Writing a Coconut Ingredient Spec Suppliers Can Actually Hit

A coconut ingredient spec usually fails at the laboratory bench, not at the mill. The number was never reachable.

An enforceable spec pairs every limit with a named test method, a stated test portion, an expression basis, and a tolerance band wider than that method can resolve between two laboratories. Most buyers get the first part right and the last part wrong. They copy Codex limits into a purchase order, tighten them for comfort, and write a clause no honest supplier can hit twice running.

The cost lands as disputed containers, retests, and a supplier who prices the argument into the next offer.

Why does a tolerance tighter than the test method fail?

Because two competent laboratories testing the same homogenised sample are entitled to disagree, and the standards say by how much.

ISO 5725 defines the reproducibility limit, written R: the difference between two results from different laboratories, with different operators and equipment, that will be exceeded no more than 5% of the time. R runs at roughly 2.8 times the between-laboratory standard deviation. A gap inside R is noise, not non-compliance.

Coconut oil has a published figure, and it is worse than most buyers assume. The collaborative trial behind ISO 3960, run across 23 laboratories in nine countries and reproduced in the International Olive Council’s peroxide value method, used coconut oil as a test sample. At a mean peroxide value of 0.55 meq/kg, the reproducibility limit came out at 0.53 meq/kg, a relative standard deviation of 34.7%.

Read that as a commercial fact. On fresh coconut oil, two honest labs can differ by nearly the whole measured value, and neither is wrong.

Moisture behaves the same way. The ISO 712 trial puts wheat moisture R between 0.49 and 0.68 percentage points, and Codex caps desiccated coconut moisture at 4% m/m. A clause reading “moisture 3.0% plus or minus 0.2%” sets a band of 0.4 points, narrower than the method resolves on an easier matrix.

A one-sided maximum is safer. It still needs a rule at the boundary.

Who pays for measurement uncertainty?

Whoever the contract says. Codex is explicit that the question exists and silent on the answer.

Codex CAC/GL 54-2004 states that measurement uncertainty matters when deciding whether a sample meets a specification. It sets out four situations around a maximum level: the result clears or breaches the limit by more than the uncertainty, or by less. The two middle cases are where containers get stuck.

Codex does not allocate the burden in those middle cases. It requires only that the decision rule be agreed in advance. The companion guidelines, CAC/GL 59-2006, give the benefit of the doubt to the party accused of non-compliance, but that is regulatory practice, not a term in your contract.

An agreement silent on the rule carries an unresolved argument inside every borderline lot. Three lines settle it: name the decision rule, name the tie-break analysis, name who pays for it. The desk has set out those mechanics in multi-origin contracts.

Note also that sampling uncertainty is often larger than analytical uncertainty, and frequently decides conformity outright.

Does the number name a method, a portion and a basis?

A method alone is not enough, and for coconut milk the standard you cited may not carry one.

Codex CXS 240-2003 fixes total solids, fat, non-fat solids and pH across four coconut milk and cream tiers. Its methods clause says only to see the relevant Codex texts. The numbers sit in one document and the methods sit in CXS 234-1999. A purchase order reading “per CXS 240-2003” hands a laboratory four limits and no instructions.

Follow the trail and it goes somewhere unexpected. Codex assigns coconut milk fat to ISO 1211 and IDF 1, the Rose-Gottlieb gravimetric method written for dairy. A laboratory defaulting to a Soxhlet fat method runs a different test on the same sample and returns a different number, legitimately.

Non-fat solids is the sharper trap. Codex derives it by subtracting one gravimetric result from the other. Specify the two measured inputs rather than the calculated figure, or you inherit the uncertainty of both methods inside the single parameter you reject on.

Expression basis is the third leg. Codex sets desiccated coconut acidity at 0.3% m/m of the extracted oil, measured as lauric acid. A line reading “FFA max 0.1%” omits both the extraction step and the acid the result is expressed against. Coconut oil under CXS 210-1999 is controlled by acid value in mg KOH per gram, not by a percentage at all.

Then state the test portion. Codex writes foreign matter as absence in 100 g, not as absence.

Which clauses can you actually reject on?

Not all of them, and Codex publishes the ranking.

Every method in CXS 234 carries a type from I to IV. Type I is a defining method, where the value is whatever that method returns. Type IV is tentative. For desiccated coconut, moisture by AOAC 925.40, oil by AOAC 948.22, ash by AOAC 950.49 and acidity by ISO 660 are all Type I. Extraneous vegetable material is Type IV, and its method is a naked-eye count of fragments in 100 g spread against a white background.

A rejection built on a Type IV count does not carry the weight of one built on a Type I result. Supplier-side counsel know this. Few buyers price it.

Coconut milk pH shows the gap plainly. CXS 240 sets a minimum of 5.9 across all four tiers, and CXS 234 lists no pH method under aqueous coconut products. The buyer must reach into the general processed fruit and vegetable rows and pick between AOAC 981.12 (Type III), NMKL 179 (Type II) and ISO 1842 (Type IV). None is Type I. If the contract does not choose, the supplier’s laboratory chooses.

Sampling needs the same discipline. Codex CAC/GL 50-2004 scopes itself to cases where analytical error is small against sampling error, and puts the opposite case outside the guidelines entirely. A tolerance narrower than R lands you there. The same text warns against composite samples on anything but cost grounds: compositing recovers the lot mean and destroys the variation between drums.

Pathogen absence clauses run on the same statistics. Absent in 25 g describes five samples, not a container, which is why kill-step validation is worth more in a spec than an end-product certificate.

Does the spec catch palm kernel oil?

Most coconut oil specs do not, and closing the hole costs nothing.

Buyers pin lauric acid because it reads as the signature of coconut. Codex puts coconut C12:0 at 45.1 to 53.2% of total fatty acids and palm kernel oil at 45.0 to 55.0%. A spec verifying identity on lauric alone verifies very little.

Three parameters in the same analysis separate them cleanly.

Caprylic and oleic acid come off the gas chromatography run a buyer already pays for. Iodine value is a cheap titration needing no chromatography at all.

What this means for buyers

• Procurement, CPG and food manufacturing: audit your spec for two-sided bands. Any band narrower than the method’s R generates disputes indefinitely, whatever the supplier does.
• Procurement: add a decision rule stating whether a result inside the measurement uncertainty of the limit passes or fails, and who funds the referee analysis.
• Brand owners scaling past a co-packer: for coconut flour and coconut sugar there is no Codex standard at all. Your specification is the only standard in the transaction, so a missing method column leaves nothing to enforce.
• Both: name the method edition, not just the method. Codex amended its fats and oils method list in 2024.
• Both: check identity parameters, not only quality parameters. Lauric acid alone will not separate coconut oil from palm kernel oil.

FAQ

What is a reproducibility limit? The largest difference expected, 95% of the time, between two results from different laboratories testing the same sample with the same method. It is roughly 2.8 times the between-laboratory standard deviation. Differences inside it are measurement noise, not a quality failure.

Can a buyer just tighten the Codex limit? Yes, and a one-sided maximum tighter than Codex is normal commercial practice. The problem is two-sided bands narrower than the test method resolves. Those reject compliant product at a steady rate and cannot be fixed at the mill.

Does a Codex reference make a spec enforceable? Not on its own. CXS 240-2003 carries composition limits but no methods; they sit in CXS 234-1999. Cite both, or name the method and its edition directly in the specification.

Which parameters detect palm kernel oil in coconut oil? Caprylic acid C8:0, oleic acid C18:1 and iodine value. Lauric acid ranges for the two oils overlap almost entirely under Codex CXS 210-1999, so lauric alone is not an identity test.

Need one spec held across four origins? Send the desk an RFQ.