Solanabean/meta-ai

fix(phase2): add id and grader blocks to openenv.yaml tasks for validator discovery

Phase 2: align with judge API_KEY, Qwen 72B default, score formatting and resilient inference

fix(phase2): match organizer [START]/[STEP]/[END] key=value format per task

fix(phase2): terminal grade on timeout, sanitize END scores, task grader metadata

fix(phase2): map grader scores to open interval (0, 1)

fix(phase2): emit [START]/[STEP]/[END] on stdout for validator parsing

feat: Qwen default for HF Inference, dotenv load, LiteLLM remap fix

fix(phase2): LiteLLM-compatible default model when API is not HF router

fix(phase2): ensure MODEL_NAME is never missing from os.environ

feat: shared START/STEP/END stdout (checklist); stderr for logs; pyproject inference_log

feat(inference): START/STEP/END stdout; os.getenv layout; LOCAL_IMAGE_NAME

fix(inference): default MODEL_NAME for Phase 2; top-level exception handler

Improve reliability and bounds: scenario camouflage, grader bonus parity, safer env

Trim unused dev extras; drop GET /reset /step helpers (documented in README)

Explain GET vs POST for /reset /step; document stateless HTTP /state

Redirect GET / to /docs so HF Space root URL is not Not Found

Remove redundant e2e script; exclude uv.lock from Docker context; refresh uv.lock

Add e2e_production_check.py for HTTP + WebSocket smoke tests

Align inference.py with hackathon env vars; document API_KEY and default API base

Fix HF Space card: colorFrom must be allowed value (gray)

Update README

Forensic Financial Investigation OpenEnv environment