result update

Experiment result update

Tasks from openclaw

docs: add comprehensive CyberGym+MEMRL experiment report

Revert 'add stratified sampling script and --subset-file support'

add stratified sampling script and --subset-file support

add list of 32 tasks that successfully found PoCs in previous Round 1

apply memrl_engine_patch_v4: unified Q_init=0.0, remove ε-greedy

docs: add MemRL engine patch guide for server-side apply

align MemRL with paper: symmetric Q-attribution, unified Q_init=0, λ=0.5, α=0.3, δ=0.3, greedy selection; differentiate success/failure memory storage and retrieval display

docs: add memory_service.py patch for task_id persistence

docs: add MemRL architecture documentation

memrl update

add hot restart investigation doc for server-side Synergy

eliminate MEMRL replay: rebuild caches directly from checkpoint textual_memory.json