top of page

Introducing the SEER Power Index (SPI)



SEER is a proprietary data analytics platform developed by TRVtools that leverages machine learning to forecast TRV champ outcomes. Every hour, SEER analyzes and re-analyzes over 15 million hit-by-hit interactions to rate each TRV champ across a panel of 5+ underlying traits (i.e. Strength, Defense, Attack, Accuracy, Dodge, and more). SEER then forecasts future performance outcomes by identifying historical statistical relationships amongst these traits (and other variables) and observed results across the thousands of champs that have entered the arena.


Beginning today, this performance projection will be published on TRVtools champ pages (and on TRVtools Champion Scouting Reports) as the SEER Power Index (SPI; formerly referred to as the 'SEER Elo Prediction'). SPI, akin to a power ranking, is TRVtools’ minimally biased statistical assessment as to how each champ stacks up against the competition. Champ SPI scores range from 0 (lowest win probability) to 100 (highest win probability) and are unlocked by 50 battles in stance. These measures are far less volatile than Elo-based performance measures and reach equilibrium far more quickly than Elo (usually before the 50 battle unlock threshold). SPI scores will be updated daily.

In essence, SPI is a leading indicator of future champ outcomes that outperforms Elo as a stable measure of champ quality. SPI is also highly useful in identifying stance preferences, because each stance is assessed and rated independently of the others. For most champs, SPI scores of non-preferred stances tend to cluster in a relatively tight range, whereas preferred stances constitute outliers to the upside (e.g., see Swiftfury's SPI scores above).


SPI values DO NOT simply represent percentiles. They are more informative: differences in SPI amongst champs translate directly to win probability according to the equation:


For example, all else equal, a champ with an SPI value of 90 would have a 58% probability of defeating a champ with an SPI of 70 on any given interaction*.


*Note that SPI offers a neutral champ quality baseline, and other factors (e.g. hot/cold streaks; differences in character class preferences, etc.) may cause deviations in observed win probability between two particular champs on any given interaction.

The inputs to SPI are also informative, and a subset are represented graphically in the TRVtools Champion Scouting Reports (CSRs). Relative to raw damage analytics (e.g. damage dealt/received, mean net damage, etc.), SEER-inferred traits are far more predictive of champ outcomes, because SEER accounts for opponent quality.


For example, 175 damage dealt to a high-quality defender is more impressive than the same amount dealt to a poor defender. In fact, 125 damage dealt to a tank may be more impressive than 175 dealt to a squib. SEER Strength, Defense, Accuracy, Dodge, and Attack measures control for these nuances introduced by opponent quality and random number generation.


Accordingly, SEER traits are more “pure” than raw descriptive statistics, which are confounded by noise and inconsistency. These measures (scaled to a champ's character class) can be used to inform summoning decisions.

SEER exists in a perpetual state of learning and self-assessment. SEER is fully automated and continues to refine its performance assessments as more blood is spilled in the arena. More to come!


Scuer.eth

TRVtools.com


Recent Posts

See All
bottom of page