The GOLD Score is a quality benchmark score ranging from -50 to +50 that evaluates soybean meal against the reference population. 0 is the per-pillar benchmark anchor. The typical lot does not score 0, though — because two pillars reward absolute quality, the global median lands around +8, so a lot is read as above or below that typical lot (as its scorecard labels it), not against 0 (see “Reading the Score” below).
The GOLD Score uses a quantile-based benchmark method across 6 scored pillars (10 individual metrics): CP 20% · Reactive Lysine 20% · Essential Amino Acids 20% · Sugars 15% · KOH 15% · TIA 10%. Reactive Lysine is scored on the total reactive-lysine amount — a dedicated digestibility pillar, separate from total lysine, which stays within the Essential Amino Acids pillar. For protein, sugars, amino acids and reactive lysine, each metric is positioned within the benchmark distribution using quantile interpolation, with inputs clipped at the 1st/99th percentile. Crude protein earns full credit at the 95th percentile. KOH and TIA use fixed utility curves, and TIA ≥ 6.0 mg/g triggers a hard-reject safety gate. The system is designed around two core principles: simplicity — a single number that captures multi-dimensional quality — and transparency — every weight, threshold, and formula is documented and inspectable.
The reference population is the rolling 3-year (trailing 36-month) window of the Evonik laboratory dataset for the GOLD-reference origins (Argentina, Brazil and USA). Paraguay and Bolivia are available as reportable origins for exploration but are not part of the reference benchmark.
Example: a lot scoring 12.4 — above the typical lot (median score ≈ +8)
The distribution is left-skewed: the most common score (the curve's peak) is around +21, but a long lower tail pulls the median down to +8(half of lots score below it) and the mean to +5.6. The dashed line marks the median — the typical lot — not the peak.
Two soybean meal lots can both pass purchase specifications yet deliver meaningfully different nutritional outcomes. Traditional pass/fail specifications don't capture the quality gradient within “acceptable” ranges. The GOLD Score addresses this gap by measuring how much a lot deviates from the Reference Population benchmark — turning a binary pass/fail into a continuous quality signal.
The GOLD Score follows the same relative evaluation paradigm used in two well-established domains:
Expected Progeny Differentials (EPDs)
In livestock breeding, animals are evaluated relative to their peers rather than on an absolute scale. An EPD of +5 means the animal is expected to produce offspring 5 units above the breed average. The GOLD Score applies this same concept to soybean meal quality.
Corn Suitability Ratings (CSRs)
In agricultural land evaluation, CSRs rate land relative to a reference standard rather than assigning an absolute productivity number. This relative approach makes comparisons meaningful across different contexts — the same principle behind the GOLD Score.
The GOLD Standard methodology was informed by a survey conducted by MIT that tested whether reactive lysine should become its own quality pillar and how a six-pillar weighting would behave against the Evonik reference data. Three candidate designs were scored against the empirical Evonik distribution:
Option 1 — the adopted GOLD Standard model
Keeps crude protein and the amino-acid profile and adds a standalone reactive-lysine pillar: CP 20 · Reactive Lysine 20 · 5 EAA 20 · Sugars 15 · KOH 15 · TIA 10. It adds a dedicated available-lysine (digestibility) pillar while tracking the underlying survey closely, and preserves the expected ordering across origins.
Option 2 — KOH-consolidated (not adopted)
Has no separate reactive-lysine pillar; instead KOH is raised to 35% as a stand-in for digestibility. This only works under the explicit assumption that KOH solubility carries the reactive-lysine role — KOH and reactive lysine are related but different measurements — and it inflates the whole score scale.
Option 3 — Simplified index (not adopted)
Drops the crude-protein and amino-acid pillars entirely, scoring only reactive lysine, KOH, sugars and TIA. This is a materially different product — it no longer reflects protein level or the amino-acid profile — so it was not adopted.
The GOLD Standard scores the reactive-lysine pillar on the total reactive-lysine amount (Evonik “Lys Reactive”) — the unblocked, chemically available lysine content — which measures how much digestible lysine the meal actually delivers to the animal. Reactive lysine is a required input — a lot cannot be scored without it (scored measures are never neutral-imputed). In the reference data the assay covers ~77% of lots, and the complete-panel ruler uses only lots that report it.
Reactive lysine captures protein quality that crude protein and the amino-acid profile alone can miss: two lots with the same total lysine can differ in how much of it is still available after processing.
The GOLD Score is a continuous deviation from −50 to +50. 0is the benchmark anchor — a lot at the median on the rank-based pillars and neutral on the absolute-quality curves. A positive score is net above that anchor, a negative one net below — but the typical lot sits well above 0 (≈ +8, next), so “above/below” is judged against the typical lot, not 0.
The typical lot does not score 0. Across the reference population the global median is about +8.2 (mean +5.6), and roughly 63% of lots score at or above 0. This is by design, not a miscalibration: two of the six pillars — KOH solubility and TIA — are judged against absolute agronomic quality thresholds, not relative rank. Most soybean meal is genuinely good on both, so it legitimately earns positive points there. Of the other four pillars, three are rank-based — crude protein, the five amino acids, and reactive lysine — so a median value contributes 0; sugars is scored on a fixed asymmetric curve whose median sits just below neutral (−0.75, see the table).
Two numbers to keep straight: the median of actual lot scores is about +8.2 — the honest “typical lot.” Scoring a single all-median profile gives about +9.8, higher because a real lot is usually penalized on its weakest attribute, which an idealized all-median profile avoids.
As a safety gate, a lot with TIA ≥ 6.0 mg/g is flagged Rejected regardless of its raw score, and the raw score is still computed and shown for diagnostics.
The GOLD Standard benchmarks each lot against the rolling 3-year reference distribution and converts the result into a single score from -50 to +50. Inputs are clipped at the 1st/99th percentile before scoring.
1. Position within the benchmark distribution
For each amino acid and for reactive lysine (the total reactive-lysine amount), estimate where the lot sits within the benchmark distribution using the benchmark quantiles: q05, q25, q50, q75, q95. Percentile positions are clipped to the 1st/99th percentile.
Then convert that percentile position into a utility score:
A value at the 5th percentile of the benchmark scores 0, the median scores 0.5, and the 95th percentile or above scores a full 1 — values in between scale proportionally.
Crude Protein uses diminishing returns above the median: below q50, standard percentile-to-utility; above q50, utility flattens through anchors q50→0.50, q75→0.75, q95→1.00, so crude protein earns full credit at the 95th percentile.
Sugars uses an asymmetric curve: zero utility at or below q25, rising to 0.45 at q50, 0.85 at q75, and 1.00 at q95. Low sugars are penalized more heavily than high sugars are rewarded.
A utility score of 0 means clearly weak versus benchmark
A utility score of 0.5 means roughly benchmark-average
A utility score of 1 means clearly strong versus benchmark
2. KOH practical optimum range
KOH is not treated as “higher is always better.” Instead, the GOLD Standard uses a practical optimal range informed by industry processing-quality feedback:
0 utility at 70 or below
Utility rises linearly from 70 to 78
Full utility from 78 to 85
Utility falls linearly from 85 to 88
0 utility at 88 or above
This captures both under-processing and over-processing. The wider plateau (78–85) reflects domain consensus that the practical optimum extends higher than originally modeled.
3. TIA plateau + staged penalty + reject gate
TIA is not rewarded indefinitely for becoming lower and lower. The GOLD Standard uses a plateau with staged penalty and a hard reject gate:
Full utility from 2.0 to 3.0 mg/g (practical optimum)
No extra reward for TIA below 2.0 — utility stays at 1.0
Decline to 0.50 at 4.0
Steeper decline to 0 at 5.5
TIA ≥ 6.0 mg/g → REJECTED regardless of other metrics
Beyond the utility curve, elevated TIA also trims the combined score: for TIA above 3.5 mg/g, roughly four percent of the amount over 3.5 is subtracted before scaling. Additionally, when TIA is below 1.5 mg/g and KOH is below 75%, a small penalty is applied to the TIA pillar — this combination may signal over-processing where excessive heat has deactivated TIA but also damaged protein solubility.
4. Amino acid subweights
Each amino acid is scored separately against the benchmark distribution and contributes directly to the composite with its own explicit weight:
Lysine (LYS) — 8%
Methionine (MET) — 8%
Cysteine (CYS) — 2%
Threonine (THR) — 1%
Tryptophan (TRP) — 1%
These five AA subweights sum to 20%, the total Essential Amino Acids pillar weight. Reactive lysine is scored as its own 20% pillar (the total reactive-lysine amount), separate from total lysine here. Lysine and Methionine carry the most influence as the two most limiting amino acids in soybean meal for monogastric nutrition.
5. Center each component and build the weighted composite
Convert each utility score to a centered score:
Each utility is re-centered around the benchmark so a benchmark-average value becomes 0, weaker values go negative, and stronger values go positive (a −1 to +1 scale):
−1 = weak
0 = benchmark-neutral
+1 = strong
Each centered score is then multiplied by its pillar weight and summed: Crude Protein 20%, Reactive Lysine 20%, Essential Amino Acids 20% (Lysine 8, Methionine 8, Cysteine 2, Threonine 1, Tryptophan 1), Sugars 15%, KOH 15%, and TIA 10%.
6. TIA composite haircut + final GOLD Score
Before scaling, elevated TIA trims the combined score once more: for TIA above 3.5 mg/g, about four percent of the amount over 3.5 is subtracted.
Finally, the combined score is scaled onto the −50 to +50 range to give the GOLD Score.
If TIA ≥ 6.0 mg/g, the displayed status is Rejected regardless of the raw score. The raw score is still computed for diagnostic purposes.
How to read the score
0 = per-pillar anchor (typical lot ≈ +8)
positive = above the anchor
negative = below the anchor
Contribution of each metric to the final score (must sum to 100)
Primary indicator of nutritional value. Below the benchmark median, higher protein earns proportional credit. Above the median, returns diminish, and crude protein earns full credit at the 95th percentile. Amino acids and reactive lysine carry their own pillars, so protein level and protein quality are scored separately.
Scored on the total reactive-lysine AMOUNT (Evonik 'Lys Reactive', typically ~2.3–2.7%) — the unblocked, chemically available lysine content of the meal. Higher reactive lysine means more digestible lysine reaches the animal. Reactive lysine is a required input — a lot cannot be scored without it (scored measures are never neutral-imputed). In the reference data the assay covers ~77% of the window; the complete-panel ruler is built only from lots that report it.
Composite of 5 essential amino acids with explicit subweights: Lysine (8%), Methionine (8%), Cysteine (2%), Threonine (1%), Tryptophan (1%). Each AA is scored against the benchmark distribution and contributes directly to the composite with its own weight.
Energy proxy contributing to metabolizable energy in the feed ration. Low sugars (≤q25) receive zero utility — a meaningful penalty. Above the median, sugars earn strong credit up to full utility at q95.
Measures protein digestibility after processing. Scored with a practical optimum curve: full utility at 78–85%, linearly decreasing outside this range. Below 70% or above 88% receives zero utility.
Anti-nutritional factor that inhibits protein digestion. Full utility at TIA 2.0–3.0 mg/g (practical optimum). Decline to 0.50 at 4.0, steeper decline to 0 at 5.5. TIA ≥ 6.0 mg/g results in automatic rejection regardless of other metrics. A 0.10 utility penalty is applied when TIA is below 1.5 and KOH is below 75% (under-processing signal).