Notebooks — the reproducibility backbone

Every figure and number in an advocacy brief is produced by a notebook here, from the raw inputs. make reproduce runs them end to end; a claim no notebook can regenerate is not published.

The notebooks are deterministic (seeded) and cover:

• Hotspots — kernel density surfaces (labeled as report intensity unless exposure-normalized) and Getis-Ord Gi* significant clusters.
• Trends — change over time, with intervals.
• Exposure sensitivity — how conclusions shift under different exposure denominators, stated openly rather than hidden.

Each notebook records its inputs, the exposure source and date, and the thresholds used, so a figure traces back through statistic → cleaned dataset → raw reports. See docs/METHODOLOGY.md.

Teaching module — "How to lie with heat maps"

teaching/ is a hands-on workshop for journalism and civic-data audiences: three bilingual (EN/ES), deterministic notebooks built entirely on the synthetic Davis decoy fixtures, plus a 90-minute facilitator guide. It shows how a raw-count heat map misleads (the busy decoy seg-03), how exposure normalization corrects it, and how re-segmentation can manufacture or dissolve a "significant" hotspot. Run them with make teach; a CI job executes them on every push.