ADR-006 — No composite score in Phase 1
Status: accepted (2026-09-08)
Context
A single "impact" or "priority" score per cancer is attractive for headlines and for the homepage, and the specification foresees composite indexes (CI-IMPACT etc.) in a later phase (§353). Composite scores, however, hide weighting choices, mix units and uncertainty levels, and are the values most likely to be misread as clinical or policy verdicts. In Phase 1 the platform still lacks the burden layer for most of the world (GLOBOCAN license review, SEER credentials), so any composite would be dominated by whichever inputs happen to exist.
Decision
Phase 1 ships single-input, fully explained metrics only — the catalog in
packages/database/src/seed-data/metrics.ts (burden counts and rates, mortality-to-incidence ratio,
5-year survival, trial counts, publication counts and growth, curated evidence and gene counts,
genomic cohorts) plus two two-input gap indexes (trial_gap, research_gap) whose formula is a
transparent percentile difference within one scope. No weighted sum across metrics is computed,
stored or displayed. The rankings.breakdown column and the composite category in
metric_definitions exist in the schema but remain unused.
Consequences
- Every ranking on the site answers a precise question ("which cancers have the most recruiting trials?"), with scope, year, source and formula version visible and "Why this rank?" available.
- The homepage cannot present a single master ordering; it presents several rankings side by side with their scopes, and the methodology page states explicitly that composites are not computed yet.
- When a composite is introduced (Phase 2+), it must: publish weights and components in
rankings.breakdown; be versioned like any formula; be computed only where all components exist in the same scope (no imputation); be markedexperimentaluntil reviewed; and never be labelled as a measure of importance of a disease or a judgement on research communities (spec §324).