documents/CHANGELOG.md

Changelog

All changes to the Invoice Exception Handler environment are recorded here.

---

[1.0.0] — 2025-01-20

Added

• Initial implementation of InvoiceExceptionEnv with full OpenEnv API
• Three tasks: task1_price_variance, task2_duplicate_tax, task3_compound_fraud
• Pydantic v2 typed models for all environment objects
• FastAPI HTTP endpoints for HF Spaces validation
• Gradio UI for interactive exploration
• inference.py using OpenAI client with [START][STEP][END] log format
• openenv.yaml spec file
• Dockerfile for HF Spaces deployment

Design decisions

• Used pure Python simulation (no external databases) for portability and determinism
• Compound fraud task has four signals to prevent simple greedy agents from scoring well
• Channel selection in Task 3 (phone vs email) tests policy knowledge, not just anomaly detection
• Grader uses sub-scores to allow partial credit for partial solutions