Fix progression stalls and add normal class trainer visits

deploy/playable/README.md

@@ -232,3 +232,33 @@ By default:
 The project checkout keeps AzerothAI's local D1 state in `.wrangler/state`.
 
 Do not delete these directories if you want the realm and character history to survive restarts.
+
+## Class training and leveling
+
+`ale/azerothai_training.lua` checks each managed resident after a level change and
+rechecks every minute. It visits a nearby class trainer when the core reports
+eligible spells, then uses the normal trainer purchase transaction. Purchases
+require interaction distance and sufficient character gold; spells and money are
+never granted. Combat, recovery, player control, and mixed parties suspend visits.
+Stalled visits release paused work and retry later. The monitor reports pending
+spells, trainer, blocked reason, and verified purchases separately.
+
+This requires the native ALE methods installed by:
+
+```bash
+python3 deploy/playable/patch-native-training.py /path/to/azerothcore
+```
+
+Rebuild and deploy worldserver before enabling the training Lua module. The patch
+also restores the native RPG training trigger and prevents sightseeing flights
+for residents below level 60. Normal travel flight routes remain available.
+`install-lifecycle.py` copies the Lua module with the other resident services.
+Without the native API, the monitor reports that training is unavailable.
+
+Training state and recoverable strategy leases live in
+`acore_auth.azerothai_training`, one row per resident. Existing trainer purchases
+made outside this workflow are not retroactively included in its purchase count.
+
+Flight guards and telemetry use `HasUnitState(0x00000100)` (`UNIT_STATE_IN_FLIGHT`).
+ALE's `IsTaxi()` tests the flight-master NPC flag and must not be used to detect
+a player riding a taxi. Model decisions also yield to an active native flight.