⚽ ConvocadosDocsllms.txtGitHub

Rating & Ladder Glossary

Canonical terms. Use these names everywhere; avoid the listed synonyms.

Rating math

Rating

A single number estimating a competitor's latent strength, used to predictfuture results. A rating is a forecast, not a scoreboard.Avoid: points, score, ELO (as a noun for the number).

Skill Rating

Convocados' lifetime, hidden Elo value used to balance teams. Never resets.Avoid: rank, level, MMR.

Expected score (E)

The probability-weighted points a player is forecast to take from a game, between 0 and 1; a draw counts as 0.5. Elo uses E = 1 / (1 + 10^((R_opp − R_you)/400)).Avoid: win probability (E is an average score, not P(win)).

K-factor

The maximum points a rating can move in one game; scales every update. Large K = fast, volatile ratings; small K = stable, slow. Avoid: multiplier, sensitivity.

Update rule

R′ = R + K · (S − E), where S is the actual score (1 win / 0.5 draw / 0 loss). Winning above expectation raises the rating; meeting expectation changes nothing.Avoid: formula, algorithm.

Rating period

A batch of games treated as occurring simultaneously, after which ratings update together. Common in Glicko systems. Avoid: round, matchday.

Rating deviation (RD)

The uncertainty of a rating estimate. High RD = little evidence (rating moves a lot); low RD = confident (rating moves little).Avoid: confidence, variance, error bar.

Volatility (σ)

Glicko-2's measure of how erratic a player's performances are; rises after unexpectedly strong or weak runs. Avoid: consistency, form.

Systems

Elo

The baseline system: one rating per player, fixed K, expected score from the rating gap. Assumes every rating is equally precise.Avoid: the Elo system's name as an acronym; it is a surname.

Glicko-2

Elo extended with a rating deviation and volatility per player, updated over rating periods. Handles provisional players and skill change without a manual provisional window. Avoid: Elo2, improved Elo.

TrueSkill

Microsoft's Bayesian model that tracks each skill as a Gaussian (mean + variance), models draws, and infers individual skill from team results.Avoid: Xbox rating.

OpenSkill

An open, faster implementation of the Weng–Lin Bayesian approximation, tuned for asymmetric multi-team games like TrueSkill.Avoid: open TrueSkill, clone.

Ladder & season layer

Season Rank

Convocados' visible, 0-based per-Season ladder. Derived on read, moves only in counted Season games, and soft-resets between Seasons.Avoid: MMR, skill rating, points.

Soft reset

Moving a player part-way back toward a seed value between Seasons, so rank is not inherited fully but skill is not discarded.Avoid: wipe, decay, restart.

Provisional

A player's first few counted games, during which the ladder shows an unlock counter instead of a Rank because the estimate is not yet stable.Avoid: placement, calibration, unranked.

Tier

A named band of Rank (e.g. Bronze…Master) whose edges are fixed at launch. Gives a ladder its sense of place and progression. Avoid: division, league, medal.

Placement

A deliberate, high-K set of games that establishes a starting Rank. Distinct from Provisional, which is passive. Avoid: qualifiers, seeding games.