Somebody chose the beans in that pouch. That one sentence explains most of the disappointment in vanilla importing. A sample is not a random handful of a lot; it is a selection, and a selection made by the party trying to sell you the lot. Ten beautiful pods arrive, you approve them, and six weeks later a 25 kg carton turns up wetter, shorter, and speckled with splits and mould you never saw. Nobody necessarily lied. The sample did exactly what a sample does when you let the seller choose it: it showed you the best the lot could offer, and nothing about the rest.
A vanilla sample misleads you because it is small and hand-selected. Ten beans cannot reveal a defect rate: if four percent of a lot is split, a ten-bean sample shows no splits roughly two times in three. To make a sample represent the shipment, ask for about 250 g drawn from across the lot (not hand-picked), require it to arrive with the lot’s own paperwork, measure it at your bench against defined criteria, and write the sample’s figures into the contract so the shipment is held to them.
This is not an argument against sampling. It is an argument for sampling correctly — asking in a way that makes selection hard, then measuring what arrives against what you were promised. Done right, a sample becomes the most powerful lever you have over the shipment. Done the usual way, it is theatre.
The Statistics Nobody Mentions
The core problem is sample size. Defects in a lot — splits, mould, short beans, foreign matter — occur at some rate across the whole mass. A small sample is a poor estimator of a low rate. Suppose four percent of beans in a lot are split. In a ten-bean sample, the chance that no split bean appears at all is about 0.96 to the tenth power, roughly 66 percent. So two times in three, a genuinely four-percent-split lot shows you a clean ten-bean sample. You conclude “no splits,” and you are wrong about the lot while being right about the sample.
Raise the sample to around 250 g — on the order of fifty to seventy beans depending on size — and the maths changes completely. The probability of missing a four-percent defect across sixty beans drops to a few percent. You start to see the distribution, not just the headline. This is the entire reason the serious ask is “send me 250 grams,” not “send me a few beans.” You are buying statistical power, and at current prices the beans cost the supplier less than the courier.
A sample with no defects in it does not tell you the lot is clean. It tells you the sample was small, or sorted, or both.
Five Ways a Sample Misleads You
The supplier picks the best beans. No dishonesty required — anyone sending a sample naturally sends good ones. The lot regresses to the mean; the sample does not.
Too few beans to reveal a low defect rate, as the statistics above show. The sample can be honestly drawn and still mislead purely because it is small.
Some suppliers keep a small quantity of showpiece beans for samples and buy the actual consignment later, from elsewhere, once you commit.
A sample from last season’s best, against a shipment from this season’s bulk. Moisture and suppleness can differ markedly.
None of these requires a villain. They are the default behaviour of a sampling process left unmanaged. Your job is to manage it — to make selection difficult and obvious, and to measure the result so that even a cleverly chosen sample gives up its secrets at your bench.
How to Ask for a Sample That Represents the Lot
The request does most of the work. A well-written sample request does several jobs at once: it specifies the quantity, it specifies that the sample be drawn from across the lot rather than selected, it requires the lot’s own paperwork to travel with it, and it sets expectations for the shipment to follow. Each clause closes a gap that an open-ended “please send a sample” leaves wide open.
- Ask for about 250 g, and offer to pay for it — paying makes the request serious and removes the excuse to send a token handful.
- Specify that the sample be drawn from several points in the lot, not hand-selected, and say so in writing.
- Require the sample to arrive with the lot’s certificate of analysis and a lot or batch identifier you can match to the shipment later.
- State that the shipment will be measured against the sample, and that the sample’s figures will form part of the contract.
- Note how you will handle a pre-shipment sample — a second sample drawn from the actual consignment before it sails.
The request that the sample be drawn from across the lot rather than hand-picked is the one that separates suppliers. A confident producer agrees without fuss. Hesitation, or a counter-offer of a smaller “representative” sample, tells you the lot and the showpiece beans are not the same thing.
Measuring It at Your Bench
A representative sample is only useful if you measure it the same way every time. You do not need a laboratory. A kitchen scale, a steel rule and a clean glass jar will give you most of what matters in about ninety minutes, and the numbers you record become the baseline the shipment must match.
| Measurement | What you need | What it tells you |
|---|---|---|
| Length band and spread | Steel rule | Whether the lot is uniform or a mix of grades sold as one |
| Beans per kilogram | Scale, count | A hard, objective proxy for size and grade that is easy to verify on arrival |
| Moisture, by feel and weight | Scale, time | Whether you are paying vanilla prices for water — the costliest hidden variable |
| The cut test | Knife | Seed density, oiliness and interior condition the exterior hides |
| The jar test | Sealed jar, 24–48 h | Aroma development, and any sign of fermentation or mould the surface did not show |
| Defect count | Eyes, patience | The split, mouldy and foreign-matter rate across the whole 250 g — the number a ten-bean sample could never give you |
Record every figure on one sheet, score it against defined bands, and apply a small number of hard stops — mould above a threshold, a vanillin or moisture figure outside the agreed range — that fail a sample regardless of how it scores elsewhere. The point is consistency: the same bench, the same criteria, the same sheet, so that two suppliers’ samples and, later, the shipment itself are all judged on identical terms.
Making the Sample Binding
The final step is the one most buyers skip, and it is the one that makes all the measuring worthwhile. The figures you recorded from the sample should go into the contract as the specification the shipment must meet: a moisture range, a minimum bean count, a maximum defect rate, a vanillin floor on a stated basis. Then a pre-shipment sample, drawn from the actual consignment, is measured the same way before the container sails. If it misses, you have a defined remedy — and leverage, because you have not yet paid in full.
If the sample’s figures never enter the contract, you have no mechanism to reject a shipment that fails to match it. The supplier can truthfully say they never promised the lot would equal the sample — because you never wrote it down. The measuring only pays off when the numbers become terms.
What to Accept When a Supplier Won’t Send 250 g
Not every supplier will send a full 250 g sample, and the reasons vary from legitimate to revealing. A small producer early in the season may genuinely have little loose stock; a large trader who refuses while sitting on tonnes is telling you something else. The workable middle ground is a smaller sample paired with stronger paperwork: accept, say, 100 g if it arrives with a lot-specific certificate of analysis and a clear lot identifier, and make the pre-shipment sample — drawn from the actual consignment — the real gate. What you should never accept is a hand-selected showpiece handful with no lot reference and no paperwork, offered in place of anything representative, because that is a sample engineered to tell you nothing.
It also helps to read the refusal itself. A supplier who explains the constraint, offers an alternative, and agrees to a pre-shipment sample is negotiating in good faith. One who simply insists their beans are “all Grade A, no need to test” and resists any representative draw is signalling that the lot and the sample are not the same thing — which is the entire problem this process exists to catch.
The Sample Request & Evaluation Pack is built to run exactly this sequence: five ready-to-send emails (including the drawn-not-selected request and the pre-shipment follow-up), a ten-measurement bench protocol with a defect atlas, an Excel scoring sheet that returns the weighted score and any hard stop, and the contract wording that ties the shipment to the sample. It turns a pouch of pretty beans into a test the whole container has to pass.
Frequently Asked Questions
Why doesn’t a vanilla sample match the shipment?
Because a sample is small and usually hand-selected by the seller. A few beans cannot reveal a lot’s defect rate or variance, and the beans chosen for a sample are naturally the best ones, so the lot regresses toward its average while the sample does not. The fix is a larger sample, drawn from across the lot rather than selected.
How big should a vanilla sample be?
Around 250 grams — roughly fifty to seventy beans depending on size. That is enough to reveal the distribution of length, moisture and defects, whereas a ten-bean sample can miss a four-percent defect rate about two times in three purely because it is too small.
Will suppliers actually send 250 g of vanilla as a sample?
Most will, particularly if you offer to pay for it, because at current prices the beans cost them less than the courier. The important part is specifying that the sample be drawn from across the lot rather than hand-picked, and watching how the supplier responds to that condition.
Can I evaluate a vanilla sample without lab equipment?
Yes. A kitchen scale, a steel rule and a clean sealed jar let you measure length and spread, beans per kilogram, moisture by feel and weight, a cut test and a jar test, plus a defect count across the whole sample — in about ninety minutes. Lab analysis is a separate step for vanillin and safety parameters.
What is a pre-shipment vanilla sample?
A second sample drawn from the actual consignment just before it ships, measured the same way as the first. It catches the gap between the showpiece sample you approved and the lot that will actually travel, while you still have the leverage of unpaid balance to insist on a remedy.
How do I make a vanilla sample binding on the supplier?
Write the sample’s measured figures — moisture range, minimum bean count, maximum defect rate, vanillin floor on a stated basis — into the contract as the specification the shipment must meet, then test a pre-shipment sample against them. Without this, the supplier never promised the lot would match the sample.
Does a flawless vanilla sample mean the lot is clean?
No. A sample with no defects may simply be small, sorted, or both. A clean ten-bean sample is fully consistent with a lot that has a meaningful defect rate; only a larger, drawn sample and a defect count give you a reliable read on the lot itself.
Further reading: Codex Alimentarius sampling plans