Why one industry index can never be right for every producer.

This is the central problem Genemap exists to solve. It sounds obvious when stated. The maths behind it is straightforward — Hazel published the foundation in 1943, the discounted gene flow extension was worked out by Brascamp in the 1970s, and the bioeconomic literature has had a settled answer to "how do you derive economic weights for traits" for forty years. What's been missing is the infrastructure to actually do it per producer instead of once per breed.

This piece walks through why the per-farm derivation matters, what changes when you do it, and what it means for the way producers should evaluate breeding stock.

The selection-index foundation

Selection-index theory, in one sentence: the most profitable animal to breed from is the one whose breeding values, multiplied by the dollar value each trait actually delivers in your operation, sum to the highest number.

The maths is just a dot product:

$Index = Σ (EBV_i × economic_weight_i)

The hard part has never been the dot product. It's the economic weights. Those are the dollar value per breeding-value unit on your specific farm, and they depend on:

• Your breeding system — terminal sire, F1 cross, self-replacing straight, self-replacing composite, retained finisher, trading.
• Your calving time — spring, autumn, split, year-round.
• Your joining length — 6 weeks vs 12 weeks materially changes the value of fertility EBVs.
• Your retention strategy — what % of heifers you keep, what % of progeny you finish vs sell.
• Your target grid — MSA grassfed, MSA grainfed, Wagyu long-fed, weaner sale, live export, EUROP-graded EU.
• Your gross margin per head — the calibration anchor that keeps the whole index honest.

Put two operations side by side and almost every coefficient in the equation differs. Yet for fifty years the industry has computed one set of weights per breed and applied them to every producer in that breed. The "average" producer that index is calibrated to does not actually exist — it's a statistical fiction, useful for breed-society publication but inadequate for individual decision-making.

Two operations, one breed, two completely different answers

Take a concrete example. Two real Angus operations both reading the same Angus Australia BREEDPLAN evaluation:

The marbling EBV is worth 35× more on the long-fed grid composite operation than on the pastoral commercial. The fertility EBV is worth 2.3× more on the pastoral. Mature cow weight is 3× more punitive on the pastoral — every extra kg of cow has to be fed across a much wider seasonal feed gap. And so on.

Now apply the same Angus Selection Index to both operations. It's calibrated to "the average Angus producer" — somewhere statistically between these two operations, but materially different from both. The composite breeder ranks bulls in an order that systematically under-weights marbling. The pastoral commercial ranks them in an order that systematically under-weights fertility and over-weights growth-related carcase traits.

Both breeders end up with sub-optimal genetics for their actual operations — by industry-index design.

The discounted gene flow correction

The above weights are per-cow per-year of direct expression. They under-count maternal traits, because milk and fertility EBVs are expressed not just in this calf but in every retained daughter, and then again in every granddaughter the daughter produces. Brascamp's 1978 contribution was the framework for valuing every gene's lifetime expression, NPV-discounted at a sensible rate (Genemap defaults to 5% p.a.).

This matters most for self-replacing operations. A self-replacing herd retaining 30% of heifers is going to express a milk EBV in the original calf, plus 30% of cases in a daughter, plus ~9% of cases in a granddaughter, plus a tail of further generations. The lifetime gene flow can be 1.7–2.4× the per-cow per-year value depending on retention rate, generation interval and the discount rate.

For a terminal sire operation, where every progeny is slaughtered, the daughter pathway delivers nothing. Maternal traits revert to roughly zero economic weight. Industry indexes that quietly assume some baseline retention rate get the answer wrong for both ends of this spectrum.

The closed-loop calibration step

So far this is all first-principles. But "first principles" is also "what we think traits are worth." What they're actually worth on your operation is what you realise at the kill floor.

This is the step that takes per-farm bioeconomic indexing from "better than industry default" to "actually right." Every realised slaughter outcome — HSCW, MSA index, EUROP grade, USDA Quality grade — is an empirical observation of how your specific operation converts breeding values into dollars.

Multivariate ridge regression with VIF diagnostics is the standard machinery here. You fit:

realised_$/head ~ EBV_WW + EBV_YW + EBV_MARB + EBV_REA + EBV_FAT + ...

Across enough kill records, the regression coefficients become the empirical economic weights for your farm. The first-principles defaults from Phase 5.0a/b stay in as priors; the longer you use the platform, the more weight the regression takes from the priors. This is Bayesian calibration in the sense Hayes and others have written extensively about.

The longer you use Genemap, the more your weights come from your own data, not the textbook.

The per-farm GWAS layer

Phase 5.0d goes one further. Where producers have SNP genotypes on their animals, Genemap fits per-trait DGV (direct genomic value) ridge regressions against the producer's own realised profit. This is per-farm GWAS — same statistical machinery as the world's most advanced commercial breeding programs (PIC, Aviagen, Cobb), applied at the individual producer level.

The output is a per-animal, per-trait DGV that's specific to your reference population. It complements the central ssGBLUP coming from your breed society's genomic evaluation pipeline. Critically, it lets producers in regions or breeds with thin national reference populations still get useful genomic prediction — just calibrated against their own data.

What this means for breeding decisions

What this means for breed societies

Industry indexes aren't going away. They serve a real purpose — a single number that everyone can compare bulls against, useful for breed publication, sale catalogue ranking, and as a default for producers who don't want to go deeper. Breed societies will keep publishing them, and producers without specific operation context will keep using them.

What changes is that every producer with specific operation context now has the option to do better. The infrastructure to derive a per-farm bioeconomic index, calibrate it against realised slaughter outcomes, layer in per-farm genomic prediction and re-derive weights daily as the market moves — that infrastructure now exists. It's no longer a research-paper curiosity; it's a self-serve product.

The breed society's role evolves from being the only publisher of selection signals to being one (very important) input into a per-farm engine that combines society EBVs, the producer's economics, the producer's climate, the producer's market and the producer's outcomes. The genetic evaluation work the breed societies do remains foundational — Genemap reads their evaluations natively for 18 countries and via AI translation for the long tail. The per-farm derivation sits above that work, not in competition with it.

The mathematics is open

Every coefficient in the Genemap engine is documented as code. The bioeconomic derivation lives in core/js/ai-enterprise-cattle.js; the discounted gene flow extension is wired in alongside; the closed-loop calibration is in calibration-bayesian.js and calibration-online.js. Override any coefficient on the rank page and watch the rankings shift in real time. The engine is the product, and the product is fully open to the producer who runs it.

The full methodology page documents the architecture; the research & references page lists the academic citations behind every load-bearing module.