01
Why a photograph is a poor instrument
The Melanin Layer. Light crosses the epidermis twice, and melanin sits on both legs of the journey.
See the graphic › Read piece 01 ›Graphic built
Programme 01 · Research
A reading sequence on how skin diagnostic models are built, who they are tested on, and what a sensor measures that a camera cannot. One piece every ten to fourteen days, for about three months. Written for someone with no background in any of it.
The Sequence
Each piece ends with the same question the programme is trying to answer, so the sequence has a spine rather than a theme. If a paragraph needs a degree to parse, it is the wrong paragraph.
01
The Melanin Layer. Light crosses the epidermis twice, and melanin sits on both legs of the journey.
See the graphic › Read piece 01 ›Graphic built
02
Performance across skin tone. An average stops describing anyone in particular once you split it.
Read piece 02 ›Graphic built, held
03
The scoping review itself. This is the piece that cites our own preprint.
Read piece 03 ›Waiting on the review
04
Federated learning. Data stays put and only gradients travel.
Read piece 04 ›Graphic to build
05
Uncertainty. A model reporting how sure it is rather than giving a bare answer.
Read piece 05 ›Graphic to build
06
No graphic needed
07
The Barrier. Water leaves through the lipid matrix, and how much is a state rather than a type.
Read piece 07 ›Held until the filing
Join
Cohorts are named and they start together. There is no queue, no position number and no allocation ledger, because there is no hardware to allocate yet. When there is, the rule gets published alongside it: priority follows the thinness of the training data, and the running totals are public.
Until then this is exactly what it says it is. Seven pieces, then it ends.
The sign up form is deliberately not built yet. It arrives with the first piece, not before it.