Strengthen grading transparency and reproducibility evidence

Revamp README and WEB_README for judge rubric alignment

Align success semantics with resolution and harden SG action handling

Make inference task-aware and deterministic under action-cost rewards

Add model fallback selection and token diagnostics for inference

Add action-cost RL dynamics, metadata lookup, and hard-mode cascading failures

Avoid rounded endpoint scores in inference END logs

Fix full validator cross-platform inference log handling

Enforce strict (0,1) task scores for Phase 2 validation

Upgrade README and WEB_README for hackathon-quality documentation

Add Gemma 3 vs Gemma 4 baseline comparison

Add uv.lock for OpenEnv submission readiness

Add required OpenEnv discovery metadata tag

Force /app/README.md to web docs for OpenEnv /web renderer

Fix /web README source and speed up Docker rebuilds

Use dedicated WEB_README for /web documentation rendering

Include README.md in Docker image for /web documentation

Set HF Space app_port to 8000 for Docker runtime

Prepare HF Space submission validation and compliance.

initial commit