the error is in the scoop, not the kit

Problem statement

Aflatoxins and other mycotoxins are distributed extremely unevenly through a lot of grain or nuts — a few heavily contaminated kernels carry most of the toxin — so the concentration a laboratory reports depends less on the assay than on which kernels ended up in the sample. FAO's own sampling calculator shows that even the internationally agreed Codex sampling plans for maize and wheat "could result in total measurement error equivalent or greater than 90% of the current and proposed maximum levels" for aflatoxins in maize and ochratoxin A in wheat. Those plans were designed for large, mechanically handled trade consignments. In the bagged, mixed-origin lots that move through smallholder aggregation points and informal markets — where the disease burden is highest — there is no widely validated, affordable sampling design, so a rapid test can pass or fail the same lot depending on the scoop.

Why this matters

Sampling error is not an academic nuisance: it decides whether a contaminated lot enters the food supply (a false accept) or a clean farmer's lot is rejected and dumped onto the unregulated market (a false reject). Donnelly and colleagues' 2022 systematic review concludes that "sampling is the major source of error in the accurate assessment of aflatoxin levels in food," that "the sampling step is the most crucial step, as this is the largest contributor to error and variability," and that "there is a lack of evidence to support this or indicate the current utilization of the reviewed procedures" — meaning that even in well-resourced settings nobody knows whether recommended sampling is actually practised. In low-income countries the trade-off is sharper still: "Developing countries have a limited mycotoxin sampling budget. Hence, the risk between sampling cost and effective monitoring of AFs in commodities is difficult, posing a further challenge." Considerable investment has gone into cheaper, faster field assays; if the sample they receive carries error near the regulatory limit, that investment buys precision on the wrong step.

What’s been tried and why it hasn’t worked

Codex and national authorities publish sampling plans (numbers of incremental samples, aggregate sample mass, grinding and sub-sampling rules) derived from decades of variance studies, and FAO's online Mycotoxin Sampling Tool (2013–) lets a user see the operating-characteristic curve of any plan for 26 mycotoxin–commodity combinations without running costly trials. The Tittlemier–Whitaker analysis shows the limitation: applying the standard plans still leaves total error comparable to the maximum level, and shrinking the plan to fit a small trader's budget makes it worse. Reviewers recommend "increasing sample size and frequency, automatic dynamic sampling techniques, adequate storage, and ensuring the complete homogenization of aggregate samples," but automatic dynamic (falling-stream) samplers presuppose bulk conveyance that bagged, hand-traded lots do not have, and larger aggregate samples presuppose grinding capacity and toxin-lab access that informal markets lack. The prescription for developing countries — "an optimum AF sampling procedure ... must be cost-effective but will also produce accurate results" — has been stated as a requirement rather than delivered as a design. Meanwhile, rapid lateral-flow and reader-based kits have improved analytical precision on the sub-sample, leaving sampling as the untouched dominant term.

What would unlock progress

The unlock is a sampling design and physical toolkit sized to the smallholder lot — bag-level probing patterns, low-cost compositing and grinding, and decision rules that treat a rapid-test result as an interval rather than a number — with its operating characteristics quantified using the same variance framework FAO's tool already implements. Adjacent precedents: acceptance-sampling theory in manufacturing quality control, which routinely designs plans for small lots under cost constraints, and portable milling/mixing devices developed for on-farm seed and feed processing.

Entry points for student teams

A student team could use FAO's Mycotoxin Sampling Tool to characterize how misclassification risk changes as sample size and increments are reduced toward what a village aggregator can afford, and then design and field-test a bag-probing and compositing protocol whose measured variance beats the naive single-scoop practice. A hardware team could prototype a hand-powered grinder-splitter that produces a homogenized test portion from a 1–5 kg aggregate sample at aggregation-point cost. A decision-science team could design a sequential-testing rule (test, and if near the limit, resample) that minimizes false accepts and false rejects for a given testing budget. Relevant skills: statistics, mechanical design, agricultural extension, food chemistry.