The Half-Life of Ninety-Five Percent

In late 2020, two large trials reported vaccine efficacy against symptomatic COVID-19 at around ninety-five percent, and the number entered the culture as a property — a fact about a vial, printed on the side like a proof rating on a bottle of whiskey. It was not that. It was a measurement of the relationship between two populations, both of them moving, taken at one moment, against one version of a virus, under one set of human behaviors, with confidence intervals attached that nobody read aloud on television.

This week the photograph aged.

Researchers at Public Health England reported in the New England Journal of Medicine what happens when you point the same vaccines at B.1.617.2 — Delta. Two doses of the Pfizer vaccine: 93.7 percent effective against symptomatic disease with Alpha, 88.0 percent with Delta. Two doses of AstraZeneca: 74.5 percent against Alpha, 67.0 percent against Delta. A single dose against Delta: 30.7 percent, down from 48.7. The two-dose numbers are modest slippage. The one-dose number is a cliff.

Here is the part worth sitting with — carefully, because two different comparisons are in play and only one of them is clean.

The clean one is inside the paper. 93.7 against Alpha, 88.0 against Delta: same study, same design, same population, same window. Nothing about the vaccine changed between those two numbers. Not one molecule was altered. What changed was the other half of the relationship — a virus with a mutation rate, running trillions of replications a day across a planetary substrate of human bodies, doing arithmetic nobody authorized.

The messier comparison is the one in the headlines, and in the title of this piece: ninety-five down to eighty-eight. Those numbers do not come from the same world. Ninety-five was efficacy — randomized, tightly controlled, short follow-up, largely against the ancestral strain. Eighty-eight is effectiveness — observational, test-negative, a different population behaving differently in a different season, and, the part nobody enjoys saying out loud, considerably more time elapsed since the second dose. At least three things moved between those numbers and the virus is only one of them. Waning immunity is another. The measuring instrument is a third.

Which illustrates the point better than the clean version would have. Effectiveness is not stored in the syringe. It lives in the gap between many moving things, and the ruler is moving too.

The number moved, and a great many people experienced that as betrayal.

That is the genuinely strange thing here — stranger than the biology. The biology is doing precisely what biology does; evolution is not a scandal, it is Tuesday. The strange thing is that a functioning epistemic system, publicly revising a number in response to new evidence within nine months, was widely received as proof that it had been lying all along. The revision was the system working. It got read as the system failing. We had been handed what looked like a stone and were furious to discover it was a river.

It is tempting to file that under human nature, and human nature is genuinely involved — but reaching for the cognitive quirk first lets the more legible cause walk quietly out of the room.

The error bars were always in the papers. They were rarely in the sentence. That was not a failure of translation; it was the designed output of a channel optimizing for compliance rather than comprehension. An apparatus that measures its success in uptake rates has every incentive to hand over a provisional figure as a carved one, because certainty moves people and confidence intervals do not. It is a loan against a future you have to repay with interest — because when the revision comes, as it always does, you have already trained your audience to read movement as lying. The betrayal reaction was built in. Not on purpose, but as reliably as if it had been. You cannot spend a year insisting a number is a stone and then expect applause for admitting it is a river.

That is an environment problem rather than a character flaw, which is the more useful diagnosis, because environments can be rebuilt. Scolding a population for wanting certainty is a much worse strategy than not handing it out.

And underneath the incentives, the older thing is real too. We want reality to hold still long enough to build on top of — and it mostly does, for the eight or nine decades any of us is around to check, which lets us forget that every constant we lean on is a measurement with a shelf life. The kilogram was a lump of platinum-iridium in a vault outside Paris until 2019, when it became clear the lump was quietly drifting and the unit was redefined in terms of Planck's constant instead. Note what that story actually is: not knowledge dissolving but knowledge relocating to a better footing — the revision made the kilogram more stable, which is exactly what a working system looks like from outside. Your body, meanwhile, swaps out most of its atoms on a rolling schedule; the thing you answer to when someone says your name is a pattern maintaining itself against turnover, not an object sitting still.

You are already a river. You simply don't get the update emails.

Provisional knowledge is not a weaker grade of knowledge. It is the only grade there is. The alternative — a claim structurally incapable of revision — is not sturdier; it has merely stopped being connected to anything. A number that never moves has stopped listening.

The sharpest demonstration of that sits in the same table as the numbers above. Those single-dose figures, 48.7 and 30.7, did not land on an accidental population. Britain had one-dosed millions of people on purpose the previous winter, when its vaccine committee chose to stretch the gap between shots to as much as twelve weeks in order to get a first dose into as many arms as possible. That was a bet placed on a provisional number, under real uncertainty, for defensible reasons — and then the target moved, and the bet was called by a variant that had not existed when it was made. This is what acting on provisional knowledge looks like from the inside: not paralysis, not false certainty, but a wager you must be willing to be seen revising. Making the bet was never the failure. Describing it as a sure thing was.

And eighty-eight percent, incidentally, is extraordinary. Two years ago this virus had a zero attached to it, because it had not been described. We built something that works on roughly nine out of ten encounters with a pathogen that was not in the literature when most of the researchers involved started their careers — and then felt let down that it wasn't also a permanent constant of nature.

The universe does not issue permanent constants. It issues rates, gradients, and patterns that hold for a while, which is a category that includes stars, species, and you. That is not a downgrade. That is just what's here.

Get the second dose. The cliff is real and the river is fine — and if that needs revising in a year, you'll have learned exactly the right thing about how any of this works.