We back fields we think are ready to move, then check whether they do. Field velocity is that rate of change: how fast talent enters, capital forms, tool costs fall, and output ships. Inflection points are one of the markers we read — dated, falsifiable shifts an accelerating field should produce.
We measure velocity the same way we do the work: as a research program. Whether field acceleration works is itself the open question, tested across all four focus areas.
Learn more about our methodologyPick a focus area. The summary above reads that field’s velocity across the instruments that apply to it; the inflection points below are the specific markers we track, each with its live signal.
The method is the meta-research design of how we do field acceleration. We name the interventions we run, then read field velocity as the result. Same design, every focus area.
A fixed toolkit we bring to every field. Pick the ones a field is missing, then push.
We don’t claim these interventions directly cause a field to accelerate: attribution at the field level isn’t cleanly identifiable. For now we name what we run and watch whether the field moves; making that link clearer is work we intend to do, and to publish here later.
The interventions are the input. Field velocity, the rate a field is moving, is what tells us whether they landed. We read it through five instruments.
Not every instrument fits every field. A field with no single gating unit cost has no performance curve to read, and a field no forecast market has priced has no market signal. We show the instruments that apply and name the ones that do not.
A reading is only reported as current for about a year after the observation it refers to. Past that we keep the number but flag it stale and drop the trend arrow: a metric we have not re-measured is unmeasured, not flat. We also never draw a direction from fewer than three data points, and we separate when a figure was last measured from when the pipeline last ran.
These instruments read the research and capital sides of a field. They do not observe invention directly: prototypes, designs, datasets, and negative results largely lack identifiers to count. Closing that gap is itself part of the work.