Spent to know

the beaver is summoned by the sound of its failure — a stored seed fails in silence, looks exactly like a living one, and can be found out only by spending some of it

R810’s beavers never had to go looking for damage: the damage announced itself, in the sound of water getting away, and the makers came to the noise. Here is the keeping that gets no such summons. A seed in a vault is either alive or dead, and from outside the packet there is no difference to see. It does not rot loudly, leak, sag or call. Its failure is a fact about the inside of a grain, and it can accumulate, seed by seed, for years, in perfect quiet, in a room doing everything right.

The keepers are seed banks, and their whole discipline is organised around refusing to trust that quiet. Seeds are dried and frozen — Kew’s Millennium Seed Bank stores at −20°C, Svalbard at −18°C — because cold and dryness slow ageing almost to a stop. Almost is the load-bearing word: how long viability actually lasts under optimal storage, species by species, is still not fully known. So the collections are not simply left. At the Millennium Seed Bank, a collection’s initial viability is established about a month after it enters cold storage, and it is then retested every five or ten years, depending on whether the species is expected to be short- or long-lived. The partnership’s own standards set monitoring at least every ten years, and tie the decision to regenerate or recollect to viability falling toward 85 per cent of what it started at — you act while there is still plenty alive to act from, not when the packet is nearly spent.

And the test itself is the strange part. There is no gauge to read and no inspection that leaves the thing inspected intact: to find out whether seeds are alive, you take a sample out and germinate it. The seeds you test are spent by the testing — a seed that sprouts in the laboratory is a seed the vault no longer has. Knowledge here is not gathered by looking; it is bought, in the very currency being kept. Every monitoring round slightly shrinks the collection in order to certify the rest of it, and the schedule is a wager: test too rarely and the silent loss may cross the line where regeneration gets hard; test too often and you spend the collection proving, again and again, that it was fine.

The wager can be stretched to a century. In August 2020 a longevity experiment went into the Svalbard Global Seed Vault: seed sets of fourteen important crops, the first of them barley, pea, wheat and lettuce from the German genebank at Gatersleben. Identical samples are to be withdrawn and tested in 2030, and then every tenth year until 2120. Nobody who sealed those packets will read the last result. The experiment is, in form, exactly the discipline above slowed down until the keepers themselves have to be replaced by their successors — makers who must go and look for a failure nothing announces, on appointments made for them before they were born.

Look, at a price. A toy, with its arithmetic stated plainly: a collection of 1,000 seeds goes into the cold at a true viability of 96 per cent. In this toy the true viability falls by 4 points a year, silently — the packet’s outward state never changes, and the bar below shows only what you know, which is nothing until you test. A test germinates a sample of 25 seeds: those 25 are spent, and the test reports the true viability at that moment. Regenerating grows the remaining seeds out and banks a fresh collection of 1,000 at 96 per cent — but only if some are still alive; at zero, there is nothing left to grow. The 4-points-a-year fall is the toy’s own invention, far faster than a real vault’s, chosen so a decade of neglect fits on one page.

Last tested viability (untested = no reading)

Run it both ways. Test every year and you will watch the line cross 85 in the third year, with the collection barely dented — and you will have spent seeds to learn, twice, what one more year of patience would have told you once. Never test, and the page stays serene: same packet, same count, nothing to see — while the hidden figure walks down past the rescue line, and past the point where “regenerate” has anything left to work with. The beaver’s failure was generous: it made a noise. This one makes none, so its keepers replaced the noise with a calendar.

Three narrowings. First, the fall is neither smooth nor universal in real life: longevity varies enormously by species and by the quality of the seed lot going in, some well-stored seed is expected to last centuries, and the toy’s straight line at 4 points a year is a shape, not a measurement. Second, regeneration is not a perfect restore: the plants a late rescue can grow are only the survivors, so a collection regenerated after deep decline may come back narrowed — alive, but carrying less of the variation it was banked for — which is why the standards push action at a high threshold rather than at the edge. Third, Svalbard itself is mostly a backup safe-deposit: the vault holds duplicates for depositing genebanks, and routine viability monitoring is the depositors’ work on their own collections — the 100-year sets are a deliberate exception, packets placed specifically to be opened by appointment. Holding, even here, is not the same job as checking.