Benchmark 001 is published, losses included
Our first training run cut forecast error 17.8% against its base model. It is not yet ahead of a simple last value baseline, and we publish that too.
On 14 October 2026 we published Benchmark 001 at benchmarks.eigna.ai. Our first training run cut forecast error 17.8% against its base model. It is not yet ahead of a simple last value baseline, and we publish that too.
The model forecasts hourly BTC and ETH 24 hours ahead. It was built for our first client, which shares founders with Eigna. On 243 held-out days its forecast error is 1.439% against 1.750% for its open base, Kronos-small. Three simple baselines still score lower: last value 1.392%, EWMA 1.398% and GARCH(1,1) 1.407%.
Its forecast bands held 57.3% of outcomes against a target of 80%, and on paper it traded worse than its base: +4.4% against +15.0%. No candidate was released. It is an internal diagnostic on two assets, not a product claim.
Read the full report, Case 001, or why we publish the losses.
