approach · how the lab works
afterwave studies one question: when a piece of content meets a population, what happens — to whom, for how long, and with what second-order effect. we answer it by simulation, and we hold every answer to the standard of a measurement: reproducible, calibrated, and reported with its error.
thesis
the public record of what content does is almost entirely first-hour engagement — likes, reposts, dwell time. it is abundant, cheap to collect, and a poor proxy for the thing that matters: the downstream effect a sentence has on a reader days later, and on a community you did not know you were addressing.
we treat content as a perturbation to a population, and forecasting its consequence as a scientific problem. that framing carries obligations. a forecast must be calibrated against outcomes we did not train on. it must come with uncertainty. and it must be honest about the regime in which it stops working. an estimate that cannot fail is not a measurement — it is a marketing claim.
principles
a synthetic population reads the draft. not keywords — readers, each with a prior, an attention budget, and a place in the network. response is aggregated from individuals, so the tails survive.
every forecast is backtested against held-out outcomes and ships with error bars and a coverage score. outside the validated regime, we widen the interval and say so.
we measure the downstream effect — what a post changes weeks later — not the engagement it spikes in the first hour. consequence over engagement, in the method, not just the slogan.
method · draft to forecast
we sample a synthetic population to match a target audience — its priors, its attention, its network structure — against measured behavior on held-out content.
each agent reads the draft to the extent its attention budget allows, forms a response, and — conditioned on graph position — propagates or terminates it.
we run the simulation forward to estimate projected reach, response half-life, breakout probability, and any latent backlash risk — each with a population-derived interval.
the forecast is returned with error bars, a coverage score, and an explicit flag when the draft falls outside the calibrated regime. no point estimate without its uncertainty.
afterwave is an independent research lab, pre-launch and hiring researchers and engineers who care about information propagation, agent simulation, and calibration. tell us where to reach you.
✓ received. we read every note and will reply.
research · engineering · collaboration