docs: state filtered accessor cost and evidence semantics precisely

filtered_learning_curve's signature mirrors learning_curve, which is cheap
per key — so the mirroring trained callers to assume this one is too. It is
a full forward pass per call, making the natural loop over competitors
O(competitors * events). The doc now says so in complexity terms and points
multi-key callers at the plural form.

filtered_log_evidence claimed each event is scored "using only what was
known before it". That is exact for a slice holding one event, but events
sharing a timestamp inform each other through the within-slice sweep, so
the honest claim is "before that time". The behaviour is deliberate and
matches log_evidence's own convention; only the promise was too strong.

This branch exists because a feature's documentation was quietly false.
Shipping it with two more overstated doc comments would be a poor joke.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01T5SYDExxL4vZgvunrcNSMc

src/history.rs

@@ -383,7 +383,9 @@ impl<T: Time, D: Drift<T>, O: Observer<T>, K: Eq + Hash + Clone> History<T, D, O
     /// time — "what we knew then". Contrast `learning_curves`, whose points
     /// are smoothed and so incorporate rounds played later.
     ///
-    /// Runs a full forward pass per call and caches nothing.
+    /// Runs a full forward pass per call and caches nothing. This is the
+    /// entry point for multi-key work — see `filtered_learning_curve` for
+    /// why calling that once per key is far more expensive.
     pub fn filtered_learning_curves(&self) -> HashMap<K, Vec<(T, Gaussian)>> {
         let mut data: HashMap<K, Vec<(T, Gaussian)>> = HashMap::new();
 
@@ -401,9 +403,11 @@ impl<T: Time, D: Drift<T>, O: Observer<T>, K: Eq + Hash + Clone> History<T, D, O
     /// Filtered learning curve for a single key: (time, posterior) pairs in
     /// time order.
     ///
-    /// Runs the same full pass as `filtered_learning_curves` and keeps one
-    /// key, so asking for several keys individually costs a pass each — use
-    /// the plural form for that.
+    /// Despite mirroring `learning_curve`'s signature, this is not the cheap
+    /// per-key lookup that method is: it runs a full forward pass, O(events),
+    /// discarding every posterior but the requested key's. N keys fetched
+    /// this way costs O(N * events); use `filtered_learning_curves` for
+    /// multi-key work instead — it computes the same pass once.
     pub fn filtered_learning_curve<Q>(&self, key: &Q) -> Vec<(T, Gaussian)>
     where
         K: Borrow<Q>,
@@ -484,10 +488,13 @@ impl<T: Time, D: Drift<T>, O: Observer<T>, K: Eq + Hash + Clone> History<T, D, O
 
     /// Total log-evidence under forward-only (filtering) information.
     ///
-    /// Each event is scored using only what was known before it, which is the
-    /// right quantity for prequential scoring and model comparison. Contrast
-    /// `log_evidence`, whose per-event priors carry information from events
-    /// that had not happened yet.
+    /// Each event is scored using only what was known before that *time*,
+    /// which is the right quantity for prequential scoring and model
+    /// comparison. Events sharing a timestamp still inform each other
+    /// through the within-slice sweep, so within one slice this is not a
+    /// guarantee that event A is scored independently of simultaneous event
+    /// B. Contrast `log_evidence`, whose per-event priors carry information
+    /// from events that had not happened yet.
     ///
     /// Runs a full forward pass per call and caches nothing. The result does
     /// not depend on whether `converge` has been called.