🚧 Beta version - The site is under active development. I am aware of many bugs and fix them daily. Any feedback is welcome - data, functionality, UI, security, or anything else. · feedback welcome Contact

Terminology

Glossary of terms and metrics used on DanceStats.eu. Click any linked term in the platform to jump directly to its explanation.

Ratings & Statistics

Elo Rating Statistics

The Elo rating is a system for measuring the relative strength of couples, originally developed by Hungarian-American professor Arpad Elo for chess. In dance sport, it works the same way - every couple starts with a base rating of 1500 points, which changes after each competition based on results against other couples.

Key principle: When a weaker couple (low Elo) beats a stronger couple (high Elo), it gains more points than if it had beaten a similarly rated couple. Conversely, losing to a stronger opponent costs fewer points.

  • Base rating: 1500 points (new couples)
  • Beating a stronger opponent = larger point gain
  • Losing to a weaker opponent = larger point loss
  • The K-factor determines how quickly ratings change (see K-factor)
Example: A couple with Elo 1650 beats a couple with Elo 1800 - gains ~25 points. If they had lost, they would only lose ~7 points.

How Elo is tracked

The Elo rating is not attached to an individual dancer but to a couple and a specific discipline:

  • Tracked per couple (two partners) - each new partnership starts with its own rating.
  • Separate ratings for Standard, Latin and Combination - each discipline evolves independently.
  • A new couple partially inherits the rating from each dancer's previous partnerships; the remainder is filled in with the 1500 base.

Elo across classes - caveats

Classes (D, C, B, A, M) form segregated competition pools - a C-class couple competes almost exclusively against other C-class couples, an M-class couple against M couples. Elo, however, is calculated on the same scale for everyone, so ratings cannot be directly compared across classes.

A strong C-class couple can, after dozens of wins, reach an Elo similar to (or even higher than) a newly promoted M-class couple still finding its feet - yet the M couple would still win a head-to-head with ease. The gap between classes reflects Dance class.

Real-world distribution (median Elo in ČSTS Standard, couples with 5+ competitions):
M = 1574, A = 1524, B = 1517, C = 1503, D = 1461, E = 1452. The medians are close, yet a typical M couple beats a typical C couple almost every time - the difference is not in the number itself but in the fact that the M couple earned M class through a long chain of promotions.

How to read the number

  • Within the same class and discipline: higher Elo = more likely winner head-to-head.
  • Across classes: use class as the primary ranking signal and Elo as secondary - a higher class almost always means a stronger couple regardless of the exact Elo value.
  • New couples (few competitions) have ratings that have not yet stabilised. A freshly promoted couple keeps its rating - promotion only widens its uncertainty, which settles again over the next ~3-5 competitions.
K-Factor Statistics

The K-factor determines how much a couple's Elo rating changes after a single competition. A higher K-factor means faster rating changes.

  • New couples (< 10 competitions): high K-factor - rating stabilises quickly
  • Moderately experienced couples: medium K-factor
  • Experienced couples (30+ competitions): low K-factor - stable rating
Momentum (form) Statistics

Momentum measures a couple's current form based on the trend of their recent results. If a couple has been improving their placements in recent competitions, they have positive momentum (rising form).

Momentum is one of the 8 factors that adjust the Elo prediction in Model B. A couple on the rise is rated higher than their base Elo would suggest.

PageRank Statistics

PageRank is an algorithm originally developed by Google for ranking web pages, adapted for dance sport. It measures "strength of victories" - not just how many opponents a couple has beaten, but how strong those opponents were.

A couple that regularly beats strong opponents (who themselves beat other strong opponents) earns a higher PageRank than one that only wins against weak opponents.

Adjudication & Marks

Cross mark Adjudicating

A cross mark is the basic unit of scoring in preliminary rounds of dance competitions. Each adjudicator awards crosses to couples they believe should advance to the next round - the more crosses a couple receives from the panel, the better their assessment.

An adjudicator has a limited number of crosses available (typically half or two-thirds of the field). A cross is binary - either a couple receives it (✓) or not.

Cross-rate Adjudicating

Cross-rate is the ratio of crosses an adjudicator has awarded versus the total number of assessments. It represents an adjudicator's "generosity" - how easily they give crosses. An adjudicator's global cross-rate is the average across all their assessments.

Cross-rate is a key metric for analysing adjudication behaviour. An adjudicator with a 45% cross-rate gives crosses to less than half of all couples they assess.

Example: An adjudicator assessed 200 couples and gave 90 crosses → cross-rate = 45%. For a specific couple: 8 crosses from 10 assessments → cross-rate = 80%.
Used on: Predictor, Adjudicator profile, Trainer
Adjudicator Style Match Adjudicating

Style match measures how well an adjudicator's scoring patterns align with a specific couple's style. It is calculated as the difference between the adjudicator's cross-rate for a given couple and their global cross-rate.

Style match is a normalised value where:

  • Positive value: adjudicator gives the couple more crosses than their average (higher compatibility)
  • Zero: adjudicator rates the couple in line with their average (neutral)
  • Negative value: adjudicator gives the couple fewer crosses (lower compatibility)
Used on: Predictor, Adjudicator profile, Dancer profile
Club Compatibility Adjudicating

Club compatibility measures whether an adjudicator's scoring patterns differ for couples from a particular club. It analyses the adjudicator's historical crosses for couples from specific clubs and compares them with their overall average.

The figure only shows how far an adjudicator's crosses for one club's couples differ from their own average. It describes a difference, not its cause.

Used on: Adjudicator profile, Adjudicator analysis
Correlation Adjudicating

Correlation measures the agreement between two adjudicators' assessments. High correlation means two adjudicators give crosses to similar couples; low correlation means independent assessment.

DanceStats.eu calculates adjudicator correlation using statistical methods that compare their shared assessments at competitions where they sat on the same panel.

Used on: Correlation
Panel score Adjudicating

Panel score is an aggregated metric measuring how compatible the entire adjudication panel is with a specific couple. It is calculated as the average compatibility of all adjudicators in the current panel with the given couple.

A value of 50 is neutral - the panel as a whole has no particular leaning. Above 50 means a more favourable panel, below 50 less favourable.

Final mark (placement) Adjudicating

In the final round of a competition, adjudicators do not give crosses but placement marks - each adjudicator assigns each couple a rank from 1 (best) to the number of finalists. The final order is determined by the sum of these marks.

The system for determining final placements from marks is defined by ČSTS/WDSF rules and uses the so-called skating system to resolve ties.

Head-to-Head (H2H) Statistics

Head-to-Head is a direct comparison of two couples based on their historical results at shared competitions. It measures how many times one couple has beaten the other.

H2H is the foundation of Model A and an important factor in Model B. The more shared competitions two couples have, the more reliable the H2H prediction.

Example: Couple A beat Couple B at 7 out of 10 shared competitions → H2H = 70% in favour of Couple A.

Prediction Models

Model A - H2H + Panel Analysis Model

A statistical model based on direct results between couples and detailed adjudicator analysis. Combines historical win/loss record (H2H) with the current panel's compatibility.

  • Head-to-Head: Calculates historical results between couples - how many times one couple beat the other at shared competitions
  • Adjudicator Compatibility: An adjudicator's specific affinity towards a couple minus their global generosity (how many crosses they give this couple vs. all couples)
  • Panel score: Average compatibility of the entire panel with a couple (50 = neutral). H2H probability is adjusted by the difference in panel compatibility

Best for: Couples with a rich history of shared competitions and a known panel.

Used on: Predictor
Model B - Elo Rating + 8 Factors Model

An Elo rating system (known from chess) adapted for dance sport with 8 adjustment factors. Blend: 60% Elo + 40% H2H (if history exists).

  • Momentum (form): Trend of recent results (rising/falling form)
  • Club Compatibility: Difference between an adjudicator's marks for a club's couples and their own average
  • Class (uncertainty): Promotion to a higher class widens the rating's uncertainty instead of deducting points (settles over ~3-5 competitions)
  • Age category transition: Effect of transitioning between age categories
  • Inactivity: Penalty for long breaks between competitions
  • PageRank: Strength of opponents the couple has beaten (transitive quality)
  • Transitive H2H: Indirect results (A beat B, B beat C → A stronger than C)
  • Partnership heritage: Rating transfer from previous partnerships of both dancers

The resulting score (0–100) indicates the percentage of opponents the couple should beat based on adjusted Elo.

Best for: New couples, unknown opponents, broader couple strength context.

Model C - Neural Network (Transformer) Model

A deep neural network (PyTorch Transformer) trained across federations rather than on a single one: the model carries a federation alongside every couple, so results from dozens of federations reinforce each other instead of being learned separately. Evaluates each discipline (STD/LAT) separately, considers 42 per-couple numerical features and a cross-attention layer over the adjudicator panel.

  • Elo rating + form: Current rating, peak rating, momentum, volatility, gap vs. field median, finals streak
  • Discipline activity: Starts in the discipline over 180/365 days and lifetime, age-category experience, new-couple flag
  • Marks and crosses: Cross-marking rate by adjudicators, average final placement, round advancement rate, win rate, recent form (last 3 competitions)
  • Championships: Participation and best placement at Czech championships
  • Ranking: ČSTS ranking position and discipline specialisation
  • Partnership: Current partnership length and stability, peak Elo from both dancers' previous partnerships (cold-start for new couples)
  • Context: Field strength, club statistics, class, age category, number of class transitions
  • Adjudicator analysis: Cross-attention over five features per adjudicator on the panel (couple cross-rate, global generosity, compatibility, shared competitions), familiarity with the current panel, head-to-head wins
  • Embeddings: Learned representations for couples, dancers, adjudicators and clubs. For new couples the representation is composed from both dancers' embeddings (cold-start).
  • Transformer attention: The model learns which factors are most important in any given situation. Incremental fine-tuning on new data without retraining from scratch.

Best for: Discovering hidden patterns, predictions for couples with rich history in a given discipline.

Used on: Predictor

Competition Structure

Dance class Class

A dance class (E, D, C, B, A, M) determines the level of a dance couple in the ČSTS system. Couples start in the lowest class and progress upward based on competition results.

  • E - entry class
  • D - beginners
  • C - intermediate
  • B - advanced
  • A - excellent
  • M - international class (highest)

Class vs. Elo rating

Class and Elo measure strength differently. Class is the federation's promotion system: an M-class couple earned its way through E→D→C→B→A→M, each promotion requiring concrete results. Elo, by contrast, starts everyone at 1500 regardless of class and is kept per discipline and per federation, not per class. Moving up a class never subtracts rating points; it only widens the uncertainty for a while. So a C couple with Elo 1750 isn't necessarily better than an M couple at Elo 1600: class tells you the level the couple has reached, Elo their current form. See also Elo Rating.

Used on: Elo Rating, Promotion
Promotion Class

A promotion is a couple's transition to a higher dance class. Promotion requirements are defined by ČSTS Competition Rules and depend on the number and quality of placements at competitions of the appropriate level.

DanceStats.eu tracks promotions and displays them on dancer profiles and club overviews. There are different promotion types - regular (meeting point thresholds), extraordinary, and transfers from other federations.

Used on: Dancer profile, Promotions
Age category Competition

The age category determines which group a couple competes in. Categories are defined by the birth year of both partners (the older partner's age is decisive).

Main categories: Children I/II, Juniors I/II, Youth, Under 21, Adults, Seniors I/II/III/IV/V. Transitioning between categories can temporarily affect a couple's performance.

Competition round Competition

A competition consists of several rounds. In preliminary rounds, adjudicators award crosses and qualifying couples advance. In the final, adjudicators give placement marks.

Typical structure: 1st round → 2nd round → quarter-final → semi-final → final. Special rounds include re-dance (repeat round for borderline couples) and hope round.

Federation Competition

DanceStats.eu aggregates data from dozens of dance federations. The most prominent ones:

  • ČSTS - Czech Dance Sport Federation (main Czech federation)
  • SZTŠ - Slovak Dance Sport Federation (main Slovak federation)
  • WDSF - World DanceSport Federation (international federation)
  • WDC-AL - WDC Amateur League (amateur arm of the World Dance Council)
  • DTV - German Dance Sport Federation (Deutscher Tanzsportverband)
  • SUT - Czech Dance Teachers Association (Czech federation)
  • DSI - DanceSportInfo (independent international database)

Plus further national federations across Europe and beyond (69 in total).

0.1