Scientific safeguards

The following rules are part of the product contract. They are not optional presentation choices.

Measurements stay separate from estimates

Observed data and inferred surfaces use separate tables, layers, provenance, and visual marks. A caller cannot request a view that makes their status indistinguishable.

Unknown is an output

Map cells marked as unknown—or as driven mainly by model assumptions rather than observations—are excluded from summary statistics, and the excluded fraction travels with the result. An unmeasured place is not treated as zero. A visually empty area is explicitly labeled rather than silently filled.

Zoom follows evidence

Map resolution responds to support, not interface zoom. Zooming past the evidence reveals gaps; it does not manufacture local precision.

Assumptions remain explicit

Coordinates, location uncertainty, recruitment, cohort identity, allele orientation, penetrance, and denominators never receive convenient scientific defaults.

Clinical scope is enforced server-side

The burden system excludes behavioral, cognitive, and anthropometric traits for every caller. Prompt wording, frontend state, and configuration flags cannot bypass that policy.

Refusal beats plausibility

When the evidence cannot support a number, genomeOS returns no number and says why. Published results for haemoglobin S (HbS), glucose-6-phosphate dehydrogenase (G6PD) deficiency, and screening programmes are the validation controls; a plausible-looking map is not.

Read the complete repository invariants.