Blog: All links section at end, dedupe inline URLs

Blog: fix typo, clarify reward CSV paths and D1

Blog: clearer problem synopsis, Colab links, resolve URL for chart

docs(Blog): CSV stats, 0.6 by task, tools list, reward_curve copy.png figure

docs(Blog): Colab + Space tree links, Qwen3-0.6B run, human read on ~160/~0.6 curve

docs: add Blog.MD writeup for HF Space; README links Hub + YouTube placeholder

docs: sample GRPO reward log + curve, human README section, plot colors

docs(readme): Space-first table for live app, /play, healthz, hf.space vs Hub

feat(ui): DriftShield /play HumanAgent + State observer console

feat(space): enable OpenEnv /web play UI, root redirect, docs and sample rewards CSV

docs(colab): save zip, judging CSV, clarify grader vs LLM judge

fix(packaging): bump 0.1.1 so pip upgrades; verify colab_grpo after install

docs(colab): prerequisite + clear error if support_ops_env missing

feat(colab): A100 30min preset, vLLM logprobs TRL patch, full reward heads (sid-rp-style)

Add Colab reward analysis graphs

Improve GRPO prompts for tool-call rewards

Expand Colab preset tasks to target episodes

Simplify Colab notebook for HF Space training

Add Colab GRPO memory presets

Tune Colab GRPO defaults for 22GB GPUs

Match TRL reward function keyword API

Handle OpenEnv validation errors in GRPO

Harden GRPO rollout action parsing

Remove incompatible torchao from Colab setup

Pin Transformers below 4.57 for Colab training

Re-pin Colab stack after OpenEnv install

Pin Colab training dependency versions

Pin Colab install to latest helper commit

Fix DriftShield Colab GRPO reward logging

Expand Colab notebook for multi-episode reward curves

Disable vLLM sleep mode by default in Colab

Disable 4-bit loading for vLLM GRPO

Pin Colab setuptools for vLLM

Fix invalid low-VRAM GRPO generation count

Clean up Colab dependency conflicts

Tune Colab GRPO notebook for low-VRAM GPUs

fix(colab): drop expandable_segments for vLLM cumem; fix notebook JSON

fix(colab): reduce GRPO train VRAM (4bit NF4, vLLM sleep, checkpointing, smaller G)

fix(colab): cap vllm_max_model_length for colocate KV cache (Qwen3 context)

fix(colab): wrap stdout/stderr for Jupyter fileno with GRPOTrainer

docs(colab): full GRPO training cell scaffold

fix(client): reject hub page URL for WebSocket base_url

fix(server): add GET / and favicon 204 for HF Space probes

README: clarify HF Space hub URL vs hf.space ENV_URL and paste typos

Colab: document correct pip URLs for Space vs OpenEnv; single install line

Dockerfile: install git for pip VCS install of openenv-core

Space: healthcheck /healthz, document git push to HF

DriftShield OpenEnv: grader, server, Docker Space, notebooks