TrueSkill - Through Time

Bayesian skill rating over a time axis.

Where plain TrueSkill gives each competitor one running estimate, TrueSkill Through Time treats a whole history as a single model and infers skill at every point in time. Evidence flows both directions: a result today sharpens the estimate of who someone was last year, so early estimates stop being frozen guesses and comparisons across eras become meaningful.

A Rust port of TrueSkillThroughTime.py.

Install

[dependencies]
trueskill-tt = "0.8"

Optional features, both off by default:

• approx — approx's equality traits for Gaussian. Useful in tests.
• rayon — parallelises the within-slice sweep and the per-slice passes of learning_curves / log_evidence. Results stay bit-identical regardless of worker count; just determinism asserts it at 1, 2, 4 and 8 threads.

Quickstart

Record results, converge, then read off skills.

use trueskill_tt::History;

let mut history = History::default();

history.record_winner(&"alice", &"bob", 1)?;
history.record_winner(&"bob", &"carol", 2)?;
history.record_winner(&"alice", &"carol", 3)?;

history.converge()?;

let alice = history.current_skill("alice").unwrap();
assert!(alice.mu() > 0.0, "alice won every game she played");
# Ok::<(), trueskill_tt::InferenceError>(())

The third argument is the time. It is what makes this Through Time rather than plain TrueSkill: skill is inferred at each of those moments, not once at the end. learning_curve reads the whole trajectory back.

# use trueskill_tt::History;
# let mut history = History::default();
# history.record_winner(&"alice", &"bob", 1)?;
# history.record_winner(&"bob", &"carol", 2)?;
# history.record_winner(&"alice", &"carol", 3)?;
# history.converge()?;
// `None` means the key is unknown; `Some(vec![])` means known but unplayed.
let curve = history.learning_curve("alice").unwrap();
for (time, skill) in &curve {
    println!("t={time}: {:.2} ± {:.2}", skill.mu(), skill.sigma());
}

// Everyone's latest posterior in one pass — the leaderboard query.
let latest = history.current_skills();
assert_eq!(latest.len(), 3);
# Ok::<(), trueskill_tt::InferenceError>(())

Teams, rankings and draws

Anything beyond one-versus-one goes through the fluent event builder. An event is only recorded by the terminal .commit().

use trueskill_tt::History;

let mut history = History::builder().p_draw(0.1).build();

history
    .event(1)
    .team(["alice", "bob"])
    .team(["carol", "dave"])
    .ranking([0, 1])   // lower is better; equal values are a tie
    .commit()?;

history.converge()?;
# Ok::<(), trueskill_tt::InferenceError>(())

A tie needs a positive p_draw. A p_draw of zero asserts draws cannot happen, so a tied result has no representable likelihood and is rejected rather than fitted to something else:

use trueskill_tt::{History, InferenceError};

let mut history = History::default(); // p_draw defaults to 0.0
let err = history.record_draw(&"alice", &"bob", 1).unwrap_err();
assert!(matches!(err, InferenceError::TieWithoutDrawProbability { .. }));

This also catches Outcome::winner(w, n) for three or more teams, which ties every loser.

Which entry point?

You want to	Use
One match, two competitors	record_winner / record_draw
Teams, explicit ranks, scores, per-member weights	history.event(t)…commit()
A batch you already have as values	add_events(iter)
Score a hypothetical with no history at all	Game

Game is the odd one out and worth being explicit about: it is a single match's factor graph, it does not participate in a History, and nothing it computes is remembered. Reach for it to evaluate a matchup in isolation; reach for History for everything that accumulates.

converge is strict

converge returns Err(NotConverged) if the sweep hits max_iter with the step still above epsilon, and Err(NonFiniteResult) if a sweep produces NaN.

It used to return Ok with converged: false, which was the worst available shape. A fit that stops short is wrong by a little: every posterior is finite, the ordering looks sensible, and nothing about the output says the numbers were still moving. Detection was opt-in, and let _ = h.converge() silently opted out — which is how a real defect hid in this crate's own test suite.

The default max_iter is high enough that reaching it means something is genuinely wrong rather than that the history is large; the loop exits at epsilon long before, so raising the cap costs nothing when it is not needed. Use converge_partial when a deliberately capped, unconverged fit is the point.

Predictions are strict for the same reason: every predict_* method reads skills through one gate that refuses a NaN-poisoned fit, rather than returning a plausible number computed from it.

Drift

Skill drift models how a competitor's true skill can change between appearances. Each time they reappear after a gap, their skill uncertainty is widened by the drift model before the new evidence is incorporated.

Drift is represented by the Drift trait (src/drift.rs), generic over the history's time type:

pub trait Drift<T: Time>: Copy + Debug + Send + Sync {
    fn variance_delta(&self, from: &T, to: &T) -> f64;
    fn variance_for_elapsed(&self, elapsed: i64) -> f64;
}

Both methods return the amount to add to σ², not to σ. variance_delta works from two timestamps; variance_for_elapsed takes an already-computed elapsed count, and is used on the paths that cache it. Gaussian::forget applies the result entirely in variance space — from_mv(mu, variance() + variance_delta) — taking no square root.

That block is a quotation rather than a doctest. The custom-drift example below is compiled by CI, so it is what actually pins the signature.

ConstantDrift

The built-in ConstantDrift implements a linear random walk — skill uncertainty grows proportionally to time:

variance_delta = elapsed * γ²

This is the standard TrueSkill Through Time model. Pass a ConstantDrift::new(gamma) when constructing a Rating:

use trueskill_tt::{ConstantDrift, Gaussian, Rating};

// gamma = 0.1 means skill can shift ~0.1 per time unit.
let rating: Rating<i64, ConstantDrift> =
    Rating::new(Gaussian::from_ms(0.0, 6.0), 1.0, ConstantDrift::new(0.1));

assert_eq!(rating.drift().gamma(), 0.1);

The type annotation is load-bearing: ConstantDrift implements Drift<T> for every T: Time, so without it T is ambiguous.

Custom drift

Implement Drift<T> to express any other model. For example, a drift that saturates after a long absence, with uncertainty growing as the square root of elapsed time instead of linearly:

use trueskill_tt::{Drift, Gaussian, History, Rating, Time};

#[derive(Clone, Copy, Debug)]
struct SqrtDrift {
    gamma: f64,
}

impl<T: Time> Drift<T> for SqrtDrift {
    fn variance_delta(&self, from: &T, to: &T) -> f64 {
        let elapsed = from.elapsed_to(to).max(0) as f64;
        elapsed.sqrt() * self.gamma * self.gamma
    }

    fn variance_for_elapsed(&self, elapsed: i64) -> f64 {
        (elapsed.max(0) as f64).sqrt() * self.gamma * self.gamma
    }
}

// On a single Rating:
let rating: Rating<i64, SqrtDrift> =
    Rating::new(Gaussian::from_ms(0.0, 6.0), 1.0, SqrtDrift { gamma: 0.5 });

// Or for a whole History, via the builder:
let history = History::builder().drift(SqrtDrift { gamma: 0.5 }).build();

assert_eq!(rating.beta(), 1.0);
assert_eq!(history.log_evidence(), 0.0);

HistoryBuilder::drift is the only way to set a history's drift model; there is no gamma() shorthand. The default is ConstantDrift::new(GAMMA).

Per-competitor drift

A History has one drift model, but individual competitors can scale it. Member::with_drift_scale(s) multiplies the drift variance that competitor accumulates, so s is in the same units as gamma: ConstantDrift::new(g) at scale s behaves exactly as ConstantDrift::new(g * s) would, for that competitor alone.

0.0 pins a competitor still. That is what makes a fixed reference point expressible in the same graph as moving competitors — a bot at a known strength, a rating floor, a course difficulty:

use trueskill_tt::{ConstantDrift, Event, History, Member, Outcome, Team};

let mut h = History::builder().drift(ConstantDrift::new(0.1)).build();

h.add_events(vec![Event {
    time: 0,
    teams: [
        Team::with_members([Member::new("player")]),
        // A course does not improve. Pin it, and the round's evidence
        // lands on the player instead of being split between the two.
        Team::with_members([Member::new("layout_7").with_drift_scale(0.0)]),
    ]
    .into_iter()
    .collect(),
    outcome: Outcome::winner(0, 2),
}])
.unwrap();

h.converge().unwrap();

Like with_prior, the scale is competitor configuration, not a per-event value: it applies to the competitor for the whole history, and it applies whenever it is supplied — including on a key the history already knows. Configuring one late still refits the whole history rather than taking effect only from that event onward, because converge refits from competitor state. Repeating the same value is inert; supplying two different values for one competitor within a single batch is InferenceError::ConflictingCompetitorConfig, since events in a batch have no order. The scale must be finite and non-negative; ingestion otherwise fails with InferenceError::InvalidParameter.

The fluent EventBuilder reaches this too: .team([...]) is the common case and leaves both unset, while .members([...]) takes Member values directly, so h.event(t).members([Member::new("layout_7").with_drift_scale(0.0)]) is equivalent to the typed shape above.

Scored outcomes

Use Outcome::scores([...]) when you have continuous per-team scores rather than just ranks. Adjacent score margins flow into a MarginFactor that adds soft Gaussian evidence about the latent performance diff. Configure HistoryBuilder::score_sigma(σ) to control how much you trust the margins (smaller σ = more trust).

use trueskill_tt::History;

let mut h = History::builder().score_sigma(2.0).build();
h.event(1)
    .team(["alice"])
    .team(["bob"])
    .scores([21.0, 9.0])
    .commit()
    .unwrap();
h.converge().unwrap();

Prediction

predict_outcome gives the full distribution over finishing orders. Each entry is a rank vector in the same shape Outcome::ranking takes — equal ranks mean a tie — so an outcome feeds straight back into inference.

use trueskill_tt::History;

let mut h = History::builder().p_draw(0.1).build();
h.record_winner(&"alice", &"bob", 1).unwrap();
h.converge().unwrap();

let p = h.predict_outcome(&[&[&"alice"], &[&"bob"]]).unwrap();

// Probabilities are exhaustive and disjoint, so they sum to one.
assert!((p.total() - 1.0).abs() < 1e-6);

let (best, likelihood) = p.most_likely().unwrap();
println!("most likely: {best:?} at {likelihood:.3}");
println!("draw:        {:.3}", p.probability_of(&[0, 0]));

Supports any number of teams. Because the outcome space grows factorially, the full distribution is capped at MAX_PREDICTED_TEAMS; two cheaper entry points stay available at any size:

• predict_win_probabilities(teams) — P(team i finishes strictly first), quadratic in team count.
• predict_ranking(teams, ranks) — one specific finishing order.

Unknown keys are an error by default, not a silent omission: a team the history has never seen cannot produce a confident-looking probability. The error names the key, and every key must already be known — pre-filter with lookup or current_skill if your caller cannot guarantee that.

If predicting for competitors you have never seen is the point rather than a mistake, say so once:

use trueskill_tt::{History, UnknownKeys};

let h = History::builder().unknown_keys(UnknownKeys::Prior).build();

An unknown competitor is then answered from the configured prior, which is the honest reading — you have no evidence about them — and correctly widens a team that contains one. There is deliberately no "skip the member" mode: a team's performance is the sum of its members, so dropping one would make the model more certain because it knows less.

Asking about one competitor

Gaussian answers tail questions directly, which is what a stopping rule needs:

use trueskill_tt::History;

let mut h = History::default();
h.record_winner(&"alice", &"bob", 1).unwrap();
h.converge().unwrap();

let skill = h.current_skill("alice").unwrap();

// "How sure am I that this is below the cutoff?" — a probability, not a
// `mu + z * sigma` band whose confidence drifts as sigma changes.
let _ = skill.probability_below(20.0);

// Use this rather than `1.0 - probability_below(x)`: the complement cancels
// away every digit in the upper tail, which is where a stopping rule lives.
let _ = skill.probability_above(30.0);

Which match to play next

quality() measures whether a matchup is fair. That is not the same as whether it is informative, and the two only coincide for two evenly matched competitors. When each observation costs something, ask expected_information_gain instead — the outcome-weighted divergence between what you believe now and what you would believe afterwards.

use trueskill_tt::History;

let mut h = History::default();
for t in 1..=10 {
    h.record_winner(&"veteran", &"regular", t).unwrap();
    h.record_winner(&"regular", &"veteran", t + 100).unwrap();
}
h.record_winner(&"veteran", &"newcomer", 500).unwrap();
h.converge().unwrap();

let settled = h.expected_information_gain(&[&[&"veteran"], &[&"regular"]]).unwrap();
let unknown = h.expected_information_gain(&[&[&"veteran"], &[&"newcomer"]]).unwrap();

// Playing the newcomer teaches you more than replaying a settled rivalry.
assert!(unknown > settled);

The result is in nats, and is bounded by the entropy of the outcome: at most ln 2 ≈ 0.693 for a two-way result, ln 3 once draws are possible, ln k for k outcomes. A value near zero means you already know how it ends.

This costs one full inference pass per possible outcome, so it is far more expensive than quality(). Scoring every pairing among n competitors is O(n² × outcomes) passes — shortlist with quality() or predict_win_probabilities first, then score only the shortlist.

Other implementations

• ttt-scala
• ChessAnalysis #F
• TrueSkillThroughTime.jl
• TrueSkillThroughTime.R
• TrueSkill Through Time: Revisiting the History of Chess
• TrueSkill Through Time. The full scientific documentation

Status

Every box on the old todo list is ticked, so it has been retired; open work lives in the issue tracker instead. The crate is in use and the API is still moving — breaking changes are batched into minor releases rather than dribbled out, and CHANGELOG.md records them.

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.