Coral Compass

Overview

Joins your commits, calendar and chat onto one hourly timeline, scores the day 0–100, and asks a model to name the fragmentation pattern in the data.

The problem

Most focus tools watch a single signal, which is why their advice never lands. The cost of a fragmented day isn't the meeting itself — it's the refocus tax on either side of it, and you can only see that by joining calendar against work that actually happened.

Approach

Three Coral data sources feed one SQL join that buckets the day by hour into commits, meetings and pings. The scoring is plain arithmetic over those rows — context switches, refocus tax, longest uninterrupted block — and only the final naming of the pattern goes to a model.

What it does

• Joins three signals — GitHub commits as deep work, calendar as scheduled fragmentation, chat pings as unscheduled
• A single SQL join stitches them into hourly rows, then scores context switches and refocus tax at roughly 15 minutes per switch
• Provider-agnostic insight layer — Gemini or Anthropic behind one env toggle

What I learned

Shipped as a hackathon build and labelled as one. The mock-data path is solid and the live join works end-to-end on GitHub and chat; calendar is wired but rough around timezone edges. Being explicit about that in the README was more useful than polishing it into looking finished.