We use Google Analytics to understand which pages help players. No names, emails or match footage are ever sent. See our privacy notice.
How the PADEXA Index works.
The Index is a 0–100 score of how you actually play — built from every rally of your filmed match, scored against published padel research, and honest about its own uncertainty. This page explains the method: what goes in, how it's scored, what we deliberately don't claim, and how we're testing it. With citations.
Your results know if you won.
They never know why.
Results-based ratings are good at their job — matchmaking. But in doubles, half of every result belongs to your partner, and all of it is blind to behaviour. The scientific case for watching the player instead is now peer-reviewed: in a 2026 study of 180 club players across 50 matches, ratings derived from match video tracked a blinded panel of expert coaches strongly (r = 0.78) and edged the results-based app level on classification accuracy — which over-rated players by nearly a full level. [11] More on how padel levels are measured.
Scoreboard in, level out.
Wins and losses, averaged with your partner's, lagging your actual play by months. Structurally unable to tell which of the four players on court made the errors that decided the match — no scoreboard system can.
Behaviour in, level out.
Every rally observed: who hit what, from where, with what outcome. The Index scores the padel you produced — blended with verified results so a stylish loser can't outrank an effective winner — and it moves when your behaviour moves.
Five pillars. Each one earns its place in the literature.
The pillars aren't our opinion of what matters in padel — they're what three decades of match-analysis research says separates winning pairs from losing ones. Every pillar is scored against published reference values, conditioned by sex and level, because the science is clear that the norms differ. [1][9] The underlying numbers live in the padel statistics evidence base.
Net control
Padel is a territory game and the net is the territory. We score how often your pair takes, holds and retakes the attack position — not how pretty your volleys are.
About 80% of professional points are won from the net zone [1]; holding the net zone gives the highest point-win probability in the men's pro game — 82% on serve, 74% on return [2]; losing pairs fail to mount an attack in 47% of points. [1]
Error economy
The pillar that decides most club matches, because club matches aren't won — they're donated. We score your unforced errors as rates, priced by the scoreboard situation they happened in, and mapped by where they died.
Around 65% of club-level points end on unforced errors [3]; winning club pairs give away half the return errors of losing pairs (9.3 vs 18.7 per match) [3]; roughly half of all errors finish in the net. [4]
Transitions
How you move between defence and attack — above all, the lob. We score whether your lobs actually flip the court, and how often you surrender a position you'd just won.
Lobs are 85% of the shots that get past the net pair, win the net position 70% of the time — and the evicted pair takes it back half the time, so territory is rented, not owned. [5]
Finishing
When the chance arrives, do you convert it — and do you take it from the right place? We score winners per opportunity, not per match, plus overhead discipline: which smash, from where, aimed where.
When a smash ends a pro point it's a winner ~89% of the time for men, ~75% for women [6]; flat smashes produce 75% of smash winners while slices cause 39% of smash errors [7]; overheads produce ~55% of winners in the men's pro game. [8]
Serve & return
Neither shot is glamorous; both are structural. We score the serve as net access (what it buys, not what it aces) and the return as survival — the free points you refuse to donate.
The serving pair wins 59.3% of points, an advantage that decays from 71% in short rallies to 47% by shot thirteen [2]; club players land 78.8% of first serves against a 92.9% elite reference [3]; elite returns come back over 90% of the time. [10]
Eight steps between your footage and your number.
This is the shape of the system — stated plainly, because a score you can't interrogate is a score you shouldn't trust. The exact weights, reference values and update constants are PADEXA's protected core; everything else is on the table. See what a finished report looks like in padel match analysis.
Observe
The AI watches every rally of every set and logs structured events — shots, outcomes, positions, phases — each with a timestamp and a confidence rating. What it couldn't see clearly is excluded, not guessed.
Turn counts into rates
Nothing is scored as a raw count. Every metric has a denominator — errors per shot, winners per opportunity, net points per rally — so a long match and a short one are comparable.
Price the context
An error at 30–30 is not an error at 40–0. Every event is weighted by how much it actually moved the game, using published win-probability data by scoreline. [9]
Respect small samples
One match is thin evidence. Estimates are statistically shrunk toward the reference for your sex and level until your own data outweighs the prior — a standard, published technique, not a fudge.
Compare to the right people
Your rates are placed against reference values conditioned by sex and level — because pro-normal behaviour isn't club-normal, and men's and women's padel are measurably different games. [1]
Aggregate the pillars
The five pillar scores combine into a performance score. The weights are evidence-seeded and re-fitted against predictive validity as our match corpus grows — they are not published.
Update, don't jump
Your Index is a running estimate, not a per-match verdict. New evidence moves it in proportion to its reliability; one bad night reads as noise until the pattern repeats. Uncertainty grows if you stop playing.
Blend with results
Finally, verified results are blended in — weighted by opposition strength, with credit split by observed contribution, not split evenly with your partner. Performance leads, results anchor.
Engineered to admit what it can't see.
Confident-sounding nonsense is how players lose trust in AI analysis. So the Index's uncertainty handling isn't a disclaimer — it's load-bearing architecture, and it shows up in the product.
Your Index displays with a band that narrows as evidence accumulates. Early scores are labelled “early read”. Single-camera vision is genuinely less accurate for the far pair and at the net — published accuracy research says so [12] — so those observations carry less weight and wider bands, by rule.
A pillar doesn't contribute until enough of the right events have been observed. Too few service points? The serve pillar sits out, marked as an early read, rather than pretending a handful of serves is a verdict.
The number never moves on one observation, and every move decomposes into pillar deltas you can see. If your Index changed, you can find out exactly which behaviour changed it — and jump to the clips.
Because behaviour is scored, not margins: managing the scoreline does nothing, tanking dead points carries near-zero leverage, and playing scared drains your net-control and transition pillars faster than it protects your error economy.
Pre-registered validation, published either way.
A methodology page is claims; a validation study is evidence. The strongest published benchmark in this field is a 2026 peer-reviewed study that rated 180 club players three ways — expert coaches, a video AI, and a results-based app — and reported exactly how closely each pair agreed. [11] We are holding ourselves to that same protocol, and we're stating the targets before we run it.
PADEXA will replicate the published coach-panel protocol: a blinded panel of qualified coaches will independently rate a cohort of PADEXA users from their match footage, and we will report the agreement between the PADEXA Index and the coach benchmark — Pearson correlation, concordance, mean absolute error, and classification accuracy. Targets: match or beat the published video-AI result (r = 0.78; 74.3% accuracy). The results will be published on this page, whether they flatter us or not.
Known limits, stated plainly.
- Published norms are mostly professional and sub-elite. Peer-reviewed amateur padel data barely exists. Our reference tables start from the published values and are progressively replaced by PADEXA's own amateur corpus — until a cell is filled by real amateur data, scores against it carry wider bands.
- One phone is not a lab rig. Split-second events — net cords, double bounces, line calls — are not reliably decidable from a single camera behind the glass, so the Index never uses them. Multi-second patterns (positioning, rally shape, error patterns) are the reliable ground, and that's where the Index lives.
- A single match measures a day, not a player. Sleep alone can slow elite players ~11% overnight. [13] That's why the Index updates gradually and why one-off dips are treated as noise until they repeat.
- The published ceiling is agreement with coaches, not perfection. The best published video-AI agreement with expert coaches is r = 0.78 — and coaches themselves disagree with each other. Anyone claiming a perfect padel rating is selling something. Ours comes with error bars on purpose.
The evidence base.
[1] Martín-Miguel, Escudero-Tena, Muñoz & Sánchez-Alcaraz (2023). Performance Analysis in Padel: A Systematic Review. Journal of Human Kinetics.
[2] Prieto-Lage et al. (2024). Assessing the Probability of Winning a Point in Men's Padel. Applied Sciences, 14(15), 6642.
[3] Ungureanu, Lupo & Brustio (2022). Padel Match Analysis: Notational and Time-Motion Analysis during Official Italian Sub-Elite Competitions. IJERPH, 19(14), 8386.
[4] Ramírez-Ortega, Sánchez-Pay & Sánchez-Alcaraz (2025). Distribution of Winners and Errors in Professional Padel. Padel Scientific Journal, 3(1).
[5] Escudero-Tena, Fernández-Cortes, García-Rubio & Ibáñez (2020). Use and Efficacy of the Lob to Achieve the Offensive Position in Women's Professional Padel. IJERPH, 17(11), 4061.
[6] López-Sierra, Escudero-Tena, Ibáñez & Muñoz (2025). Data-Driven Decision Trees for Tactical Shot Selection in Professional Padel. Applied Sciences, 15(4), 2198.
[7] Escudero-Tena et al. (2024). Analysis of the Smashes of the Qatar Major, Premier Padel 2023. (880 smashes, validated OASP instrument.)
[8] Escudero-Tena, Sánchez-Alcaraz, García-Rubio & Ibáñez (2022). Analysis of Errors and Winners in Men's and Women's Professional Padel. Applied Sciences, 12(16), 8125.
[9] Martín-Miguel (2026). Influence of Scoreline on Point and Game Outcomes in Professional Men's Padel. Padel Scientific Journal, 4(2). (2,077 Premier Padel points.)
[10] Cerrillo-Lafuente, Conde-Ripoll, Escudero-Tena & Sánchez-Alcaraz (2025). Technical-Tactical Analysis of the Return in U-18 High-Level Padel. Padel Scientific Journal, 3(2).
[11] Fernandez-de-Osso & Sánchez-Trigo (2026). AI video ratings vs expert coaches and results-based levels: 180 players, 50 matches, 9 clubs. International Journal of Sports Science & Coaching.
[12] Javadiha et al. (2021). Estimating Player Positions from Padel High-Angle Videos. Sensors, 21. (Position accuracy and per-zone error amplification.)
[13] Díaz-García et al. (2025). Sleep Restriction and Padel Performance in Elite Youth Players. Padel Scientific Journal.
These thirteen are the load-bearing citations for this page; the full PADEXA research corpus runs to 61 published sources, maintained in the padel statistics evidence base. Related reading: padel match analysis · what's my padel level? Academic or press queries about the methodology: hello@padexa.ai — we answer them.
Get your number.
Film your next match. The Index starts with match one — early read, honest band, and the one fix that matters most.
Analyse my match — first match free£8.99/month after. Cancel any time.