VERIFIEDfull tape · lead-right

Anthony Romero

258

verified strikes (swept)

≥179

reviewed attempts (sweep, lower bound)

fights on file

verified fights

machine fights

Ask the Analyst

coverage

tendencies computed from 3 round-scoped fights (vs Aviles, vs Dao, vs Frangulyan); 2 clip-logged fights contribute volume & per-fight detail only

The tape in one line

the strongest measured read on file — receipts attached

when hit

47%of incoming strikes get an answer within 2s

When strikes come at Anthony Romero, he fires back within 2 seconds 47% of the time. That's 1.7× what his own pace would produce by chance. He usually answers with a left punch (71% of his responses).

Confidence interval [40%, 54%] · lift interval [1.2×, 2.6×] · n = 167 opponent strikes · 79 responses

1.7× vs chance · interval [1.2×, 2.6×]

consistent across 3 fights

left punch71% (56)

right punch24% (19)

other5% (4)

n = 79

Watch the Film Study▾

vs Aviles1:502:032:083:048:13vs Dao0:250:270:291:201:23vs Frangulyan1:011:031:09

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

response = this fighter's next strike within 2.0s of the opponent's strike (two-sided stream, single-opponent context)

lift = observed rate ÷ a circular-shift null that keeps this fighter's own pace and burst structure but scrambles timing vs the opponent

no stance asset — response classes are physical left/right

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 167 eventsbootstrap-2000 rate CI; circular-shift null ×2000 for lift (seeded per cell·fight)pre-registered gate: n ≥ 25 for shares/conditionals

Signature tendencies

the strongest measured patterns in his verified tape — nothing editorial, receipts on every line

paceHis output climbs round over round: 7.6 → 5.5 → 20.3 strikes per minute.n = 247 strikes

bursts59% of his output comes in bursts.n = 247 strikes

after right punchAfter a right punch, his next strike is a left punch 73% of the time.Confidence interval [60%, 85%] · n = 52

after left punchAfter a left punch, his next strike is a right punch 55% of the time.Confidence interval [42%, 67%] · n = 66

What follows what — the top read

his most predictable pattern — the full table lives on the Fight IQ tab

After a right punch, his next strike is a left punch 73% of the time.

Confidence interval [60%, 85%] · n = 52 follow-ups within 2s

consistent across 3 fightspre-registered gate: n ≥ 25 for shares/conditionals

Output by round

pooled pace — read the trend, not single-round precision

events per minute by round↑ output rises by round

Anthony Romero throws more as the fight goes on: 7.6 strikes a minute in round 1, up to 20.3 by round 3.

n = 247 strikes across 3 rounds

When he gets hit · bursts

what he does when he gets hit, and how his output clusters — each measured against what his own pace would produce by chance

bursts

59%of his output arrives in bursts

59% of Anthony Romero's output comes in bursts — strikes less than 1 second apart.

Confidence interval [53%, 65%] · n = 247 strikes

consistent across 3 fights

vs Aviles57% (median gap 2.4s)

vs Dao59% (median gap 2.0s)

vs Frangulyan80% (median gap 0.6s)

Watch the Film Study▾

vs Aviles1:471:471:561:562:09vs Dao0:290:291:251:261:28vs Frangulyan1:081:091:091:10

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

burst = inter-strike interval < 1.0s

lift = observed burst share ÷ a null that keeps this fighter's event COUNT but redistributes onsets uniformly over his active in-round time

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 247 eventsbootstrap-2000 seeded per cellpre-registered gate: n ≥ 25 for shares/conditionals

Output by round — pooled

↑ output rises by round

Anthony Romero throws more as the fight goes on: 7.6 strikes a minute in round 1, up to 20.3 by round 3.

n = 247 strikes across 3 rounds

strike mix

94%of his strikes are punches

Punches make up 94% of what Anthony Romero throws; the other 6% is kicks and everything else.

Confidence interval [91%, 97%] · n = 247 strikes

consistent across 3 fights

punch94% (232)

kick4% (9)

other2% (6)

n = 247

Watch the Film Study▾

vs Aviles1:341:471:471:501:53vs Dao0:270:290:291:201:25vs Frangulyan1:031:081:091:091:10

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 247 eventsbootstrap-2000 seeded per cellpre-registered gate: n ≥ 25 for shares/conditionals

side mix

57%of his strikes come from the left side

Anthony Romero throws 57% of his strikes from the left side. His kicks come off the left leg 100% of the time.

Confidence interval [51%, 63%] · n = 247 strikes (9 kicks)

consistent across 3 fights

not enough data to break down the mix — 9 of the 25 needed

Watch the Film Study▾

vs Aviles1:471:471:562:032:41vs Dao0:270:291:201:251:26vs Frangulyan1:031:081:09

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

no stance asset for this fight — sides are physical left/right, not lead/rear

stance derived from a validated pose asset (ankle-vote, 12/12 vs human truth; experiments/pose_stance_validation) — jab/cross and lead/rear ride on it

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 247 eventsbootstrap-2000 seeded per cellpre-registered gate: n ≥ 25 for shares/conditionals

strike types

39%of his strikes are the straight

Anthony Romero's most-thrown strike is the straight — 39% of everything he throws.

n = 247 classified strikes

single bout — observed vs Dao only

straight39% (97)

jab14% (34)

hook13% (32)

cross13% (31)

uppercut11% (26)

overhand5% (12)

body kick1% (3)

head kick1% (3)

leg kick1% (2)

kick0% (1)

other2% (6)

n = 247

Watch the Film Study▾

vs Dao0:290:291:201:341:34

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

stance derived from a validated pose asset (ankle-vote, 12/12 vs human truth; experiments/pose_stance_validation) — jab/cross and lead/rear ride on it

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 247 eventspre-registered gate: n ≥ 25 for shares/conditionals

output by round

6.2 / 6.8strikes per minute, round by round

Anthony Romero's pace holds steady: 6.2 strikes a minute in round 1, 6.8 in round 2.

n = 65 strikes across 2 rounds

single bout — observed vs Aviles only

round 16.2/min (31 strikes)

round 26.8/min (34 strikes)

Watch the Film Study▾

vs Aviles1:341:471:471:501:53

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

per-round counts sit on a Poisson floor of roughly ±18–25% — read trends, not single-round precision

engine provenance ▸

source handlog · 65 events

output by round

9.2 / 4.4 / 20.3strikes per minute, round by round

Anthony Romero throws more as the fight goes on: 9.2 strikes a minute in round 1, up to 20.3 by round 3.

n = 177 strikes across 3 rounds

single bout — observed vs Dao only

round 19.2/min (57 strikes)

round 24.4/min (26 strikes)

round 320.3/min (94 strikes)

Watch the Film Study▾

vs Dao0:270:290:291:201:25

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

per-round counts sit on a Poisson floor of roughly ±18–25% — read trends, not single-round precision

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 177 events

output by round

5.4strikes per minute, round by round

Anthony Romero threw 5.4 strikes a minute in round 1 — the only round on file.

n = 5 strikes

single bout — observed vs Frangulyan only

round 15.4/min (5 strikes)

Watch the Film Study▾

vs Frangulyan1:031:081:091:091:10

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

per-round counts sit on a Poisson floor of roughly ±18–25% — read trends, not single-round precision

engine provenance ▸

source handlog · 5 events

landed rate

41%of his strikes land

41% of Anthony Romero's strikes land, as judged by our annotator watching the tape.

Confidence interval [34%, 47%] · n = 241 labeled strikes

consistent across 3 fights

punch41% of 229

kicknot enough data (6 of 25)

n = 235

Watch the Film Study▾

vs Aviles1:531:562:093:003:01vs Dao0:271:341:361:441:56vs Frangulyan1:081:09

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

landed/missed is the ANNOTATOR's judgment on the hand log — it is not machine-derivable at current capture and carries no defended/absorbed nuance

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 241 eventsbootstrap-2000 seeded per cellpre-registered gate: n ≥ 25 for shares/conditionals

what lands on him

39landed strikes absorbed

Anthony Romero absorbed 39 landed strikes across 3 fights — 25 punches, 13 kicks, 1 knee.

n = 39 landed strikes (one opponent per bout)

consistent across 3 fights

punch25 punches absorbed

kick13 kicks absorbed

other×1

Watch the Film Study▾

vs Aviles3:045:509:099:119:40vs Dao0:311:291:401:492:28vs Frangulyan1:031:09

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

received = the OPPONENT's landed strikes in this bout (single-opponent context); annotator-judged landed only

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 39 events

sweep-verified

Computed from a human sweep (no hand-log). Sides are physical left / right — stance unavailable, not lead / rear; a per-fight stance asset would unlock lead/rear.

the top read

73%after a right punch: left punch

Confidence interval [60%,85%] · n = 52

consistent across 3 fightspre-registered gate: n ≥ 25 for shares/conditionals

What follows what

each row shows what he throws next after that strike — deeper red means it happens more often; click a row for the film. rows without enough tape stay visible with their reasons — an honest “not enough data” beats a confident guess

consistent across 3 fights

after…

left punch

right punch

left kick

right kick

left punch

35%[23%,47%]

55%[42%,67%]

3%[0%,8%]

0%[0%,0%]

After a left punch, Anthony Romero's next strike is a right punch — 55% of the time.

Confidence interval [42%, 67%] · n = 66 follow-ups within 2s

right punch 55% [42%,67%]video link not yet verified

vs Aviles1:472:593:003:013:09vs Dao0:271:251:281:341:44vs Frangulyan1:081:09

greyed: no playable video registered

pre-registered gate: n ≥ 25 for shares/conditionals

right punch

73%[60%,85%]

23%[12%,35%]

2%[0%,6%]

0%[0%,0%]

After a right punch, Anthony Romero's next strike is a left punch — 73% of the time.

Confidence interval [60%, 85%] · n = 52 follow-ups within 2s

left punch 73% [60%,85%]video link not yet verified

vs Aviles1:562:092:412:572:59vs Dao0:291:341:562:423:10vs Frangulyan1:09

greyed: no playable video registered

pre-registered gate: n ≥ 25 for shares/conditionals

left kick

1 of 25needs 24 more

only 1 left kick events were followed by another strike within 2s — 25 needed

right kick

0 of 25needs 25 more

only 0 right kick events were followed by another strike within 2s — 25 needed

Sequences

consecutive strikes with every gap inside the fixed window — ranked by frequency

his most common 2-strike combination

Anthony Romero's go-to 2-strike combination is straight → straight — he threw it 35 times, 29% of his 2-strike sequences.

n = 120 sequences

single bout — observed vs Dao only

01straight → straight35× (29%)

02jab → cross6× (5%)

03uppercut → hook6× (5%)

04straight → hook6× (5%)

05cross → jab5× (4%)

06hook → hook5× (4%)

07jab → hook3× (2%)

08jab → uppercut3× (2%)

Watch the Film Study▾

vs Dao0:291:341:563:383:42

greyed: no playable video registered

pre-registered gate: n ≥ 25 for shares/conditionals

his most common 3-strike combination

Anthony Romero's go-to 3-strike combination is straight → straight → straight — he threw it 12 times, 20% of his 3-strike sequences.

n = 59 sequences

single bout — observed vs Dao only

01straight → straight → straight12× (20%)

02straight → straight → hook5× (8%)

03uppercut → jab → cross2× (3%)

04overhand → straight → straight2× (3%)

05straight → hook → uppercut2× (3%)

06straight → hook → hook2× (3%)

07hook → straight → straight2× (3%)

08jab → uppercut → hook1× (2%)

Watch the Film Study▾

vs Dao12:4112:5313:0214:0814:08

greyed: no playable video registered

pre-registered gate: n ≥ 25 for shares/conditionals

When he gets hit

how often Anthony Romero fires back within 2 seconds of eating a strike. The “vs chance” number compares that to what his own pace would produce at random — so it measures deliberate answering, not just busyness. This is about his offense, not how open he is to being hit.

when he gets hit

47%of incoming strikes get an answer within 2s

Confidence interval [40%, 54%] · lift interval [1.2×, 2.6×] · n = 167 opponent strikes · 79 responses

1.7× vs chance · interval [1.2×, 2.6×]

consistent across 3 fights

left punch71% (56)

right punch24% (19)

other5% (4)

n = 79

Watch the Film Study▾

vs Aviles1:502:032:083:048:13vs Dao0:250:270:291:201:23vs Frangulyan1:011:031:09

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

response = this fighter's next strike within 2.0s of the opponent's strike (two-sided stream, single-opponent context)

lift = observed rate ÷ a circular-shift null that keeps this fighter's own pace and burst structure but scrambles timing vs the opponent

no stance asset — response classes are physical left/right

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlog · 167 eventsbootstrap-2000 rate CI; circular-shift null ×2000 for lift (seeded per cell·fight)pre-registered gate: n ≥ 25 for shares/conditionals

⟶ Measured reads

the strongest conditional signals on file — receipts attached, nothing editorial

He fires back at 47% of the strikes thrown at him within 2 seconds — 1.7× what his pace would produce by chance. He usually answers with a left punch (71% of his responses).Confidence interval [40%, 54%] · n = 167 opponent strikes

59% of his output comes in bursts.n = 247 strikes

After a right punch, his next strike is a left punch 73% of the time.Confidence interval [60%, 85%] · n = 52

After a left punch, his next strike is a right punch 55% of the time.Confidence interval [42%, 67%] · n = 66

Ask the analyst →

What lands on him

the openings an opponent actually found — annotator-judged landed strikes absorbed

39landed strikes absorbed

straight×13

leg kick×7

hook×7

body kick×6

uppercut×3

single-opponent context per bout — a hole an opponent found, not a population rate

the gate

pre-registered gate: n ≥ 25 for shares/conditionals. Rows below the gate stay visible with their reasons — an honest "not enough tape" beats a confident guess.

limitedDefense telemetry is early: the signal is real and eyeball-confirmed, but capture coverage — not the metric — is the blocker. More verified tape unlocks it.

defensive tell — guard recovery

insufficient data

0 of 25needs 25 more

guard telemetry coverage too thin under the frozen validity rules for this fighter

unlock: a defense-telemetry upgrade is pre-registered — these cells re-grade automatically after the next data pass

pre-registered gate: n ≥ 25 for shares/conditionals

fight volume

72human-verified strikes (swept)

72 human-verified strikes (swept) in this fight — 68 punches, 2 elbows, 1 kick, 1 knee.

n = 72 strikes

punch68 punch

kick1 kick

other×3

Watch the Film Study▾

vs Aviles1:341:471:471:501:53

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

no round/segment scoping available for this fight — raw fight-window counts only

engine provenance ▸

source handlogsweep 72

fight volume

181human-verified strikes (swept)

181 human-verified strikes (swept) in this fight — 167 punches, 8 kicks, 6 knees.

n = 181 strikes

punch167 punch

kick8 kick

other×6

Watch the Film Study▾

vs Dao0:270:290:291:201:25

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

no round/segment scoping available for this fight — raw fight-window counts only

romero_dao: census-grade review (2524/2551 in-scope judged); counts are lower bounds — 27 unreviewed candidates may include real strikes

engine provenance ▸

source handlogsweep 181

fight volume

insufficient data

0 of 25needs 25 more

no swept events

pre-registered gate: n ≥ 25 for shares/conditionals

fight volume

5human-verified strikes (swept)

5 human-verified strikes (swept) in this fight — 5 punches.

n = 5 strikes

punch5 punch

Watch the Film Study▾

vs Frangulyan1:031:081:091:091:10

greyed: no playable video registered

standup-scoped, in-round hand-log events (single annotator; ~11–15% estimated strike undercount — every level is a floor)

no round/segment scoping available for this fight — raw fight-window counts only

engine provenance ▸

source handlogsweep 5

fight volume

insufficient data

0 of 25needs 25 more

no swept events

pre-registered gate: n ≥ 25 for shares/conditionals

Verified verdict profile

Operator verdict-ledger evidence, kept separate from temporal tendencies. Complete reviews may support totals and gated rates; partial reviews remain labeled samples; reviews below the coverage floor expose only their withholding reason.

verdict evidence on file

Complete-review pooled totals

91 own real attempts across aviles romero, romero frangulyan · 50 opponent real attempts. Partial and withheld fights are excluded from these totals and every rate.

Own clean-landed rate: 39/91 (43%) · Opponent clean-landed-on-him rate: 17/50 (34%)

Strike mixes, combinations, and conditionals for these complete-review fights are computed once, by the cell engine, and rendered in the Overview, Striking, and Fight IQ tabs — this panel repeats none of them.

aviles romero

816/816 reviewed (100%) · Complete review

85 own real attempts · 39 opponent real attempts — full analytics render in the main tabs.

5 unresolved attribution/validity verdicts — a floor on missingness.

romero dao

3472/3505 reviewed (99%) · Census review

232 own real attempts · 190 opponent real attempts — full analytics render in the main tabs.

32 unresolved attribution/validity verdicts — a floor on missingness.

romero dias

0/0 reviewed (0%) · No verified verdicts

0 unresolved attribution/validity verdicts — a floor on missingness.

romero frangulyan

301/301 reviewed (100%) · Complete review

6 own real attempts · 11 opponent real attempts — full analytics render in the main tabs.

1 unresolved attribution/validity verdicts — a floor on missingness.

romero lewis

2363/3966 reviewed (60%) · Partial review

Reviewed sample only — not fight totals or accuracy: 179 own real attempts · 194 opponent real attempts.

Own clean-landed rate: complete review required for an accuracy rate · Opponent clean-landed-on-him rate: complete review required for a defense rate

Sample from an in-progress review; ordering is not random — rates may be inflated and early-fight-weighted.

weapon sample

elbow 2 (1%) · fist 152 (85%) · foot shin 18 (10%) · knee 5 (3%) · unlabeled 2 (1%)

trajectory sample

arcing 17 (9%) · hooked 54 (30%) · overhand 1 (1%) · straight 85 (47%) · unlabeled 9 (5%) · upward 13 (7%)

side sample (physical, not stance)

left 91 (51%) · right 86 (48%) · unlabeled 2 (1%)

target sample

body 32 (18%) · head 138 (77%) · legs 7 (4%) · unlabeled 2 (1%)

outcome sample

blocked 12 (7%) · landed 77 (43%) · missed 56 (31%) · partial 15 (8%) · slipped 16 (9%) · unlabeled 3 (2%)

opponent outcome sample

blocked 27 · landed 77 · missed 56 · partial 21 · slipped 12 · unlabeled 1

53 unresolved attribution/validity verdicts — a floor on missingness.

The three frozen verdict gates

Per-fight mix shares require n ≥ 25 reviewed real attempts; below it, counts only.
Complete-fight pooled mix shares and own rates require n ≥ 25 real attempts.
Defense clean-absorb rates require a complete review and n ≥ 25 opponent real attempts.

Partial reviews are non-random samples: counts and eligible sample mixes are not fight totals or accuracy.
Physical side means left/right, not lead/rear stance. Jab/cross for swept strikes is stance-INFERRED from pose: a windowed ankle-vote at the strike instant when decisive (94.3% vs hand-log truth), else the fight-level lead; strikes where both abstain stay "straight" rather than guess.
A single reviewer produced these verdicts; unresolved attribution/validity verdicts are a floor on missingness.

Sweep analytics — reviewed sample

Counts from human-reviewed candidates only; they are lower bounds for the whole fight. Rates and temporal patterns stay withheld until a fight's review is complete.

lower bounds · review in progress

≥179

reviewed strike attempts

≥77

landed clean (reviewed)

≥194

opponent attempts faced

≤36

est. real strikes unreviewed

vs Lewis1852/3047 in-scope reviewed · est ≤36 unreviewed

attempts ≥179 · landed clean ≥77 · partial 15 · opponent attempts ≥194

fist 152 · 85%foot_shin 18 · 10%knee 5 · 3%elbow 2 · 1%unlabeled 2 · 1%

landed 77 · 43%missed 56 · 31%slipped 16 · 9%partial 15 · 8%blocked 12 · 7%unlabeled 3 · 2%

head 138 · 77%body 32 · 18%legs 7 · 4%unlabeled 2 · 1%

Defense rates require a complete review: the unreviewed low-confidence tail is exactly where missed opponent attempts live, so a partial denominator would flatter the defender.

Coverage & provenance

packpack-v3

vocabv1.1

data through2026-08-10

regression gatePASS

vs Aviles

complete assertion review: 816/816 (100%); temporal stream admitted; assertion review complete: 816/816 (100.0%); 5 unresolved — a floor on missingness

816/816 (100%)verified

vs Dao

census assertion review: 2524/2551 in-scope (98.9%); 27 unreviewed candidates may include real strikes — temporal stream admitted as lower-bound counts; defense rates require 100%; assertion review census: 3472/3505 (99.1%); 32 unresolved — a floor on missingness

3472/3505 (99%)verified

vs Dias

assertion review below 20%: 0/0 (0.0%); metrics withheld and no temporal stream admitted; assertion review zero_evidence: 0/0 (0.0%); 0 unresolved — a floor on missingness

0/0 (0%)verified

vs Frangulyan

complete assertion review: 301/301 (100%); temporal stream admitted; assertion review complete: 301/301 (100.0%); 1 unresolved — a floor on missingness

301/301 (100%)verified

vs Lewis

partial assertion review: 2363/3966 (59.6%); in-scope 1852/3047 (60.8%), est ≤35.91 real strikes unreviewed (calibrated); counts are lower bounds — stream admitted at census grade or 100%; assertion review partial: 2363/3966 (59.6%); 53 unresolved — a floor on missingness; Sample from an in-progress review; ordering is not random — rates may be inflated and early-fight-weighted.

2363/3966 (60%)verified

Not measurable from broadcast

the single-feed footage can’t support these — they are not a locked upgrade, they simply aren’t recoverable. Saying so is the product.

not recoverable

Landed / grazed / missed

outcome carries ~no machine signal at this capture and is unreliable even by eye from a single broadcast feed

not recoverable

Fine punch subtype (jab / cross / hook)

capped by 72% annotator self-agreement and the single-feed capture wall

not recoverable

Clinch & ground interior

the corpus is standup-scoped — a single broadcast feed can't resolve grappling exchanges

not recoverable

Feint intent

intent-without-contact is not recoverable from broadcast video

Known limits

single-annotator ground truth; measured undercount ~8.8% on the deepest-swept fight — global estimate 11–15%; every count is a floor, not an exaggeration
tendency numbers are standup-scoped and in-round; clinch/ground exchanges are not measured
stance (lead/rear) is a per-fight constant from a pose asset; a mid-fight stance switch would mislabel lead/rear symmetrically
per-round counts sit on a Poisson floor of roughly ±18–25%; product language is trends and conditionals, never single-round precision