Update README.md

Upload blog.md

Delete blog.md

Update static frontend: Experience tab, mobile nav, responsive layout, live run UI

Update blog.md

Serve SPA with no-cache headers so HF UI picks up static/index.html after deploy

Blog/Space: second-person to we/us in analogy and correlation sections

Blog: reconcile correlation narrative; add RNG/game analogy; push protham.ipynb

Refresh training metrics and dashboard image for HF Space

Embed real 77-step GRPO data into training charts

Bump version to trigger Space rebuild

Update training dashboard charts and stats

Fix Bug 4 and Bug 5 from FIXES.md: _handle_history step_log + asyncio lock for start_episode

replace unverified % estimates with actual measured numbers from notebook

fix: use actual measured numbers from protham.ipynb, remove speculative line from blog

add SPA frontend: Blog, Training, README, API tabs with Chart.js graphs

Update blog.md

Update blog.md

chore: add LFS tracking for PNG files, migrate images to LFS

add training section, result images, and protham notebook to blog

feat: 3-phase belief training — probe evidence → model generates belief from observation

fix: model generates belief — parse 2 JSONs from output, raise belief weight to 0.4

fix: simplify install to 2 cells — force-reinstall torch (NCCL fix) then unsloth

fix: adopt uv + torch-presence check for unsloth install (mirrors official OpenEnv notebook)

fix: use pip install unsloth instead of git-based install; split install/verify into two cells

fix: run env.step in thread pool inside /run_episode to prevent event loop deadlock

fix: add /run_episode endpoint + update reward_fn to use it

fix: bypass localhost HTTP for drift/probe — use direct in-process DYNAMIC_CONFIG mutation

feat: evidence-augmented prompts for true world modeling

fix: drift_trigger_step + reward_fn for proper belief scores

chore: add uv.lock for openenv deployment

Fix environment bug: adjust drift triggers and prevent premature auto-termination to avoid reward hacking; update README

Remove deprecated __init__ arguments from EnvClient

Update default HF router model to Mistral 7B to ensure standard endpoint availability

Fix inference loop for openenv-core 0.2.3 StepResult compatibility

Add HF Spaces YAML metadata to README

Add .gitignore and remove cached build artifacts

Initial commit: AdaptiveWorld Env v2.0 (Round 2)