About expressive

Most genetic-health products tell you what to do. We don't.

We surface what current research actually says about each variant we find in your DNA , effect sizes, study quality, replication status, population data , and let you decide what to do with it.

When new studies publish on a variant you carry, your report updates. It's a living index, not a one-shot PDF.

The product thesis is that every claim is anchored in real research, with proper hedging, and we don't guess. That sounds like marketing copy until you realise it forces every page through some uncomfortable filters:

• Every empirical claim ships with the PMID it came from, inline. No silent rephrasing of methodology you can't click through to.
• When the evidence is thin (single small study, in-vitro only, no replication), the page says so out loud. We'd rather be quiet than confident-wrong.
• If the source article doesn't actually discuss the variant you're reading about, the page refuses to write that section. Methodology paraphrases don't count as content.
• Gene symbols are taken verbatim from authoritative sources. No silent paralog substitutions , the difference between OR51E1 and OR51E2 matters even when it sounds small.

Each of those rules killed an entire category of bad output we'd otherwise have shipped. They live in the prompts, the data pipelines, and the renderer , multiple safety nets because brand-defining claims deserve them.

A raw genotype file from 23andMe, Ancestry, MyHeritage, a direct VCF from a sequencing service, or anything compatible. Encryption happens client-side with a key derived from your wallet signature; the file gets parsed into individual variants and then discarded. We keep the variants, not the file.

From there, your report cross-references each variant against the indexed research corpus , GWAS Catalog, ClinVar, PharmGKB, GTEx, and the underlying PMC-indexed papers , and surfaces what's actionable, what's interesting, and what's being studied but isn't settled yet.

We don't sell your data. We don't share it with insurers, advertisers, or research consortia without your explicit, specific consent (and even then, only de-identified, per study, never blanket). We don't use it to train models we don't operate ourselves. None of that is unusual to promise in this space , what's unusual is that the architecture actively enforces it. PII lives in a different database from genetic data, both encrypted at rest, and the wallet- derived key never leaves your browser unencrypted.