Fix format_comparison_metrics_html to accept run_comparison() dict directly

Add run_demo_episode wrapper to demo.py for dict-based episode results

Update metrics format with drift/oversight tracking, add colab training notebook

Add billing schema drift support with field renaming

Add drift-specific metrics: drift events, detection, adaptation rate

Add oversight accuracy and explanation quality metrics to dashboard

Improve HeuristicOversight explanations with specific data references

Add SLA policy drift support for ticketing system

Add structured explanation quality scoring for oversight agent

Fix schema drift renames to target actual Customer model fields

Fix window_ticks policy enforcement in billing refund validation

Add randomized attacker, security metrics engine, and updated Gradio dashboard

Add episode metrics computation and HTML formatting for SentinelOps Arena

Replace HeuristicAttacker with RandomizedAttacker for probabilistic attacks

Remove hackathon_env template, rewrite train.py for SentinelOpsArena

Implement Phase 3 (HTTP server) and Phase 4 (demo + Gradio app)

Implement Phase 2: environment core with MCPEnvironment base

Implement Phase 1: models, enterprise systems, attacks, rewards