Limits on lead in toys are of two kinds. Total-content limits restrict the mass fraction of lead in the coating or the substrate. In the United States the Consumer Product Safety Improvement Act set the limit for paint and similar surface coatings at 90 parts per million and that for accessible substrate materials of children's products at 100 parts per million (United States Congress, 2008), and the model law promoted by the United Nations Environment Programme recommends the same 90 parts per million for paints generally (UNEP, 2018). Migration limits, by contrast, restrict the amount of an element released into dilute acid simulating gastric fluid. This is the approach of the European standard EN 71-3 (CEN, 2019) and of the soluble-element requirements in ASTM F963 (ASTM International, 2017). XRF measures total lead. It can therefore be compared directly with a total-content limit, and only indirectly with a migration limit.

In energy-dispersive XRF a miniature X-ray tube excites the sample and a detector records the characteristic radiation emitted by its elements (Beckhoff et al., 2006). Lead is normally quantified from its L lines, Lα at about 10.55 keV and Lβ at about 12.61 keV. The method is fast and non-destructive, and field instruments have a long record in environmental work (Kalnicky and Singhvi, 2001). Three limitations bear on painted toys. Arsenic Kα coincides with lead Lα, so the Lβ line must be used where arsenic may be present, and bromine from flame retardants gives lines in the same region. A paint film is thinner than the depth from which the signal arises, so the instrument sees coating and substrate together, and a result expressed as a mass fraction is diluted by whatever lies beneath. The substrate may itself contain lead. Finally, small, curved or irregular parts do not fill the measurement window, which lowers the count rate and the reliability of the reading. For these reasons the screening result is treated throughout as an indication to be confirmed.

The survey needs a single criterion fixed in advance. Where national law sets a total-lead limit for coatings, that limit is used; where it does not, the 90 parts per million of the model law is the natural reference. Because a reading on an intact coating is biased low by the substrate, the rule should also define a lower band within which a reading counts as inconclusive, the width of the band being set from the instrument's stated precision and from the pilot work described below.

Informal markets have no register from which to sample. The proposed approach is to build a frame at the level of the selling location: periodic markets, permanent stall clusters, streets known for toy vendors, and stalls near schools and transport hubs, enumerated by district with the help of local authorities and a walk-through. Locations are then stratified by district and type and selected at random within strata, following ordinary survey practice (Cochran, 1977). Visits are spread across days and seasons, since the goods on sale change around festivals. The result supports statements about the locations sampled, and, with caution, about the districts; it does not yield a national prevalence.

Within each location, items are purchased, not measured on the stall. Purchase avoids confrontation, allows repeat and confirmatory analysis, and keeps the instrument away from bystanders. A quota ensures coverage of categories, such as painted wooden toys, metal vehicles and coated plastic figures, and of colours, with deliberate inclusion of yellow, orange, red and green coatings, in which lead chromate pigments are most likely. Each item is photographed and given a code; the price, any labelling and any stated origin are recorded. The vendor's identity is not.

Each visually distinct colour on an item is measured separately, on the flattest area available, with the window in contact with the surface and a fixed counting time. At least two readings are taken per colour at different positions. The operator notes whether the part filled the window and whether the coating appeared thick, thin or worn. Where the instrument offers both a coatings mode reporting mass per unit area and a bulk mode reporting mass fraction, the mode is fixed for the whole survey and stated. Readings for chromium are kept alongside those for lead, because the two together point to a chromate pigment, and readings for arsenic and bromine are kept as indicators of possible interference. Spectra are saved. The instrument emits ionising radiation, and it is operated only by trained staff under the applicable local licensing rules, in a stand or shielded enclosure wherever practicable.

Quality control follows the pattern of the United States Environmental Protection Agency's field XRF method (US EPA, 2007). Each session begins and ends with an energy calibration check, a blank, and certified reference paint films or plastics spanning the action level; results are plotted on control charts and a session is rejected if they drift outside preset bounds. A proportion of items is re-measured by a second operator. Detection limits are established for the counting time used and for each class of substrate.

Readings are classified by a rule that considers both the value and the conditions under which it was obtained, as shown in Figure 1. Every item classed as positive, presumptive positive or inconclusive goes forward for confirmation, together with a random subset of the negatives so that false negatives can be detected. Confirmation means removing the coating without the substrate, digesting it in acid and determining lead by atomic spectrometry, as in the standard operating procedure of the United States Consumer Product Safety Commission (CPSC, 2011). Laboratory results, not screening readings, decide whether an item exceeds the limit.

The paired screening and laboratory results also show how well the instrument performs on this kind of sample. Agreement is examined by plotting differences against means in the manner of Bland and Altman (1986), separately for each substrate class, and by tabulating how often the screening classification matched the laboratory outcome.

For mapping, each item carries the coordinates of the market where it was bought, never those of a stall. The mapped quantity is the proportion of confirmed exceedances among items tested at a location or in a district, shown with the number tested, since a proportion based on a handful of items is uninformative. Symbols scaled to sample size, or a simple table beside the map, prevent small samples from dominating the picture. A second map can show the categories or colours involved. The full item-level data, stripped of anything that identifies a seller, are published with the report so that others can reanalyse them.