Open neurosurgery,
end-to-end. For every brain.

A disease rare enough that no single hospital sees enough of it to learn from.

The hospitals stop working alone.

Federated cohorts let nine centres answer a question none of them could answer separately — without a single record leaving any of them.

The evidence exists, spread across a dozen registries that cannot speak to each other.

One study definition, run everywhere.

Written once in a shared vocabulary, executed at every site, results returned as aggregates. The definition travels; the patients' records never do.

A model works beautifully at the hospital that built it, and nowhere else.

It does not ship until it works somewhere else.

Every headline result is re-run at a site that had no hand in producing it, and the outcome is published either way — including the one that failed.

Outcomes still track postcodes, and the paper mentions it in the limitations.

Equity fails the release, not the discussion section.

Subgroup performance is a gate in the release path. A model that fails it does not ship, however good the aggregate number looks.

Patients were consulted, thanked, and shown the door before the decisions.

Two seats on the council. Real votes. Permanent.

Observer status was proposed and rejected during review. Advocates vote on the same motions as clinical and technical members.

The answer a family needed was already written — behind a paywall, in a language they don't read.

Open at the source, and in plain words.

MIT licence at public alpha, no embargo. And 204 guides plus 1,060 defined terms written for the person holding the scan, not the person who ordered it.

Four films, when we can make them honestly.

Nothing is recorded yet. Each film needs ethics cover, a site willing to be named or a fully synthetic dataset, and clinician review of every frame that shows an output — so these are commitments, not teasers.