. 100th percentile for Sig Strikes/min, 98th for Striking Accuracy. High volume and accuracy at all targets. Elite at distance and in the clinch with 88th percentile Knockdowns/15 mins.

Bro what?😲😲😲

@theHOFPicks Our model has it close as well but favouring Strickland at 62% confidence. Method: Decision at 43%

📊Despite Strickland’s dominant victory, odds remain close….

@MikesMMAPicks Our data agrees — every 5 years of age drops win rate ~5% and title winners cluster tightly at 25–34 The canonical "vet lesson" in our database is Randy Couture KO'ing Gabriel Gonzaga at 44 (15.9 year age gap) Modafferi over Barber, Glover over Cutelaba are the modern equivalents

MMA is typically a young man’s sport, though there are times when the superstar kid gets taught a vet lesson. Share the greatest example of a “vet lesson” you’ve ever seen.