Fix final phase 1 and phase 2 inference and scoring issues

readme

chore: push verified deployment fixes

Add Expert tier tasks to evaluate multi-step UUID stateful tracking and OpenAPI discovery

Implement strict judge feedback: ambiguous task descriptions, server-side payload clues, increased exploration steps

Initial submission: API Debug Environment