GOLD STANDARD

Scoring Methodology

What the GOLD Score measures

Two lots of soybean meal can both pass a purchase specification and still feed differently. A pass/fail spec cannot express the quality gradient inside the acceptable range. The GOLD Score does: it measures how far a lot sits from a reference population of 44,147 laboratory-analysed lots and expresses that as a single number from −50 to +50.

Six weighted pillars contribute, across ten measured analytes. Four are scored by rank against the reference distribution; KOH solubility and trypsin inhibitor are scored against fixed agronomic thresholds, because for those two the desirable range is known in absolute terms rather than relative to peers.

The approach is the one used for Expected Progeny Differences in livestock breeding and Corn Suitability Ratings in land valuation: rate against a defined reference rather than on an absolute scale, so that a number means something comparable.

0median +7.9−50+50
The reference population is left-skewed. Most lots score in the twenties; a long lower tail pulls the median to +7.9 and the mean to +5.3. The middle half falls between -9.9 and +21.3.

Zero is the anchor, not the average

This is the one part of the scale that is counterintuitive, so it is worth stating plainly. Zero means a lot is at the median on every rank-scored pillar and neutral on the fixed curves. It is not the typical score: the median lot scores +7.9 and 62.5% of lots score at or above zero.

That is by design rather than a miscalibration. KOH solubility and trypsin inhibitor are judged against absolute quality thresholds, and most soybean meal is genuinely good on both — so a typical lot earns real positive credit there while scoring 0 on the four rank-based pillars.

How each pillar scores a lot sitting at the reference median
PillarScored byAt the median
Crude proteinrank0
Reactive lysinerank0
Amino acidsrank0
Sugarsrank0
KOH solubilityabsolute threshold+2.8
Trypsin inhibitorabsolute threshold+5.0
A profile at every median+7.9

Two figures worth keeping apart: the median of real lot scores is +7.9 — the honest typical lot — while scoring a single synthetic profile set to every median gives +7.9. The two are close, and which is higher is not a fixed property of the scale: it depends on the fixed-curve pillars. A real lot is let down by its weakest rank-scored attribute, which an idealised profile has none of — but an all-median profile also inherits the median KOH, and since that sits below the full-credit threshold the idealised profile is docked there too. Note that the median lot's KOH solubility is 76.86%, below the 80% full-credit threshold — the typical lot does not earn full marks on that pillar.

The six pillars

Each pillar converts a measured value into a utility from 0 to 1 — 0 being clearly weak, 0.5 benchmark-average, 1 clearly strong. The curves below are drawn from the scoring code itself. The shaded band shows where the middle half of the reference population falls and the dashed line marks its median, so each curve also shows what a typical lot earns.

Crude protein

20%

Higher is better

The primary indicator of nutritional value. Scored on where the lot sits in the reference distribution, with diminishing returns above the median: full credit is reached at the 95th percentile rather than continuing to reward ever-higher protein.

00.51385347.13
  • Protein level and protein quality are scored separately — the amino acids and reactive lysine carry their own pillars.

Reactive lysine

20%

Higher is better

The unblocked, chemically available lysine content — how much digestible lysine the meal actually delivers. This is the pillar that captures protein quality that crude protein and the amino-acid profile can miss: two lots with identical total lysine can differ in how much of it survived processing.

00.511.9532.529
  • Scored on the reactive-lysine amount (Evonik “Lys Reactive”), typically 2.3–2.7%.
  • Required: a lot cannot be scored without it, and it is never substituted with an average. The assay covers about 77% of lots in the reference window, and the benchmark is built only from lots that report it.

Sugars

15%

Higher is better (provisional)

An energy contributor to the ration, scored on the same percentile scale as protein and the amino acids.

00.514139.4

The laboratory field reported as “Sugars” has not been confirmed in writing as sucrose specifically rather than total soluble sugars. Sucrose is a digestible energy source, so higher is better; total soluble sugars would also include raffinose-family oligosaccharides, which are anti-nutritional for pigs and poultry and for which higher would not be better. Read the direction of this pillar as provisional until that is confirmed.

KOH solubility

15%

Higher is better, up to 80%

How much protein remains soluble after heat processing, which tracks how much heat damage the meal took. Credit rises from zero at 70% to full at 80%, and then stops: a lot at 90% earns exactly what a lot at 80% earns, no more. So this pillar rewards escaping heat damage rather than maximising solubility. There is no penalty above the threshold either, because corrected amino-acid digestibility plateaus rather than declining.

00.51509776.86

Because the curve has no upper limit, this pillar on its own detects heat damage but does not confirm correct processing — very high solubility can accompany under-processed material. Under-processing is carried by the trypsin-inhibitor pillar instead, which does penalise it heavily. But be specific about where that leaves a gap: among lots above 85% solubility, 41.7% also have trypsin inhibitor low enough to earn full marks there, so for those lots neither measure comments on processing. Read a high KOH figure as “not heat damaged” rather than “correctly processed”.

Trypsin inhibitor

10%

Lower is better, to a point

An anti-nutritional factor that suppresses protein digestion. Unlike the other pillars this one is not rewarded without limit for falling: there is a practical optimum, below which further reduction earns nothing.

00.51092.51
  • Full credit from 2 to 3 mg/g, and no additional credit below 2.
  • The curve above is drawn at the reference median KOH solubility of 76.86%. This pillar's curve depends slightly on KOH: a small extra penalty applies when trypsin inhibitor is very low AND solubility is also low, which the drawn curve therefore does not show.
  • Above 3.5 mg/g the whole composite is also trimmed, by about 4% of the excess.
  • At or above 6 mg/g the lot is rejected outright, whatever else it scores.
  • A small penalty applies when trypsin inhibitor is below 1.5 mg/g and KOH is below 75% — together these suggest heat severe enough to have deactivated the inhibitor and damaged the protein.

Essential amino acids

20%

Higher is better

Five amino acids, each scored by rank on the same curve and each carrying its own weight into the composite. Lysine and methionine dominate because they are the two most limiting amino acids in soybean meal for pigs and poultry.

Lysine
8%
Methionine
8%
Cysteine
2%
Threonine
1%
Tryptophan
1%
00.512.23.32.856

The shared rank curve, shown for lysine: zero at or below the 5th percentile (2.711%), 0.5 at the median (2.856%), full credit at the 95th (2.949%). Reactive lysine is scored separately, as its own pillar.

How the pillars become one score

  1. 1.Each measured value becomes a utility from 0 to 1 on its own curve. Values are clipped at the 1st and 99th percentile before scoring, so a single extreme result cannot dominate.
  2. 2.Each utility is re-centred to a −1 to +1 scale, so that benchmark-average becomes 0, weaker goes negative and stronger goes positive.
  3. 3.Each centred value is multiplied by its pillar weight and the results are summed.
  4. 4.Elevated trypsin inhibitor trims the sum: above 3.5 mg/g, about 4% of the excess is subtracted.
  5. 5.The result is scaled onto the −50 to +50 range. A lot at or above 6 mg/g trypsin inhibitor is reported as Rejected whatever that number is; the score is still computed, for diagnosis.

You can work through this interactively on the score calculator, which shows each pillar's utility and contribution as you change a value.

The reference population

Every score is relative to a defined population, so what that population is matters as much as the arithmetic.

Source
The Evonik laboratory dataset.
Window
A rolling three-year window of the trailing 36 complete months, so the benchmark tracks current material rather than a fixed historical baseline.
Origins
Argentina, Brazil and the USA — the three origins that supply the large majority of internationally traded soybean meal, and 89% of the lots in this dataset. Holding the benchmark to these three keeps the yardstick stable: a thinly sampled origin cannot move it from one window to the next.
Paraguay and Bolivia
Scored and reported, but deliberately outside the benchmark. That is what makes their scores useful rather than a caveat — the yardstick stays fixed, so a Paraguayan lot is measured on the same scale as every other, against the mainstream traded origins. Read their origin medians with the sample size in mind: 1,349 Paraguayan and 4,091 Bolivian lots, against 9,567 to 18,444 for each reference origin.
Lots included
44,147 lots, de-duplicated, and restricted to those reporting all ten scored analytes. A lot missing any scored measure is excluded rather than filled in with an average.

Because the window rolls, the benchmark moves as new material is analysed. A score is therefore a statement about a lot relative to a stated window, and every certificate records the window and methodology version it was issued under.

Known limits

  • The sugars analyte is not yet confirmed as sucrose rather than total soluble sugars, which would change whether higher is better.
  • Under-processing is measured indirectly, and there is a known gap. Trypsin inhibitor carries it and does so effectively in general — lots above 3.5 mg/g have a median score of −12.3 and those above 4.5 mg/g of −22.4. But the panel includes no urease or protein-dispersibility assay, the other two standard checks; KOH solubility contributes nothing at the top of its range; and for the 41.7% of above-85% lots whose trypsin inhibitor is also low, no pillar comments on processing at all. A lot at 47.5% protein, 95% solubility and 2.5 mg/g trypsin inhibitor scores +14.8, above the reference median, and 95% solubility is in the range associated with raw or under-toasted material. Whether to close this with a cap or a dedicated processing gate is an open decision.
  • The score reflects laboratory analysis of a sample. It is not a measurement of animal performance, and it does not model price.

Methodology version v3.2. Weights, curves and thresholds on this page are read from the scoring engine, not restated by hand.