The lot was rejected on a moisture result of 0.6%. The supplier's own COA said 0.1%. Both laboratories were competent, both used the same method, and both were right: the sample sent for testing had spent four days in a half-filled polyethylene bag in a warehouse in July, and anhydrous citric acid is hygroscopic. The material was never at 0.6%. The sample was.
Sampling is treated as the cheap step before the expensive step, which is backwards. A laboratory measures a few grams with an uncertainty of a few percent; a sampling plan decides whether those grams came from anywhere near the average of twenty tonnes. The second error is routinely an order of magnitude larger than the first, and unlike the first it leaves no trace in the report.
Decision one: how many packages to open
Nearly every general sampling standard converges on a square-root relationship between lot size and sample size. The version used for chemical products in China, GB/T 6678, and the paint and raw material standard ISO 15528, both work from the same logic: sampling effort should grow with the lot, but far more slowly than the lot does. In practice the rule reads as follows.
| Packages in the lot | Square root | Packages to sample | Fraction of the lot inspected |
|---|---|---|---|
| 9 | 3 | 3 | 33% |
| 100 | 10 | 10 | 10% |
| 400 | 20 | 20 | 5% |
| 1,600 | 40 | 40 | 2.5% |
Two things follow. The first is that the square-root rule deliberately accepts a declining coverage fraction: it is a plan for detecting gross heterogeneity, not for certifying every drum. The second is that the plan only works if the packages are opened at random or by a defined pattern. A sampler who takes the top three bags off a pallet has sampled the top three bags off a pallet.
Where the lot is known to be at risk - a first shipment from a new supplier, a material with a history of segregation, a powder that has been in transit through a hot climate - the answer is not a bigger composite. It is more individual samples, tested or at least retained separately.
Decision two: where in the package
Liquids stratify. Water settles, light fractions rise, and additives that are marginally soluble come out of solution at the bottom of a drum that has been cold for a month. ISO 3170 handles this by defining sampling positions rather than a single dip: upper, middle and lower levels, taken separately or combined into an all-levels sample by lowering a stoppered bottle and opening it at the target depth. For a road tanker or a storage tank, an all-levels sample is the correct default; for a drum, top and bottom are not optional extras when the material has been sitting.
Viscous and solid-at-room-temperature products need a different approach again. Paraffin wax 60 (CAS 8002-74-2) has to be melted to be sampled at all, and the melting step is where the sample is damaged: heat it well above its congealing point and you accelerate oxidation, drive off light ends, and produce a darker wax with a higher acid value than the lot actually has. The working rule is to melt at the lowest temperature that gives a pourable liquid, keep the hold time short, and stir before taking the aliquot, because wax that has been melted and left to stand separates again.
Powders segregate by size. Vibration in transit moves fines down and large particles up, or the reverse depending on density and shape, so the top of a big bag is often a different material from the bottom in everything except chemistry. A sampling thief inserted vertically through the full depth is the minimum; a scoop from the surface is not a sample of the bag. Solids that can be handled are then reduced by quartering - pile, divide into four, keep two opposite quarters, repeat - because a riffler or a quartering step is what turns a kilogram of mixed particle sizes into a 100-gram aliquot with the same size distribution.
Decision three: composite or keep them apart
Compositing is convenient and it destroys information. Take ten increments from ten drums, mix them, and the laboratory result is the mean of the lot. If nine drums are at 0.05% water and one is at 2.0%, the composite reads 0.245% - a number that fails a tight specification and passes a loose one, while telling you nothing about the drum that is actually wet.
The rule that resolves this is to separate the purposes. Use the composite to answer "is the lot, on average, what we ordered". Use individual samples to answer "is any part of the lot unusable". Food, pharmaceutical, cosmetic and food-contact materials need the second answer, because a mean is not a safety property: a lot with one contaminated drum is not 90% acceptable.
Microbiological samples cannot be composited at all. The organisms do not average, they multiply, and mixing a clean increment with a contaminated one produces a sample whose count depends on how long the two sat together before the lab received it.
Decision four: the container is part of the sample
| Container | Use | What it does to the sample |
|---|---|---|
| Amber glass, screw cap with PTFE liner | Most liquids, fragrances, esters, products sensitive to light | Low adsorption, low permeability, but breakable and unsuitable for caustics |
| HDPE | General liquids, aqueous solutions, bulk routine samples | Permeable to some solvents and odour compounds; can adsorb hydrophobic actives |
| Tinplate or coated steel | Oils, waxes, non-aqueous products | Rust risk with any water content; lining can be attacked by esters and acids |
| Sterile bag or vial | Microbiology only | Any reuse or composite invalidates the result |
Headspace is the field most often got wrong. A bottle filled to the neck with a volatile solvent is safer than one filled halfway, because every cubic centimetre of headspace is a place for the light fraction to go. Leave the small ullage that thermal expansion needs - on the order of 5 to 10% of the container - and no more, and fill at the temperature the sample will be stored at rather than while it is still warm.
Water-sensitive powders need the opposite discipline: fill the container, close it immediately, and where the material is genuinely hygroscopic, use a container with a desiccant in the cap or double-bag with the air pressed out. Opening a bag of anhydrous citric acid in a humid room and leaving it open while the paperwork is completed is enough to move the moisture result above the specification before the sample ever reaches a balance.
Four cases where the sample produced the number
- Anhydrous versus monohydrate citric acid. The anhydrous grade (77-92-9) and the monohydrate (5949-29-1) are different products with different water contents, and anhydrous material held in a damp container drifts toward the monohydrate's water figure. A buyer comparing a 0.4% result against a 0.1% specification is often comparing a stale sample with a fresh lot. Seal it, weigh it, and test water first.
- Overheated wax. A wax sample taken by melting the whole drum at 120 C and then leaving it on a hot plate while three other drums are sampled will read high in acid value, dark in colour and low in penetration, and the supplier will be told to improve a lot that was never out of specification.
- Mesh size mistaken for chemistry. Zinc stearate (557-05-1) at 325 mesh and at 800 mesh are the same CAS and different products: bulk density, flow, dispersion time and dust behaviour all move with particle size, and a sieve analysis is the only way to tell them apart. A sample that was not protected from segregation in transit cannot support a mesh claim.
- Cross-contamination from the tool. The same stainless thief used to sample a preservative drum and then a food-grade lot carries enough active to fail a microbiological specification at the receiving end and enough odour to be rejected by a customer. A preservative that is legal at 0.1% in a cosmetic is a contaminant at 10 ppm in a food ingredient. Sampling tools belong to a product family, and the rule is easier to enforce than to explain in an audit.
Retained samples: the cheapest insurance in the supply chain
A retained sample answers one question that nothing else can: what did we actually ship, and what did we actually receive. Without it, every dispute becomes an argument about memory.
| Parameter | Common practice | Reason |
|---|---|---|
| Quantity | Three times the amount the full test set consumes | Allows the original test, a retest, and a third-party referee |
| Retention period | Through the shelf life and six months beyond | Complaints typically arrive at the end of the shelf life, not the start |
| Storage | Controlled temperature, out of light, original sealed container | A retained sample stored in a hot warehouse is not evidence of anything |
| Identification | Product, grade, batch, date, time, sampler, container, seal number | Chain of custody fails on the label far more often than on the chemistry |
| Seal | Tamper-evident, signed across the closure | An unsealed retained sample can be challenged by either side |
The one field that is almost always missing is the sampling location. "Drum 7 of 20, top and bottom increments, 15:40" is a record. "Sample of the lot" is not, and it is the version that appears in disputes.
The eight fields of a sampling record
- Product name, grade and the internal material code used on the purchase order.
- Batch or lot number exactly as printed on the package, plus the supplier's own batch if different.
- Number of packages in the lot and number sampled, with the pattern or randomisation method.
- Sampling position within the package, and whether the result is a composite or a single increment.
- Date, time, ambient temperature and, for waxes and viscous products, the melt temperature and hold time.
- Container type, volume, headspace and seal number.
- Name of the sampler and the sampling procedure reference.
- Condition of the packages on arrival: intact, dented, stained, wet, previously opened.
None of this requires instrumentation. It requires deciding, before the drum is opened, what question the sample is supposed to answer - lot average, worst case, or chain of custody - because those three questions need three different plans, and a single composite cannot answer more than the first.
Supply
We ship the same physical forms that make sampling difficult, so we write sampling instructions into the technical data sheet rather than leaving them to be improvised at the gate. Liquids in drums and IBCs: white mineral oil 32# (8042-47-5), polysorbate 80 (9005-65-6), 1,3-butylene glycol (107-88-0), phenoxyethanol (122-99-6) and butyl oleate (142-77-8). Hygroscopic and water-sensitive solids: citric acid anhydrous (77-92-9) and xanthan gum (11138-66-2). Powders where mesh matters: zinc stearate 325 mesh (557-05-1). Flake and waxy solids: stearic acid (57-11-4), paraffin wax 60 (8002-74-2) and pentaerythritol (115-77-5).
Two related problems are worth reading alongside this one: packaging compatibility, which covers what a container does to a product over months rather than days, and bulk density and powder physical form, which covers why two powders of the same chemistry behave differently in a silo, a feeder and a sample thief. For any of the products above we can supply the batch-specific COA, the sampling position used at our plant, and a retained-sample commitment. Ask through the enquiry form on the product page.