STATSWING

STATSWING assessment grades were retrodiction tested across 9 seasons: 11,474 player-grades, 8,300 validated outcomes. Players graded SW-1 (Elite) produced 3.44× the following-season per-90 output of players graded SW-6. Each tier predicted higher output than the one below.

Companion technical note, 'Do Body-Pose Features Improve Shot Outcome Prediction?' When body-pose features derived from STATSWING's kinetic chain framework are made available alongside event-context metadata, the model relies exclusively on the mechanical features for prediction. All 10 of the top 10 features by SHAP importance are mechanical - upper body twist, hip angle asymmetry, and maximum knee angle ranking highest. Proof-of-concept: statswing.com/research/mecha…

The football transfer market's predictive infrastructure operates on a single analytical layer – statistical actions – while the layer that most determines whether those actions replicate in a new environment is structural execution quality: mechanics. The field conflates what a player does, the technique they select, and the mechanical quality of their execution into a single unit of measurement. This conflation propagates through every model, composite rating, and "league exchange rate" that consumes action-level data as its input. The data infrastructure to assess mechanics already exists in football's top leagues - 29 skeletal points per player, captured at up to 100 frames per second. The analytical products to use it have not been built - though basketball has demonstrated that they can be. Our latest publication SW-R-2026-003, 'Mechanics as the Missing Variable in Transfer Prediction.' statswing.com/research/mecha…

Full methodology: what the grades measure, how confidence works, what they are not, and the validation evidence. statswing.com/intelligence/g…

The system grades 18,000+ players across 20+ leagues on four dimensions: -Statistical (position-specific, league-adjusted) -Market signal -Pattern intelligence -Proprietary assessment Six tiers, SW-1 (Elite) to SW-6 (Below Standard). Modifiers, confidence bands, trajectory.

A reader raised three methodological concerns about the possession-adjusting paper. They were substantive, so we ran three stronger designs to test them. The suppression mechanism – that centre-backs on dominant teams tackle less because they face fewer defensive actions – is real and statistically significant across all three. It is also small: A within-player fixed-effects model across 6,880 match observations estimates the effect at roughly 0.07 fewer tackles per 90 across the full gap between dominant and weak teams; less than 5% of the mean. Standard implementations correct for something an order of magnitude larger, which strengthens the original study's practical recommendations. Read the follow-up companion paper here: statswing.com/research/posse…

5/ Full methodology, all nine correlation tests, and the cross-domain comparison: statswing.com/research/posse…

4/ Basketball's pace adjustment works because possessions are discrete, countable events. Football's are not — which is why other sports moved toward evaluating each defensive action in context rather than adjusting the total count. Football's possession adjustment sits at the stage those sports moved past.

Possession-adjusting individual centre-back statistics rests on an assumption: that dominant-team defenders face fewer defensive actions, so their numbers need correcting upward. We tested that assumption across 431 centre-backs, and the relationship it assumes does not exist at the player level. 🧵:

When confidence compounds through the recruitment stack without epistemic correction, the original measurement gap propagates to the decision layer. We propose a framework for bounding recruitment certainty. SW-R-2026-002. statswing.com/research/epist…

Set pieces now produce one in four Premier League goals. The metric everyone uses to evaluate aerial ability doesn't count the majority of aerial contests. I wrote about why, and what should replace it. statswing.com/research/aeria…

The industry-standard aerial duel metric records a contest only when both players leave the ground. In a single-match case study, this definition excluded 80% of contested aerial situations. SW-R-2026-001 proposes a revised measurement framework. statswing.com/research/aeria…