Skip to main content
EducationalNot betting advice · research use only.
GameTime Picks

Bout 11 of 13

Rafa Garcia vs Rongzhu

Sat, Sep 12, 2:00 PM ET · Lightweight · 3 rounds · Noche UFC: Silva vs. Delgado · Desert Diamond Arena

Winner

Who the model expects to win

Probabilities over the two fighters, from the head that was evaluated on its own preregistered bar.

Rafa GarciaRafa Garcia19-4-0
vs
RongzhuRongzhu28-6-0
Model pickRongzhu · 51.7%Decision · round 3 or later
19-4-0Record28-6-0
11Tracked bouts6
48.3%Win chance51.7%

The model separates these two by very little — 52% is close to a coin flip, and nothing in either record breaks the tie cleanly.

Method

How it ends

Read among fights that end with a winner — a draw or no-contest sits outside this sample, so the three add to 100%.

KO / TKO26.1%
Submission13.9%
Decision59.9%
Round

How far it goes

R3+ carries every fight that reaches the judges as well as every round-3-or-later finish.

18
R1
14
R2
67
R3+
Round 1
18.3%
Round 2
14.4%
Round 3 or later
67.3%
Goes the distance
59.9%

“Goes the distance” is the method head’s decision probability, not a separate model — so it cannot disagree with the bar above it.

The fighters

What the corpus knows about each of them

Tracked bouts only — a fight outside the corpus is not counted, and a thin record says so.

Rafa Garcia
19-4-0

4-1 in the last 5. 29% of wins come by finish; 73% of fights reach the judges.

+ Durable — most tracked fights reach the judges
Too few tracked losses to name a pattern — absence of data, not absence of weakness
Last 5 tracked
WAlexander HernandezDecision R3
WJared GordonKO / TKO R3
WVinc PichelDecision R3
LGrant DawsonKO / TKO R2
WClay GuidaDecision R3
Rongzhu
28-6-0

3-2 in the last 5. 33% of wins come by finish; 50% of fights reach the judges.

+ 6 tracked bouts — too few clear tendencies to call out
Losses tend to come by finish rather than decision
Last 5 tracked
WAustin HubbardDecision R3
WKody SteeleDecision R3
LChris PadillaKO / TKO R2
LIgnacio BahamondesSubmission R3
WBrandon JenkinsKO / TKO R3
Method

Where these numbers come from

Three separate heads — winner, method and round — each fitted on a corpus of tracked UFC fights and each published only on its own PASS verdict against a bar frozen before it was fitted. This is the one model on the site that beat its baseline on every head it was tested on.

Winner 60.6% accurate over 3,557 held-out fights · baseline 50% · log loss 0.6615 against 0.6931
Method 52.7% accurate over 3,557 held-out fights · baseline 46.7% · log loss 0.9692 against 1.0255
Round 60.6% accurate over 3,557 held-out fights · baseline 55.3% · log loss 0.9148 against 0.9549
model ufc-fight-model@e657fc40b64d
corpus 8,642 fights · 1994-03-11 to 2026-08-08 · scrape_ufc_stats (GPL-3.0)
generated 2026-09-09T17:05:43Z

Educational and paper-only — not betting advice.

More

Other bouts on this card