Who Checks on the Seeds?

What had gone wrong, Lindell worked out inside 10 minutes, was ordinary. The chambers are cooled by two compressors, one running and one in reserve, and the running one had tripped at 1:40 a.m. when a fan motor began to seize. The reserve should have started on its own. It had not, because a contactor—a switch the size of a matchbox—had worn out after 19 years of clicking. Lindell, who learned refrigeration on the freezer trawlers that used to work out of Norhaven, found a spare in a drawer labeled, in his own capitals, THINGS THAT WILL BREAK, fitted it with his gloves off, and at 4:20 a.m. compressor two shuddered into life. The air in Chamber A peaked at −13.8°C five minutes later. By 10 in the morning it was back at −18.^*

“Everybody wants to hear about the alarm,” said Agnes Tolland, the repository’s curator, when I visited in September. She is 59, has worked here since 1994, and talks about the collection the way a head nurse talks about a ward: fondly, briskly, and without illusions. “The alarm is the exciting bit. It’s also the bit that’s solved. We have two compressors, a drawer of spares, a man who lives 20 minutes away, and a hill that stays at five degrees even if the whole country loses power. The seeds would take weeks to warm up to anything that frightened me. The thing that keeps me awake isn’t the alarm. It’s the list.”

The list is the part of seed banking that never appears in photographs. Every one of the repository’s own samples—it holds about 162,000, alongside some 238,000 sealed safety copies kept on behalf of other genebanks—is supposed to be brought out once a decade and tested to see whether its seeds are still alive. Those that are failing must be planted, grown, harvested, dried, and sent back up the hill as fresh seed. All of that falls to a crew of five: Tolland; Lindell; a germination technician, Josie Amadi; a records manager, Henrik Saul; and Moira Kade, who runs the regeneration fields on the valley floor. Until 2019 there were six. The list has been getting longer ever since.

A slow chemistry

A seed is a plant that has agreed to wait. Inside the coat is an embryo—a root, a shoot, a leaf or two, folded small—and a store of food to start it off, and for most of the crops people eat, the whole package has been shaped by evolution to dry out and sit tight until conditions improve. Seed scientists call these orthodox seeds, and nearly everything in the Norhaven vault is one.2 Dry them and chill them, and they will wait for a very long time. Not forever, though, and not all at the same pace.

A seed ages the way a cut apple browns: through slow chemistry. Reactive molecules, most of them involving oxygen, nick away at the fats in its cell membranes, at its proteins, at its DNA. Water speeds all of this up, because water lets molecules move and meet. Dry a seed far enough, though, and something odd happens inside it. The syrupy contents of its cells set into a glass, much as boiled sugar sets into hard candy. In that glassy state molecules can barely move at all, and the chemistry of aging slows almost to a halt. Cold slows it further still.

Seed physiologists of the mid-20th century boiled this down into two rules of thumb that every genebank worker can recite. Within the ranges where they hold, each one-point drop in a seed’s moisture content roughly doubles its storage life, and so does each 5°C drop in temperature. At Norhaven, incoming seed is dried for weeks in a room kept at 15°C and 15 percent relative humidity, until a barley grain that came off the field at around 15 percent water is down to about 5 percent.3 Then it is sealed in foil and carried up the hill. Whether the temperature rule holds all the way down to −18°C is still argued over. The honest version, Tolland says, is that cold always helps, just not quite as much as the arithmetic promises.

The second thing to understand is the shape of a seed’s decline. A packet of seed does not fade like an old photograph, a little every year. Individual seeds die at different moments, but their deaths bunch together, so a sample can test in the 90s for decades and then collapse within a few years. Seed scientists draw it as a reverse S: a long, flat shelf, a knee, and a cliff. The entire purpose of a testing program is to catch each sample on the shelf, or at worst at the knee, and grow it out before it reaches the edge. Josie Amadi, who does the testing, puts it more simply: “It holds, and holds, and then it goes.”

And the shelf is not the same length for every crop. Barley and wheat are long keepers; the repository’s barley samples that went in during the early 1980s still average 86 percent germination. Onions, lettuce, and carrots are notoriously short-lived, for reasons that are still not fully understood. Norhaven’s onion records, pooled across every sample it has tested since 1979, trace the curve almost too neatly: 93 percent when the seed goes in, 88 percent after 20 years in the vault, 67 percent after 35, and about a third after 45.^†

Fifty seeds on wet paper

Each working day begins in the cold. Saul pulls on a down parka that hangs by the airlock, takes a printed pick list and a plastic crate, and goes through two doors into Chamber A, where the air bites the inside of your nose and every metal edge wears a fur of frost. He allows himself 15 minutes at a time. The boxes are white card, labeled by accession number, and the packets inside are about the size of a postcard, heat-sealed, with a barcode and the same number written beside it in marker, “because barcodes are a fashion,” he says. He carries out the day’s samples and leaves them sealed on a bench for several hours, so that the cold seed warms slowly and no moisture condenses on it. Only then do they go to the lab.

The germination lab is a windowless room off the tunnel mouth, warm enough after the vault to fog your glasses, and it smells of damp paper. Along one wall stand four germination cabinets, glass-fronted and lit, set to 20°C. Inside, shelf after shelf, are clear plastic boxes lined with blue blotting paper, each holding 50 seeds laid out in a grid with tweezers. A test is two boxes, 100 seeds. Amadi, who is 33 and came to Norhaven from a plant-pathology job in Halbridge, can set up about 45 tests on a good day. She counts them on day six and day 12, or on whatever schedule a species demands, and she counts by eye, with a tally counter, the way it was done in 1979.

A seedling counts only if it is normal: a root, a shoot, the right proportions, nothing stunted or rotten. The rule matters, because age shows up first not as death but as weakness. Old seed germinates late, and its seedlings come up spindly and slow, as if they had to spend their first days mending damage before they could get on with growing, which in a sense they do. Amadi has learned to read the early signs. “When a sample that used to be finished by day four is still dragging on day nine, I write a note,” she says. Seeds that do nothing at all get pressed with forceps. Firm ones may only be dormant, and those are cut open and soaked in tetrazolium, a colorless salt solution that living tissue turns red.4 Soft gray ones are dead.

Each test is also a small withdrawal from the account it is meant to protect. A sample might start with 3,000 seeds; a hundred go every decade, more if the result is borderline, more again when a plant breeder or a researcher asks for some. Test too often and the seed is used up. Test too rarely and the knee goes by unseen. The repository settled on 10 years in 1979, and it was a sensible interval when the collection was small.

It is not a sensible interval now. To test 162,000 samples once a decade means about 16,000 tests a year. With six people, the lab came close. Since the second lab post was cut in 2019, after a spending review by the agriculture ministry in Corran, it has averaged about 11,000.^‡ On the morning I visited, the whiteboard by Amadi’s door read OVERDUE: 38,712. Henrik Saul, the records manager, updates it every Friday afternoon. “It used to be a joke,” he said. “Then it went past 20,000 and stopped being one.”

There is triage. Short-lived crops jump the queue; long-keeping cereals wait. But the system that tracks due dates is itself a thing that needs looking after. In 2021 the repository moved its records from a database built in the 1990s to a new one, and in the move about 2,300 samples lost the field that said when they were next due. Nobody noticed for three years.5 Most of the 2,300 were barley and oats, which forgave the delay. A few were onions.

The onion from Brannock

Accession NSR-1982-0417 arrived in two twists of brown paper. It had been collected in August 1981 by Cora Fenwick, the repository’s first curator, from a croft at Brannock on the north coast, where a woman named Elspeth Rowe, then 77, grew a small, hard, purplish onion that her family had been saving seed from for as long as anyone could remember. Fenwick typed up her collecting card that evening in a rented room. It is still in a filing cabinet at the repository, along with the pencil note she added the following spring.

Rowe’s seed tested at 96 percent when it went into the vault in 1982. It tested at 95 in 1992 and at 92 in 2002; in between, a breeder asked for some and was sent 250. In 2012 it tested at 89, comfortably above the repository’s trigger for regrowing a sample, which is a fall to 85 percent of its starting figure. Its next test was due in 2022. That was the year after the database move. Brannock red was one of the 2,300.

Amadi set up its test in February 2026. On day six there were three seedlings. On day 12 there were nine. She ran a second hundred to be sure and got 11. When she pressed the seeds that had done nothing, nearly all of them gave under the forceps like wet bread. Somewhere in the 14 years since anyone had last looked, the onion had gone over the cliff. “You can’t see it in the packet,” she told me. “They look exactly the same as they did. That’s the cruel bit. They still look like seeds.”

Fenwick’s note on the card was not a figure of speech. The repository sends safety copies of most of its holdings to genebanks abroad, but a sample of a few thousand seeds is too small to split, and about 38,000 of Norhaven’s own samples have no duplicate anywhere. Brannock red is one of them. Haugh Croft has been a vacation rental since 2003, and no one in Rowe’s family still grows the onion. What was in those packets was, as far as anyone knows, all of it.

Flies from the bait shop

The regeneration fields lie at the foot of the Knock: three hectares of leased farmland laid out in strips, with two long plastic hoop houses and rows of white mesh cages that look, from the road, like a campsite for very tidy ghosts. Moira Kade has run them since 1996. She is 63, wears a fleece the color of wet moss, and regards the vault up the hill with the mild condescension of someone who works with living things. “Up there they keep seeds,” she says. “Down here we keep plants. You can’t have one without the other, though people keep trying.”

Regrowing a sample is nothing like copying a file. Crops that pollinate themselves, such as barley, wheat, and peas, can be grown in neighboring plots with little risk of mixing. Crops that cross freely—rye, cabbages, carrots, onions—have to be grown in isolation, under mesh, with pollinators shut in alongside, or the next generation of a 19th-century kale will carry a dash of the neighbor’s cauliflower. And enough plants must flower to carry a fair share of the original’s genetic variety into the new seed. Kade’s fields regrow about 420 samples a season. At that rate it would take roughly 385 years to grow out the whole collection once, which is why nothing is grown out until a test says it must be.

Onions are among the most awkward crops she handles. They are biennial: the seed makes a bulb in its first summer, the bulb has to be lifted and chilled through the winter, and only in its second summer does it send up the tall stalk and round white flowerhead that make seed. The flowers need insects, and inside a mesh cage the most dependable insect is the blowfly. Every June, Kade buys several pints of maggots from the bait-and-tackle shop on Norhaven’s harbor, lets them pupate in a shed, and releases the flies into the onion cages.6 “They don’t sting, they don’t mind rain, and they’ll visit anything,” she says. “Bees have opinions.”

In April, Kade sowed every remaining seed of Brannock red, all 1,950 of them, in trays in the hoop house. Of those, 162 came up. By September, when she lifted them, 141 had made bulbs, small and hard and purple, and they are now in the cold store at the bottom of the hill. In April 2027 they will be planted out in a cage of their own. If the summer is kind, there will be seed by September.

Whether what comes out of that cage will still be Brannock red is a question Kade raises before I can. Rebuilding a variety from 141 plants is a squeeze, and the survivors are not a random handful. They are, by definition, the seeds that were best at not dying in a freezer, and whatever genes helped them do it will now be overrepresented in everything that follows. Seed scientists worry about exactly this, and the repository’s own rules ask for at least 100 flowering plants for a crop like onion, a bar Brannock red should just clear.^§ But Kade has read the collecting card, too. “Mrs. Rowe’s family kept that onion going for a century on 12 bulbs a year,” she says. “If 12 was enough for them, I’m not going to lose sleep over 141.”

Longer between visits

It is tempting to think of a seed vault as a place where time stops. The engineering invites it: the steel door, the rock, the cold. A repository brochure printed in 2015 calls the collection “safe for centuries,” a phrase that makes Tolland wince. Time does not stop at −18°C. It slows, by a factor that varies from crop to crop and that nobody can know for any particular sample until somebody looks. A vault does not replace the people who once kept a variety alive by planting it every spring. It only lets them come back less often.

Tolland’s version is blunter: “We’re not a library. A library doesn’t have to keep reprinting its books.” The comparison she prefers is a lighthouse, or a bridge—something valuable precisely because someone keeps turning up to look after it, and which looks most finished when it is most neglected. Last year the repository asked the ministry to restore the sixth post. It was told the request would be considered at the next spending review, in 2027. By then, at the current rate, the number on Saul’s whiteboard will be well past 45,000.

Lindell, for his part, replaced the contactor on compressor one as well after the January alarm, and ordered two more for the drawer. He has also started doing something he did not do before. Once a month, at an hour he tells no one, he switches off the running compressor himself and waits to see whether the other one starts. So far it always has. “The alarm isn’t the system,” he says. “I’m the system. The alarm just knows my number.”

On my last afternoon in Norhaven, Kade took me into the cold store to see the bulbs. They sat in two orange net sacks on a slatted shelf, each about the size of a walnut, their skins the dusky purple Fenwick typed onto her card 45 years ago. Kade picked one out and turned it over in her fingers, the way people handle a stone from a beach. “It’s only ever been people,” she said. “Somebody planting them, somebody saving the seed, somebody remembering to do it again next year. The freezer just buys you longer between visits.”

If the flies do their work next summer, the cage will yield tens of thousands of seeds, more Brannock red than Elspeth Rowe ever harvested in a single year. Saul will log them, the fifteen-fifteen room will dry them, Amadi will test them, and they will go up the hill in foil, into the dark, at −18°C. A new due date will go into the database—2033, given what everyone now knows about onions—and this time, Saul says, onto a sheet of paper taped inside the door of Chamber B as well. Somebody will have to be there to read it.