v0.2.0 — Cost Dashboard, Alerts, Auth, CI, Visibility

What's new in v0.2.0

New Features

• Cost & Latency Dashboard (/cost) — token usage, avg/P95 latency, estimated cost by model, 14-day daily chart
• Slack/Email Alerts — config-driven dispatch on non-negotiable failures or score drops. Set SLACK_WEBHOOK_URL in .env to enable
• HTTP Basic Auth — optional middleware. Set GLASSBOX_USERNAME + GLASSBOX_PASSWORD to enable. Blank = open dev mode
• GitHub Actions CI — pytest + ruff on every push/PR to main
• Judge Reasoning Viewer — reasoning strings shown in TestSuite drill-down per property
• Corpus Coverage Panel — conforming/non-conforming counts, ticket types, non-negotiables tested
• Retry Indicator — amber badge in Monitor and TestSuite when response was retried

Improvements

• Concurrent corpus execution — test suite time cut from ~15 min to ~1–2 min
• temperature=0 on all LLM calls — evaluation is now fully deterministic
• Escalation compliance fixed — explicit runtime warning when previous_contacts >= 1
• Non-negotiable corpus threshold corrected to 90%
• Alert log rows enriched with full ticket details

Schema changes (auto-migrated on startup)

• runs: input_tokens, output_tokens, retried
• snapshot_examples: property_reasoning_json, non_negotiable_reasoning_json, retried

See CHANGELOG.md for full details.