Cattle, Genetics, and Code

I raise registered Black Angus at Deluisio Angus, and I build the software that helps me breed them. This is the technical side of that work: how the genetic numbers get computed, the boundary that keeps a language model from ever inventing one, and the open-source tools behind it.

The one rule the whole system is built on

The American Angus Association computes EPDs and $Values through a weekly, breed-wide National Cattle Evaluation, with genomics folded in through a single-step method. No single herd can reproduce that. So this system does not compute EPDs. It consumes the Association’s published numbers as inputs and computes only the math a breeder can legitimately run on top, like a coefficient of inbreeding from a pedigree.

A deterministic engine produces every figure. A reasoning layer plans the work and explains it, but it never emits a genetic number, because a fluent model asked for one will produce a confident guess, and that guess would be fabricated. Every number the system reports is traceable to a calculation you can redo on paper or to a value the Association published.

Write-ups

The open-source tools

The inbreeding and relationship module is on GitHub: pure standard-library Python, no numeric third-party stack, known-answer tests against hand-calculated coefficients, and a standing cross-check that Wright’s path-coefficient method agrees with the tabular relationship matrix for every animal in a pedigree. deluisio-breeding-tools on GitHub.

The plain-language side

The breeder-facing version of all this, written for someone reading a bull’s page rather than the code, lives at Deluisio Angus: what an EPD is, how to read one, and how to use it without getting fooled. If the code side here is the how, that side is the why it matters at the sale barn.