add more train stats per driver

docs/javascript-architecture.md

@@ -476,6 +476,9 @@ Owns:
 - daily summaries, controlled-lap streak, and recent personal-best gain
 - signed personal pace and consistency trends from the latest 30 laps versus the preceding 30, with a minimum of 10 laps in each window
 - derived automatic achievements for sessions, lap totals, training days, controlled streaks, and a completed five-minute result
+- active 30-minute pace buckets using a trimmed average and omitting empty clock periods
+- persisted class-specific setup markers and 30-minute before/after comparisons
+- extended stint analysis with normal pace, best 10-lap race pace, mistake count, consistency, and start/end change
 - rule-based practice recommendation selection
 - automatic and custom milestone progress calculation
 
@@ -500,6 +503,9 @@ Rules:
 - a boundary lap whose `lapMs` contains the full break is excluded from pace metrics
 - duration-result ranking compares lap count first and elapsed result time second
 - a duration result is valid only after crossing the target and includes a final lap whose start is before the target cutoff
+- a stint needs at least 3 valid pace laps for consistency/ranking and at least 10 for Race pace 10
+- a stint mistake is a lap more than 15 percent slower than that stint's median
+- setup comparisons need at least 3 valid laps in each 30-minute side of the marker
 
 ### `src/training_views.js`
 Type:
@@ -514,10 +520,12 @@ Owns:
 - compact live-practice card and primary personal-profile action for the selected driver
 - personal trend and achievement cards that never combine classes
 - recent club progress feed ordered by latest activity, without position numbers or cross-driver performance scoring
+- collapsed Extra statistics with pace heatmap, setup-marker actions, weekday pace, featured stints, and an expandable full stint table
 
 Mutations:
 - replaces or appends entries in `state.training.goals`
 - creates, rotates, enables, or disables entries in `state.training.shares`
+- creates and removes entries in `state.training.markers`
 - always calls injected `saveState()` before rerender
 
 Route behavior:
@@ -1758,7 +1766,8 @@ This bucket is the most dynamic part of state and is mutated heavily during live
 {
   goals: TrainingGoal[],
   ledger: TrainingLap[],
-  shares: TrainingShare[]
+  shares: TrainingShare[],
+  markers: TrainingMarker[]
 }
 ```
 
@@ -1812,6 +1821,21 @@ Share rules:
 }
 ```
 
+`TrainingMarker`:
+
+```js
+{
+  id: string,
+  driverId: string, // exact driver entry; profile resolution happens on read
+  classId: string,
+  sessionId: string,
+  timestamp: number,
+  type: "setup" | "tires" | "suspension" | "car" | "note",
+  note: string,
+  createdAt: string
+}
+```
+
 Normalized by:
 - `normalizeTrainingState()` wrapper in `src/app.js`
 - `normalizeTrainingState()` implementation in `src/training_logic.js`
@@ -1831,6 +1855,7 @@ Migration behavior:
 - older state without `training.ledger` normalizes to an empty ledger
 - the next `saveState()` imports valid registered-driver laps from all current session results
 - invalid laps, anonymous transponders, and laps at or below `500 ms` are excluded
+- older state without `training.markers` normalizes to an empty marker list
 
 ## 15.9 `Settings`
 
@@ -2334,6 +2359,9 @@ npm test
 - selected-profile/class active-practice isolation, current-start filtering, and non-practice rejection
 - two-minute stint boundaries, cross-session isolation, and boundary-lap exclusion
 - best-three consecutive calculation and five-minute final-lap/average-pace behavior
+- trimmed 30-minute pace buckets and weekday aggregation
+- class-isolated setup-marker before/after comparison
+- short-stint suppression, 10-lap race pace, and mistake detection
 
 `tests/public_training_profile.test.js` covers:
 - enabled/disabled/unknown share-token handling