Reckon Research publishes on Substack. Essays cover inference economics, compute placement, measurement methodology, and the systems that allocate computation. Technical notes accompany each major release of berth with the reasoning behind design decisions and validation results.
All posts are free to read.
Why the same workload costs multiples more on one chip than another, and why the premium is not a constant. How to measure it, model it, and act on it.
The decision that precedes every inference dollar: where does this workload run? The tools and primitives that make that decision legible.
How to run a benchmark that means something. Provenance, held-out splits, pre-registered gates, and what a failed validation tells you that a passing one cannot.
Technical notes accompanying each major release of berth, with the reasoning behind design decisions and the raw validation data.