docs: refresh README and CLAUDE.md; add ingest benchmark

The CLAUDE.md architecture section still described the pre-redesign engine:
its data flow named `Batch`, `Agent`, `Player` and `message.rs`, none of
which have existed since T2, and the public API it listed did not match
`lib.rs`. It is the first thing a fresh session reads, so it was actively
misleading. Rewritten against the current module layout, with the invariants
that are easy to violate — ties needing a positive `p_draw`, NaN never being
convergence, log-space evidence, color contiguity, `forbid(unsafe_code)`,
and ingestion-order equivalence — written down.

The README Todo list had five entries that were already done, including
"Time needs to be an enum": `Time` has been a trait since T2, and the
`batch::compute_elapsed()` it pointed at no longer exists. The genuinely
open item — cross-checking `quality()` against sublee/trueskill — stays.

`benches/ingest.rs` measures one-event-per-call against a single batched
call. The rest of the suite only measured batched construction, which is why
the quadratic fixed earlier on this branch went unnoticed for so long.

`TimeSlice::log_evidence` also hashes its target set once instead of
scanning the slice per player per event, so `log_evidence_for` with many
keys is no longer quadratic.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnsaJg74eNSva3PJjK2eej

CLAUDE.md

@@ -5,42 +5,96 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 ## Commands
 
 ```bash
-cargo build                          # Build the library
-cargo test --lib                     # Run all library tests
-cargo test --lib <test_name>         # Run a single test by name
-cargo test --lib -- --nocapture      # Run tests with stdout output
-cargo clippy                         # Lint
-cargo bench                          # Run benchmarks (criterion)
+just test                            # Full suite across every feature combination CI checks
+just check                           # Fast inner loop: cargo test --features approx
+just lint                            # clippy, warnings denied
+just fmt                             # ALWAYS nightly — rustfmt.toml uses nightly-only options
+just determinism                     # Bit-identical posteriors at RAYON_NUM_THREADS 1/2/4/8
+just ci                              # Everything CI runs
+cargo test --lib <test_name>         # A single test by name
+cargo bench                          # Criterion benchmarks
 ```
 
-The `approx` feature enables `approx::AbsDiffEq` for `Gaussian`:
-```bash
-cargo test --features approx
-```
+**Run tests in release too.** `debug_assert!` is compiled out there, and that
+is where several defects have hidden — a debug-only run is not evidence.
+`just test` includes a release job.
+
+### Feature flags
+
+- `approx` — `approx::AbsDiffEq` etc. for `Gaussian`. Most numerical goldens need it.
+- `rayon` — opt-in parallel within-slice sweep and per-slice query passes.
 
 ## Architecture
 
-This is a Rust port of [TrueSkillThroughTime.py](https://github.com/glandfried/TrueSkillThroughTime.py) — a Bayesian skill rating system that tracks skill evolution over time using Gaussian message passing.
+A Rust port of [TrueSkillThroughTime.py](https://github.com/glandfried/TrueSkillThroughTime.py):
+Bayesian skill rating that infers skill at every point in time, propagating
+evidence both forward and backward across a history.
 
 ### Data flow
 
 ```
-History  →  Batch[]  →  Game[]  →  teams/players
+History  →  TimeSlice[]  →  Event[]  →  Team[] → Item[]
+                                  ↓
+                     Game (factor graph) → Schedule → BuiltinFactor[]
 ```
 
-- **`History`** (`history.rs`) — top-level container. Organizes games by time into `Batch`es, runs forward/backward message passing across batches, and exposes `learning_curves()` and `log_evidence()`.
-- **`Batch`** (`batch.rs`) — all games at a single time step. Runs `iteration()` to update skill estimates via `Game::posteriors()`, collecting `Skill` distributions per player.
-- **`Game`** (`game.rs`) — a single match. Given teams (slices of `Gaussian`), computes posterior skill distributions using Gaussian factor graphs and `message.rs` helpers.
-- **`Agent`** (`agent.rs`) — wraps a `Player` with temporal state (`last_time`, `message`). `receive()` applies time-decay (`gamma`) when the player reappears after a gap.
-- **`Player`** (`player.rs`) — static configuration: prior `Gaussian`, `beta` (performance noise), `gamma` (skill drift per time unit).
-- **`Gaussian`** (`gaussian.rs`) — core probability type. Stored as natural parameters (`pi = 1/sigma²`, `tau = mu/sigma²`). Arithmetic ops implement message multiplication/division in the factor graph.
-- **`message.rs`** — `TeamMessage` and `DiffMessage`: intermediate factor graph messages used inside `Game`.
-- **`MarginFactor`** (`factor/margin.rs`) — Gaussian observation factor on a diff variable; engaged by `Outcome::Scored`.
-- **`lib.rs`** — exports the public API (`Game`, `Gaussian`, `History`, `Player`) and standalone functions (`quality()`, `pdf()`, `cdf()`, `erfc()`). Also defines global defaults: `MU=0.0`, `SIGMA=6.0`, `BETA=1.0`, `GAMMA=0.03`, `P_DRAW=0.0`, `EPSILON=1e-6`, `ITERATIONS=30`.
+- **`History`** (`history.rs`) — top level. Interns keys, groups events into
+  `TimeSlice`s by time, runs the forward/backward sweep in `converge()`, and
+  answers `learning_curves()`, `current_skill()`, `log_evidence()`,
+  `predict_quality()`, `predict_outcome()`. Built via `HistoryBuilder`.
+- **`TimeSlice`** (`time_slice.rs`) — all events at one time. Owns a
+  `SkillStore` and a `ScratchArena`; `iteration()` sweeps its events, using
+  `ColorGroups` to partition independent ones.
+- **`Event`** (`time_slice.rs`) — one match. `compute()` runs inference reading
+  skills immutably; `apply()` folds the result back. The split is what lets a
+  color group run in parallel with no `unsafe`.
+- **`Game`** (`game.rs`) — a single match's factor graph. `run_chain` builds the
+  diff chain between rank-adjacent teams and drives it to convergence.
+- **`Gaussian`** (`gaussian.rs`) — natural parameters (`pi = 1/sigma²`,
+  `tau = mu/sigma²`). `Mul`/`Div` are the EP product/cavity: pure adds and
+  subtracts. Variance-space ops (`Add`, `Sub`, `exclude`, `forget`) go through
+  `from_mv`/`variance()` and take no square root.
+- **`factor/`** — `TeamSumFactor`, `RankDiffFactor`, `TruncFactor` (ranked),
+  `MarginFactor` (scored), over a flat `VarStore`. `BuiltinFactor` dispatches
+  by enum rather than `dyn`.
+- **`Schedule`** (`schedule.rs`) — drives factor propagation. `EpsilonOrMax` is
+  the only implementation.
+- **`Competitor`** (`competitor.rs`) — per-history temporal state (`message`,
+  `last_time`). **`Rating`** (`rating.rs`) — static config (prior, `beta`, drift).
+- **`storage/`** — `SkillStore` (per slice) and `CompetitorStore` (per history),
+  both dense `Vec`s indexed by `Index`.
+- **`KeyTable`** (`key_table.rs`) — user key ↔ `Index`, both directions O(1).
+- **`Drift`** (`drift.rs`) / **`Time`** (`time.rs`) — traits. `Time` is a *trait*
+  (`i64`, `Untimed`), not an enum.
+- **`lib.rs`** — public exports, global defaults (`MU`, `SIGMA`, `BETA`,
+  `GAMMA`, `P_DRAW`, `EPSILON`, `ITERATIONS`), and the standalone `quality()`,
+  `cdf()`, `erfc()`.
 
-### Key design points
+### Invariants worth knowing
 
-- `History` uses `IndexMap<K>` (defined in `lib.rs`) to map arbitrary player keys to `Agent` state.
-- Convergence is measured by the maximum `delta()` across all skill distributions; iteration stops when below `EPSILON` or after `ITERATIONS` rounds.
-- The `approx` feature gates `AbsDiffEq` on `Gaussian` for use in tests — the feature is optional and only needed for approximate equality assertions.
-- `time` in `History`/`Batch` is currently an `f64`; the README notes it needs to become an enum to support richer temporal states.
+- **A tie needs `p_draw > 0`.** With `p_draw == 0.0` the truncation margin is
+  zero and the two-sided tie update evaluates `0/0`. Ingestion rejects such
+  events with `InferenceError::TieWithoutDrawProbability`. This includes
+  `Outcome::winner(w, n)` for `n >= 3`, which ties every loser.
+- **NaN is never convergence.** Comparisons against NaN are all false, so
+  `tuple_gt` reads NaN as "below epsilon". Use `step_converged` /
+  `step_is_finite`, never `!tuple_gt(..)` alone.
+- **Evidence accumulates in log space.** A linear product over a long diff
+  chain underflows to zero, and `ln(0)` is `-inf`.
+- **Colors are contiguous.** `recompute_color_groups` reorders events so each
+  color occupies one range; `ColorGroups::groups_are_contiguous` asserts it.
+- **The crate is `#![forbid(unsafe_code)]`.** Keep it that way.
+- **Ingestion order must not change the answer.** Events added one at a time
+  must converge to the same fixed point as the same events batched — see
+  `tests/ingestion_equivalence.rs`.
+
+### Testing notes
+
+- Numerical goldens are cross-validated against the Python/Julia reference.
+  Some are *convergence residuals*, not exact values; treat a small movement
+  as suspicious but check whether the new value is closer to the analytic
+  truth (symmetric fixtures converge to their prior mean exactly) before
+  assuming a regression.
+- `tests/degenerate_inputs.rs` covers empty/boundary/error paths,
+  `tests/ingestion_equivalence.rs` covers batching order, `tests/quality.rs`
+  covers N-group quality, `tests/determinism.rs` covers thread counts.

Cargo.toml

@@ -33,6 +33,10 @@ harness = false
 name = "scored"
 harness = false
 
+[[bench]]
+name = "ingest"
+harness = false
+
 [dependencies]
 approx = { version = "0.5.1", optional = true }
 rayon = { version = "1", optional = true }

README.md

@@ -96,8 +96,8 @@ h.converge().unwrap();
 
 - [x] Implement approx for Gaussian
 - [x] Add more tests from `TrueSkillThroughTime.jl`
-- [ ] Add tests for `quality()` (Use [sublee/trueskill](https://github.com/sublee/trueskill/tree/master) as reference)
-- [ ] Benchmark Batch::iteration()
-- [ ] Time needs to be an enum so we can have multiple states (see `batch::compute_elapsed()`)
-- [ ] Add examples (use same TrueSkillThroughTime.(py|jl))
-- [ ] Add Observer (see [argmin](https://docs.rs/argmin/latest/argmin/core/trait.Observe.html) for inspiration)
+- [x] Generalise a time axis — `Time` is now a trait (`Untimed`, `i64`), not an enum
+- [x] Add examples (`examples/atp.rs`, `examples/scored.rs`)
+- [x] Add Observer (`Observer` / `NullObserver`)
+- [x] Benchmark the inference loop (`benches/batch.rs`, `benches/history_converge.rs`, `benches/ingest.rs`)
+- [ ] Cross-check `quality()` against [sublee/trueskill](https://github.com/sublee/trueskill/tree/master) — N-group support works and is covered by invariants, but no reference values are asserted

benches/ingest.rs

@@ -0,0 +1,62 @@
+//! Ingestion cost: one event per call versus one batched call.
+//!
+//! The rest of the suite only measures batched construction, which is why a
+//! quadratic in the incremental path went unnoticed — `record_winner` and
+//! `event(..).commit()` each ingest a single event, so a caller looping over a
+//! match feed takes that path.
+
+use std::hint::black_box;
+
+use criterion::{BenchmarkId, Criterion, criterion_group, criterion_main};
+use smallvec::smallvec;
+use trueskill_tt::{Event, History, Member, Outcome, Team};
+
+fn events(n: usize, time: i64) -> Vec<Event<i64, String>> {
+    (0..n)
+        .map(|i| Event {
+            time,
+            teams: smallvec![
+                Team::with_members([Member::new(format!("p{}", 2 * i))]),
+                Team::with_members([Member::new(format!("p{}", 2 * i + 1))]),
+            ],
+            outcome: Outcome::winner(0, 2),
+        })
+        .collect()
+}
+
+fn bench_ingest(c: &mut Criterion) {
+    let mut group = c.benchmark_group("ingest");
+
+    for n in [250usize, 500, 1000] {
+        group.bench_with_input(BenchmarkId::new("one-at-a-time", n), &n, |b, &n| {
+            b.iter_batched(
+                || events(n, 0),
+                |evs| {
+                    let mut h: History<i64, _, _, String> = History::builder_with_key().build();
+                    for ev in evs {
+                        h.add_events(std::iter::once(ev)).unwrap();
+                    }
+                    black_box(h.time_slices_len())
+                },
+                criterion::BatchSize::SmallInput,
+            );
+        });
+
+        group.bench_with_input(BenchmarkId::new("single-batch", n), &n, |b, &n| {
+            b.iter_batched(
+                || events(n, 0),
+                |evs| {
+                    let mut h: History<i64, _, _, String> = History::builder_with_key().build();
+                    h.add_events(evs).unwrap();
+                    black_box(h.time_slices_len())
+                },
+                criterion::BatchSize::SmallInput,
+            );
+        });
+    }
+
+    group.finish();
+}
+
+criterion_group!(benches, bench_ingest);
+criterion_main!(benches);

src/history.rs

@@ -312,6 +312,12 @@ impl<T: Time, D: Drift<T>, O: Observer<T>, K: Eq + Hash + Clone> History<T, D, O
         step
     }
 
+    /// Number of distinct time slices in the history.
+    #[must_use]
+    pub fn time_slices_len(&self) -> usize {
+        self.time_slices.len()
+    }
+
     /// Learning curves for all competitors, keyed by their user-facing key.
     ///
     /// Note: `key(idx)` is O(n) per lookup; this method is therefore O(n²)

src/time_slice.rs

@@ -587,6 +587,9 @@ impl<T: Time> TimeSlice<T> {
         forward: bool,
         agents: &CompetitorStore<T, D>,
     ) -> f64 {
+        // Hashed once rather than scanned per player per event, so a
+        // `log_evidence_for` with many keys is not quadratic.
+        let target_set: std::collections::HashSet<Index> = targets.iter().copied().collect();
         // log_evidence is infrequent; a local arena avoids needing &mut self.
         let mut arena = ScratchArena::new();
 
@@ -636,7 +639,7 @@ impl<T: Time> TimeSlice<T> {
                         .teams
                         .iter()
                         .flat_map(|team| &team.items)
-                        .any(|item| targets.contains(&item.agent))
+                        .any(|item| target_set.contains(&item.agent))
                 })
                 .map(|event| run_event(event, &mut arena))
                 .sum()
@@ -648,7 +651,7 @@ impl<T: Time> TimeSlice<T> {
                         .teams
                         .iter()
                         .flat_map(|team| &team.items)
-                        .any(|item| targets.contains(&item.agent))
+                        .any(|item| target_set.contains(&item.agent))
                 })
                 .map(|event| event.log_evidence)
                 .sum()