About

Reckon Research builds the tools that make inference placement legible. We measure hardware, model the economics, and publish the results. The name is a navigator's term: to reckon is to estimate your position from what you know, without waiting for a signal you may never receive.

The world spends its largest capital line on silicon, and workloads land on it by default, habit, and vendor lock. Compute is traded as a commodity, dollars per GPU-hour, but consumed as a service: useful work under latency, cost, and energy constraints. The same workload can cost multiples more in one placement than another while meeting the same service target. Frontier labs train models. Providers rent chips. No one is building the system that decides where computation should run.

That is what we are building.


How we work

Measure first

Every estimate is grounded in hardware traces. We publish the raw data alongside the model so the numbers can be checked, disputed, and improved by anyone with access to the same silicon.

Pre-register gates

We state the falsification condition before we run the experiment. Publishing only the confirming cases is not a result. We pre-register gates before we run experiments.

Open by default

The core tools are Apache-2.0. The corpus of hardware traces is public. The methodology is documented in enough detail that a reader can reproduce the measurement without asking us.

Physics, not black boxes

Every prediction is a closed-form expression you can audit by hand. No learned model stands between you and the number. The term that produced each estimate is labelled in the output.


Get in touch

If you have hardware traces, a workload that behaves unexpectedly, or a dispute with a number on this site, we want to hear from you. Disputes that arrive with data attached outrank everything else.

Contact Reckon Research