fix: clamp ALL rewards/scores to strict (0.01, 0.99) — every output path

fix: clamp all scores to strict (0,1) — never 0.0 or 1.0

fix: match EXACT sample inference pattern — HF_TOKEN or API_KEY, getenv with fallbacks

fix: test LLM proxy FIRST before env connection — ensures API call is made

fix: use exactly os.environ[API_BASE_URL] and os.environ[API_KEY] — no fallbacks

fix: bulletproof inference — never crash, always emit START/END, defensive parsing

fix: add error handling for from_docker_image + full traceback logging

fix: match reference inference pattern — HF_TOKEN, from_docker_image, no fallback URL

fix: use only platform API_BASE_URL and API_KEY, no fallbacks

add uv.lock for openenv validate

feat: 3 tasks with programmatic graders + OPENAI_API_KEY support

docs: rewrite README — adaptive curriculum front and center

feat: adaptive curriculum — environment learns from agent and gets harder

feat: real data wired in, visual map, all 5 improvements complete

feat: real data, Gymnasium wrapper, baseline comparison, research framing

feat: add live demo and baseline agent

feat: SupplyChainEnv — global supply chain disruption RL environment

initial commit