Section 1
The Capability Map
Six capability primitives. Each holds a distribution of human ability and a single threshold marking the best model in the world. The shaded region is the share of human ability the frontier has already passed. Every placement is a considered estimate — open a panel for the case behind the distribution's shape, the case behind the threshold, the cost curves on the same axis, and the specific evaluations all of it rests on.
Section 2
The Six Primitives
Every job is a weighted basket of these six. They are chosen to be separable — a task can load heavily on one and barely on another — so that a job's exposure can be read as a profile rather than a single score.
Section 3
Penetration Into the Economy
Modelled — not measured
Share of each capability's economic task volume performed by machine. These curves are an output of this framework, not an observation of the world. No dataset tracks substitution at the level of a capability primitive, so each line is a logistic curve derived from three stated inputs: where the frontier already sits, where machine cost undercuts human cost on the same axis, and a ceiling set by the work in that category that stays human for reasons no capability curve crosses. Read the ordering and the spacing between lines; do not read the individual numbers as data.
Why the ceilings differ. Knowledge retrieval is capped only by the verification burden that miscalibration leaves behind. Relational and Creativity are capped much lower by demand for provenance and for a human counterparty — a preference that survives the capability crossing. Abstract reasoning & judgment carries the accountability overlay: some decisions stay human because someone must be liable for them.
What would falsify this. Each curve makes a checkable claim about ordering, not magnitude: knowledge work substitutes first, functional reasoning follows within a few years, physical work trails everything by a decade or more. If physical work moved before functional reasoning, the map's central argument would be wrong.
Section 4
How to Read Across
A task substitutes only when every primitive it loads on has crossed — timing is set by the weakest axis, not the average. Find a job's bottleneck by reading across all six panels and looking for the lowest threshold, then check whether the capability crossing has been followed by a cost crossing, which is the event that changes behaviour. Accountability sits over the whole map as an overlay: some work stays human because someone must be liable, and no capability curve crosses that.
Companion study: The Intelligence Explosion