Release notes: 2026-05-27 — studio-sim-rl-env-pool-step¶
Status: Ready for review
Repo: li-langverse/lic
Branch: feat/studio-real-implementation
PH / REQ: PH-SIM SIM-3, Li Studio real implementation
Summary (one sentence)¶
studio_sim_step_hook for the sim_rl profile now runs env_pool_stub_step on the caller's SimSessionStub instead of a detached sim_rl_tick_stub self-test plus a single sim_step.
Changed (specific)¶
| Area | What | Evidence |
|---|---|---|
li-sim | sim_rl_session_env_pool_step — shared env-pool batch + contract checks | env_pool_stub.li, sim_rl_tick_stub |
li-studio | studio_sim_rl_step_hook; RL branch early-returns after pool step | studio_sim_rl_step_hook.li, studio_sim_step_by_profile.li |
Not changed (scope fence)¶
- PH-ML async workers, persistent
EnvPoolinli-ml-rl— not in this change. - Game physics session coupling — not in this change (
studio_game_step_hookstill uses ephemeralPhysicsWorld). - Oracle / tier-2 bench claims — none.
Breaking changes¶
None — RL profile studio_sim_step_hook now advances tick by sim_rl_env_pool_size_default() (4) per frame instead of 1; callers depending on tick==1 after one hook call should use pool size (smokes updated).
Downstream¶
| Repo | Action |
|---|---|
| Native timeline | Expect tick += pool_size per RL frame until persistent pool lands |
li-ml-rl | Re-export sim_rl_session_env_pool_step when package scaffold merges |