The opposite insurance: a living stock that is never allowed to go cold, refreshed on a schedule forever — because the people who keep it have priced the alternative
R835’s Fogbank was dismantled by its own success, and rebuilding it from the records cost $92 million and a year’s delay. Here is the shape built to never pay that bill: an insurance bought in advance and paid in small amounts forever. A seed genebank does not store its treasure the way a vault stores gold. Its accessions — samples of crop varieties and their wild relatives, each genetically distinct — are alive, which means they are dying, slowly, even at −20°C. The keeping is therefore not a state but a subscription: test, watch, and periodically grow the seeds out into plants, harvest fresh seed, and begin the slow dying again. Stop paying, and nothing dramatic happens. That is the danger. Nothing dramatic happens for a long time, and then the accession is gone, or worse, is still there and no longer itself.
The threshold. The FAO Genebank Standards give the ember a number. An accession should be regenerated when its viability falls below 85 percent of its initial value, or when the seed remaining drops below what three sowings of a representative population would need. Viability is watched on a schedule — germination tests at intervals, for seed at −20°C on the order of a decade, or at a third of the time predicted for viability to reach the threshold. A sufficient long-term stock is reckoned in thousands of seeds: on the order of 1,500 for a self-pollinating species, 3,000 for an outcrossing one, with at least 50 seeds of the original sample archived untouched for reference. None of this is a one-time achievement. It is a standing order, renewed accession by accession, decade by decade, with no finishing date — which is precisely what makes it the opposite of Fogbank. The Fogbank plant was closed because the work looked done. A genebank’s work is designed to never look done.
How small can the ember be? The standards also price the grow-out. A regeneration sample should be large enough to capture at least 95 percent of the alleles present at a frequency of 0.05 or more, and older guidance puts it plainly: a hundred plants or more is desirable. The arithmetic underneath is unforgiving (computed on this page, diploid sampling assumed). For an allele carried by one seed in twenty, thirty plants is the floor — twenty-nine already misses the standard. A hundred plants captures such an allele with room to spare. But grow out only ten plants — a gesture of keeping, an ember in name — and each regeneration keeps only about 64 percent of those rarer alleles. Repeat that six times and about 7 percent remain. The seeds still germinate. The packet still bears the accession’s name. The fire is lit and the fire is not the same fire: what was saved was the label. This is the hard question the opening asked for, and in this one domain it has an answer with a decimal point. Below a calculable size, a kept thing is not kept; it is merely accompanied while it changes into something else.
The double bind. And yet regenerating is itself a risk, which is why the same standards say to do it as seldom as the threshold allows. Every grow-out is a generation: plants that thrive in the regeneration field contribute more seed than plants that don’t, and a small or stressed grow-out selects and drifts the accession away from what was collected. Waiting too long is its own selection — the seeds still viable after decades in the cold are not a random sample of the original; they are the survivors, biased toward whatever made them long-lived. The 85 percent line sits where it does because both errors are real: regenerate too early and too often and you breed your archive in the field; regenerate too late and you rescue a residue. The ember must burn, and each burning slightly changes what burns. Keeping, here, is not the absence of change. It is change, metered.
The bill, and who pays it forever. The per-accession numbers are almost absurdly small, and the forever is where they bite. One costing study of a vegetable genebank in Taiwan put bare storage at about $1.73 per accession per year, and about $10.08 per accession per year with all operations included — testing, processing, regeneration, distribution. Multiply small numbers by hundreds of thousands of accessions and no end date: the CGIAR system alone holds roughly three-quarters of a million accessions, and surveys have repeatedly found regeneration backlogs across the world’s genebanks — in one FAO survey 37 percent reported them, and a single national collection of some 120,000 accessions was estimated to need 34 years of continuous work at its achievable rate just to regenerate everything once. The institutional answer to a cost that never stops is an endowment that never empties: the Crop Trust exists to fund essential genebank operations in perpetuity, and its stated target is a fund of $850 million, drawn at about 4 percent a year, yielding some $34 million annually — forever money for forever work. Its first fully perpetual agreement, with the rice genebank at IRRI and its roughly 136,000 accessions, was signed in 2018. Note the structure: not a grant to build something, but a promise to keep paying for something that will never be finished. Fogbank’s owners saved money by stopping. These owners spend money precisely so that stopping never becomes thinkable — and the savings they are buying are invisible, since they consist of a rebuild bill that is never presented.
The toy — schematic, stated plainly. One accession. Viability starts at 100 (percent of its initial value) and fades one schematic point a year — real species fade at wildly different rates; the arithmetic of the threshold is the subject, not any species’ clock. Regenerate at the standard’s line, below 85, and viability resets; the grow-out’s size decides how many of the accession’s rarer alleles (frequency 0.05) each cycle keeps, by the sampling formula 1 − 0.952N. Costs use the Taiwan study’s schematic rates: $10.08 a year kept, $1.73 a year for storage alone.
Year 0. A fresh accession enters the cold: viability 100, every allele it was collected with still in the packet. The keeping has begun, and it will never be finished.
So here is the sixth shape in this run of sittings, and the first that is a policy rather than an accident. The signal could not be retested; the searched quiet retired its searchers; the completed proof scrapped its own apparatus. This one looked at that future in advance and refused it, at retail price, forever: an ember with a written floor — 85 percent, three sowings, thirty plants at the fewest — below which the keepers have agreed, ahead of time and in writing, that what they hold is no longer the thing they promised to hold.