The arsenic is geogenic. It is carried into the basin on iron oxyhydroxide coatings of Himalayan sediment and released to groundwater when those coatings dissolve under the reducing conditions that develop in young, organic-rich alluvium (Nickson et al., 1998; Smedley and Kinniburgh, 2002). High concentrations are typical of shallow wells in Holocene deposits, the depth range tapped by most private hand pumps, while older and deeper sediments are more often low in arsenic. Within the affected depth range the distribution is patchy at the scale of tens to hundreds of metres, and it cannot be predicted for an individual well without testing it (Fendorf et al., 2010).

Arsenic in drinking water is classified as carcinogenic to humans, with sufficient evidence for cancers of the skin, lung and urinary bladder (IARC, 2012). Long-term exposure also produces characteristic skin lesions and is associated with cardiovascular, respiratory and developmental effects (Naujokas et al., 2013). The World Health Organization's provisional guideline value is 10 micrograms per litre (WHO, 2017), and the Indian drinking water standard sets its acceptable limit at the same level (Bureau of Indian Standards, 2012). Effects appear after years of exposure, and the risk at concentrations near the guideline value is less certain than at the high concentrations where most of the evidence was gathered.

Water is the main route of exposure but not the only one. Rice irrigated with contaminated groundwater and cooked in it contributes as well (Meharg and Rahman, 2003). Individual exposure can be assessed through biomarkers: arsenic in urine reflects intake over the preceding days, while hair and nails integrate over months, each with known limitations (Hughes, 2006).

Experience in Bangladesh shows that information changes behaviour. When households were told the arsenic status of their well, a majority of those with unsafe wells changed to another source within a year (Madajewicz et al., 2007). Switching to a nearby safe well and installing deeper wells have been the most widely adopted responses, ahead of treatment technologies (Flanagan et al., 2012). Alternatives can introduce new hazards: a comparative risk assessment found that options such as dug wells and pond sand filters may replace arsenic risk with microbial risk (Howard et al., 2006). Why households adopt or reject an option depends on perceived risk, cost, social norms, confidence and habit, the factors organised in the behaviour change model of Mosler (2012).

A suitable village lies on the Holocene floodplain in a district where arsenic has been documented, relies mainly on private hand-pumped wells, and is small enough for every well and household to be enumerated. A single case supports analytic generalisation, meaning that it tests and refines explanations, and not statistical generalisation to other villages (Yin, 2018). The choice should therefore be made and reported for stated reasons. A baseline description records the layout of hamlets, the social and economic composition of the population, landholding, existing water infrastructure and any previous testing or mitigation programme.

Before fieldwork the protocol is submitted to an ethics committee constituted under the national guidelines (ICMR, 2017). Three matters need particular care. Consent is sought in the local language from each well owner and each household, with separate consent for any biological sample. The study commits in advance to telling every owner the result for their well, since withholding a result that bears on health cannot be justified. And the team agrees with local health services how people with suspected arsenic-related illness will be referred. Marking wells and discussing skin lesions can stigmatise families, so the way results are made public is settled with the community at the start.

Every well is located by satellite positioning and recorded with its owner, reported depth, year of installation and the households that use it. Each is tested on site with a field kit of the kind that converts arsenic to arsine gas and reads the stain it leaves on a test strip. Such kits are semi-quantitative and least reliable near the thresholds that matter, so a subset of wells, chosen to span the range of kit readings, is sampled in duplicate, acidified and sent for laboratory analysis by hydride generation atomic absorption spectrometry or inductively coupled plasma mass spectrometry. Field blanks, duplicates and a certified reference material accompany each batch. Results are classified against the guideline value (WHO, 2017). Testing is repeated in a second season, because water tables on the floodplain rise and fall with the monsoon.

A structured questionnaire, administered to an adult who manages water in each household, records all sources used for drinking and for cooking in each season, how long the present well has been used, wells used previously, who fetches water, and what the household already knows about the status of its well. Rice consumption and the source of cooking water are included because of the dietary route. Linking each household to the tested wells it uses gives an exposure classification for every household. Where laboratory capacity and ethical approval allow, urine or nail samples from consenting adults add an individual measure; this component is optional and should not be attempted without a laboratory able to analyse the samples properly.

Semi-structured interviews explore what households have done about their water and why. Interviewees are chosen purposively to cover households with high and low arsenic, different strategies, different economic positions and both women and men, since women usually carry the water and men often control spending on wells. Key informants such as well drillers, health workers and elected village representatives are interviewed as well. Table 1 lists the strategies the interview guide should cover and what to record for each. Recordings are transcribed and analysed thematically following the phases set out by Braun and Clarke (2006), with the factors in the model of Mosler (2012) used as sensitising concepts and not as a fixed coding frame.

The quantitative strand yields the distribution of arsenic across wells, maps of its spatial pattern, and the proportion of households whose main drinking source exceeds the guideline value. The qualitative strand yields an account of strategies and the reasoning behind them. Integration takes place in a joint display: a table that sets each household's exposure class beside its strategy and its stated reasons, so that patterns and exceptions can be read across (Creswell and Plano Clark, 2018). Households that remain on a contaminated well despite knowing its status deserve the closest reading, since they show where mitigation fails.

Each owner receives the result for their well in writing and in person, with an explanation of what it means and what the options are. A village meeting presents the overall picture without identifying households. With the community's agreement, wells are marked as safe or unsafe for drinking, and the anonymised dataset is shared with the district authority responsible for rural water supply. A follow-up round a year later, repeating the survey of sources, would show whether disclosure changed behaviour.