The field's core instrument — not a paper on comorbidity but a queryable research substrate: 15 conditions, 34 mechanisms, and 278 cited edges. Read it as a single document, explore the graph, or download the whole thing. It handles far more than overlap: shared mechanisms, masking, amplification, and the partial-definitional cases where two labels describe one underlying process.

The whole discipline read the way it is actually built — downward. Five strata from the human surface to the mathematics, 15 domains, 83 fields, 362 topics, all openable in place and any number at once. Artificial intelligence isn't a domain here: it was too large to mean anything, so it is three, and Language Models gets the nine fields the last three years earned it. Where two domains genuinely share an edge, the map names the seam instead of picking a side.

Benches, not chapters. Each one puts a working specimen on a stage, hangs dials off it, and prints the resolved values — spacing scales, type ladders, easing curves, Mac window anatomy, touch targets, Dynamic Type — as a recipe that pastes straight into a prompt. It opens on the Sight Gap: the benches that turn something you can see into something a model can act on.

2,600 years on one chronology you fly rather than scroll. Every figure is a star whose brightness is how much of what they actually said survives, and by whose hand; the constellation lines are who read whom. Crossed by the thresholds — the press, Wundt's lab, chlorpromazine, fluoxetine, the web — each with a beam through the whole field, because everyone to the left of it lived without it. Set a birth year and the field answers what arrived too late. Every entry carries traced sources, each labelled with the epistemic claim it can actually support.