Analytic Workbench

Human-directed, AI-operated analysis. The AI computes; the human steers through a review surface (marimo notebook or Quarto doc). This is not a batch pipeline — intermediate findings are presented for redirection at every step.

Start Here

1. Read AGENTS.md or agents.md if present
2. If neither exists, create AGENTS.md and update CLAUDE.md (see below)
3. Inspect the repo before changing structure
4. Separate repo facts from agent assumptions
5. Separate frozen inputs from live pulls

Then establish per area: current mode, likely next mode, review surface, first boundary-sensitive change.

Fresh Repo Bootstrap