Finger millet was domesticated in the highlands of eastern Africa and reached India in antiquity (Hilu and de Wet, 1976). The grain is small, keeps exceptionally well and is notable among cereals for its calcium content, along with dietary fibre and polyphenols (National Research Council, 1996; Shobana et al., 2013; Devi et al., 2014). The plant tolerates poor soils and intermittent drought, which has led to its description as a crop suited to uncertain climates (Gupta et al., 2017). Work on minor millets in India has argued that revival requires attention to the whole chain from seed to processing and consumption, because production alone does not restore a crop that households have stopped eating (Padulosi et al., 2015).

Upland finger millet is traditionally broadcast, or transplanted at random from thick nurseries. Both give crowded, uneven stands that can be weeded only by hand. The system of rice intensification showed that young seedlings planted singly at wide, regular spacing, with soil aeration from a mechanical weeder, develop larger root systems and more tillers (Stoop et al., 2002). The same principles have been carried to other crops under the name system of crop intensification, and finger millet is among the cases described, including a farmer-developed practice from Karnataka in which seedlings are set on a square grid (Abraham et al., 2014). The agronomic logic is sound, since finger millet tillers freely when given space. The published accounts are largely field reports, and controlled comparisons on shallow upland soils are still few.

Most seed sown by smallholders comes from their own harvest or from neighbours, and such local systems are resilient until a run of bad seasons or a change of crop breaks the chain (Almekinders et al., 1994). A community seed bank formalises local exchange. Members deposit seed of chosen varieties, the bank stores it under better conditions than a household can, and seed is lent at sowing and repaid after harvest. The collection of case histories edited by Vernooy et al. (2015) shows banks serving conservation, access and local seed production in varying proportions, and shows that many fade once outside support ends. The seed security framework separates availability, access and quality as distinct problems calling for distinct responses (Sperling et al., 2008), and integrated views of the seed sector place community institutions beside, not against, the formal system (Louwaars and de Boef, 2012).

The programme begins by finding out, with the village, where finger millet stands. Transect walks, seasonal calendars and seed mapping in the tradition of farmer-first research (Chambers et al., 1989) establish which households still grow the crop, which named varieties survive and who holds them, how labour is divided between women and men, and what led others to give it up. The diagnosis decides the balance of the later phases. Where seed is the binding constraint, the bank comes first. Where seed is present but the crop is thought not worth the weeding, agronomy and processing lead.

Surviving local varieties and a small number of released ones are grown together on farmers' fields under farmers' management, and men and women rank them separately at flowering, at harvest and after cooking. This is participatory varietal selection as set out by Witcombe et al. (1996), whose central observation was that farmers given a choice adopt varieties quickly and select for traits, such as duration, straw and taste, that station trials weigh lightly. The varieties that rank well, whether old or new, are the ones the seed bank will multiply.

Volunteers plant part of a field by the new method beside a strip under their usual practice. The package is kept simple: a raised nursery, seedlings moved at about three weeks, planting along a marked rope at a spacing wide enough for a hand-pushed weeder, compost at planting, and two or three weeder passes. Spacing is not fixed by the programme. Farmers on shallow slopes may do better with closer rows than those on deeper soils, and the purpose of the strips is to let them find out. Labour days are recorded by task and by who performs them, since a method that raises yield by adding to women's workload at transplanting has not solved the problem it set out to address.

The bank is an institution before it is a building. Members agree in writing who may borrow, how much is repaid, how quality is judged and what follows default. The conditions under which shared resources are managed durably, including clear membership, rules made by users, monitoring by users and graduated sanctions, were identified by Ostrom (1990) and apply directly. Figure 1 shows the cycle through which seed passes. Technically, the task is to keep seed dry and cool. By the rules of thumb of Harrington (1972), storage life roughly doubles for each one per cent reduction in seed moisture and for each five degree Celsius reduction in temperature, a relation later given quantitative form by Ellis and Roberts (1980). Thorough sun drying, sealed containers and a germination test before lending serve the purpose without refrigeration.

Threshing and grinding by hand are among the reasons households gave up the crop, and they fall mainly on women. A small thresher and pulveriser operated by a women's group, with a user fee that covers repair, removes much of that burden. Demand is rebuilt nearest to home first: household consumption, then village institutions such as school meal and child nutrition centres where rules permit, and only then outside markets. This ordering follows the holistic value chain argument of Padulosi et al. (2015) and guards against a programme that raises output for which no buyer exists.

Once a year, after harvest, the committee and facilitator review a short list of indicators together: the number of households growing finger millet and the area, the share established by line transplanting, varieties and quantities held and lent by the bank, repayment, germination of stored seed, and hours of machine use. The review is held in the village in the local language, and its conclusions determine the next season's plan.