Data profile

Generated by docs/generate_plots.py from the actual built cohort — make docs-assets regenerates these charts and CI fails if the committed files drift from what it produces (PORTFOLIO_PLAN_V3.md §9.4).

The cohort is Synthea (Apache-2.0, MITRE Corporation) synthetic patients, Massachusetts, seed and population size committed in data/synthea.config.json. Every date has been shifted per-patient before this cohort was built — see data/deidentify.py and D-A12-2 in decisions.

Race

Age

Ages above 89 are capped at 90 per HIPAA Safe Harbor before this chart is drawn — see data/deidentify.py.

Illness burden proxy

Active condition count is the illness-burden signal pipeline/labels/burden_label.py builds on (Phase 4) — deliberately never touched by the access-gap injection described below.

Disclosure: the injected access gap

Synthea generates realised cost from utilisation with no differential access by race, so this cohort's utilisation and spend fields have been deliberately adjusted for one stratum, at equal illness burden, per the seeded and committed data/access_gap.config.json. This is what makes the label-choice experiment (Phase 4) demonstrate the Obermeyer et al. (Science, 2019) failure mode rather than merely claim to. See D-A12-2.