A third candidate is held at the gate
On 23 October a new version of our first client’s forecasting model met the release gate. It tied with the field, and a tie keeps the current version.
On 23 October 2026 a third candidate of our first client’s forecasting model met the release gate. It is a newer version with a shrinkage head, built after Benchmark 001 showed that our first runs were too sure of themselves.
Out of sample, in replay 018, 308 forecast windows from 3 to 22 October, it ranked 3rd of 9 forecasters, just behind Toto 2.0 and GARCH(1,1). None of the gaps was significant.
A tie keeps the current version.
Our rule says a new version goes live only when it is strictly better, and no worse on calibration. A tie is not better, so the candidate was held, like the two before it. Release 1 is still not issued, and nothing untested reached our client’s desk.
Our first client shares founders with Eigna, and we say so up front. The full record is in Case 001, and every number is at benchmarks.eigna.ai.
