Medicine knows less about women than about men.
This map shows researchers in women’s health which data is missing, who said so, and whether anyone has collected it since — then drafts the study that would collect it.
Select an area to see its problems.Select an area from the list below. The numbers on the figure match it.
Affected women:
About this platform
How the system works
A gap in women’s data closes when three groups meet: a researcher who will design the study, a fund that will pay for it, and the women it is about. The platform is what puts a named gap in front of all three. It is one stage of that loop, not the whole of it — the loop is drawn below, and each stage says whether it is built or planned.
-
Gaps are found, one at a time
Published reviews naming data that is missing about women are found, checked against the live web to see whether it has been collected since, and admitted only after a person confirms the claim against its source.
-
The platform shows what is missing
Which data is needed, what status it has, who said it was missing and when, and when the platform last checked.
-
A researcher designs the study
The platform hands over a draft design — which women, which variables, which breakdowns — labelled AI-generated and unreviewed, and withheld where the rules refuse one. Reviewing it and running the study is the researcher’s work.
-
A fund chooses a study to back
A funder sees the gap, the published claim it rests on and the proposed study in one place, and decides what to pay for. Without this stage the loop has a study nobody funds.
-
Matching women join the study
Women whose own gap the study would close take part in it.
-
The collected data closes the gap
The record’s status changes, and the loop returns to stage 2.
How a record enters the system
Stage 1 in full. The last step is a person: nothing reaches the register until somebody has read the claim against its source and admitted it.
Discover
Run by plain code
Europe PMC, searched for the sentences a review writes when it looked and found nothing.
Filter
Run by AI
One yes/no question per abstract: does this name missing data about women that the paper does not itself go on to collect?
Extract
Run by AI
Structured output against the record schema. What the source does not support comes back null, not a guess.
Verify
Run by AI
A live search for data collected since the gap was named. The check is dated, and no search proves a negative.
Validate
Run by plain code
Ordinary JavaScript, no model. A record with no source, or no check, or one claiming data exists without saying where, is rejected.
Review
Run by a person
The run stops. Nothing reaches the register until a person has decided on it, and every decision and its reason is logged.
Design
Run by AI
Drafted last, and only for a record already admitted: the spend follows the approval. Kept out of the register, and withheld entirely where the safety rules refuse one.
Most tools point a model at a problem and let it generate the answer. This one inverts that. Where the data gap is, is a claim, and every claim here is borrowed from a published source and attributed. AI is used for finding, extracting and checking — never for making the claim. Nothing the model composes is stored as a record: a study design is drafted only after a person has admitted the record it answers, labelled AI-generated, and kept out of the register entirely.
Disclosures
- When it was built. Everything in this repository was written during the Hackathon Period, 14–20 September 2026. No pre-existing code was used, and the git history is the record.
- Where AI is used. Development was AI-assisted throughout, using Claude Code. The intake run calls Claude Haiku 4.5 to screen abstracts and Claude Sonnet 5 to extract a record, run its verification search and draft a study design. The published site itself calls no model.
- Where the records come from. Every record was found through Europe PMC, which indexes the published literature this platform searches. Each record quotes one sentence from the paper it cites, to report what that paper claimed, and links to the source; those sentences stay the copyright of their publishers and authors.
- Third-party assets. The female silhouette is OpenClipart #71126, public domain under CC0, recoloured and otherwise unaltered. The GitHub mark in the footer is GitHub’s trademark, used unaltered and in a single colour to link to GitHub.
-
Licence. The code is MIT; the register is CC BY 4.0,
except the quoted sentences, which are not the author's to license.
See
LICENSE. - Repository: github.com/minimarina/her-silent-data