When the corner and the camera disagreed
A six-fight amateur and his coach thought they knew his weaknesses. The AI baseline scored the round cold — no loyalty, no ego, no recency bias — and pointed the training somewhere neither of them expected.
- Athlete
- Daniel “DV” Vargas
- Profile
- Amateur Muay Thai · 6 fights
- Gym
- Northside Muay Thai
- Timeframe
- 8-week training block
The read everyone already agreed on
Daniel had sparred out of Northside for three years and was preparing for his seventh amateur bout. His coach, Marco, had watched every round he had ever thrown and was certain the priority was power — “he needs to sit down on his kicks more.” Daniel agreed; it matched how the hard rounds felt.
The problem with that read is the problem with all film study done by the people closest to the fighter: it carries memory. Marco remembered the highlight kicks. Daniel remembered the exchanges where he felt strong. Neither of them was watching the quiet, repeated mistakes that don’t feel like anything in the moment — because nothing got countered that day.
“I would have bet money his worst area was power. The model didn’t care what I remembered. It just scored what was on the tape.”
Nine skill areas, scored without an opinion
They uploaded a single unedited three-round sparring clip. Martial Lens graded it across all nine skill areas with no context about Daniel, his record, or his coach’s opinion — the same objective pass every clip gets.
Technical rating
53/100
What the score was actually pointing at
Power generation came back at 6.0 — already one of his stronger areas. Defense scored 4.0, the lowest on the card. The leaks and habits below show why, each one timestamped to the exact moment it happened on tape.
Mechanical leaks
Fix it: Glue the lead hand to the temple through the kick’s full arc; drill kick-and-recover on the bag.
Fix it: Exhale and shorten the combo by one punch; keep the chin behind the lead shoulder.
Fix it: Reset to a 45° stance after every exchange; pivot off the lead foot rather than backing straight up.
Tactical habits
Strengths
Liabilities
Coaching the data, not the memory
Marco threw out the power-focused plan. The objective baseline reframed the eight weeks around defense and the specific mechanical leaks the score exposed — not the ones either of them felt.
Lead-hand discipline
Daily kick-and-recover drilling with the lead hand pinned high, filmed weekly and re-scored to confirm the leak was closing on tape, not just in feel.
Defensive footwork
Replaced straight-line retreats with angled exits off the lead foot; live rounds capped pressure so he could rebuild the habit under control.
Setting up kicks
Every kick in pad work now had to follow a hand touch or feint, attacking the “naked kicks” habit the model flagged eight times.
Re-scored blind, eight weeks later
At the end of the block they uploaded a fresh sparring round under the same conditions. Martial Lens re-scored it blind — it had no memory of the first session and no stake in showing improvement. The gains were concentrated exactly where the objective data had pointed.
Baseline
53/100
After the block
68/100
| Skill area | Baseline | After | Change |
|---|---|---|---|
| Stance & Guard | 5.0 | 7.0 | +2.0 |
| Hands / Punches | 6.0 | 6.5 | +0.5 |
| Kicks | 6.0 | 7.0 | +1.0 |
| Knees & Elbows | 5.0 | 5.5 | +0.5 |
| Clinch | 5.0 | 5.5 | +0.5 |
| Footwork | 5.0 | 7.5 | +2.5 |
| Defense | 4.0 | 7.0 | +3.0 |
| Balance | 5.0 | 6.5 | +1.5 |
| Power Generation | 6.0 | 6.5 | +0.5 |
Defense climbed from the lowest score on the card to a 7.0, and footwork moved with it — the two areas the objective baseline had isolated. Power generation, the thing everyone wanted to chase, barely moved, and it didn’t need to. The AI didn’t coach Daniel. It just made sure the coaching pointed at the truth.
Score your own baseline