Where Do Safety Limits Come From? Mostly From a Whiteboard, Sometimes From a Lawsuit

The Magic Number Nobody Measured

Somewhere in a federal database, there's a number that decides how much arsenic is allowed in your water. Another one governs the aflatoxin in your peanut butter. A third sets the ceiling on formaldehyde in your furniture. These numbers feel authoritative. They're printed in the Code of Federal Regulations. Inspectors enforce them. Companies get fined over them.

But if you trace most of them back to their origin, you don't find a definitive experiment. You find a committee, a deadline, a lawsuit, or — and this is the part that should make you spit out your coffee — a number another agency already used, borrowed for convenience because nobody wanted to start from scratch.

American safety thresholds are a museum of educated guesses. And the interesting question isn't whether the guesses were bad. Some were remarkably good. The question is what it tells us about how much of the quantitative world around us was built by people doing their best with the data they had, which was often almost none.

One in a Million: The Number That Came From a Philosophy Paper

Start with the most famous threshold in American regulation: the "one-in-a-million" lifetime cancer risk. If a chemical in your drinking water raises your lifetime cancer risk by less than one in a million, regulators generally shrug. Above that, the paperwork begins.

Where does a million come from? Not from an epidemiologist counting tumors. It came from a kind of moral reasoning. In the 1970s and 80s, as regulators tried to figure out how much risk was "acceptable," they needed a number small enough that the public would consider it negligible. One in a million had a nice property: it's about the risk of being struck by lightning, roughly. It felt like background noise.

It's a defensible choice. But it's a choice, not a measurement. A different era might have landed on one in a hundred thousand and built an entirely different regulatory state around it. The threshold isn't a fact about arsenic. It's a fact about how much risk a democracy decided to tolerate — a value dressed up in a lab coat.

The Five-Second Rule, Bureaucratically

Then there are the numbers that are even less scientific: the rules of thumb. Food safety is full of them. Cook poultry to 165°F. Keep hot food hot and cold food cold. Time-temperature combinations that inspectors recite like scripture.

Some of these have real microbiology underneath. Many are compromises between what would actually kill the pathogen and what a kitchen could realistically do without ruining the food. The 165°F standard for poultry, for instance, is more conservative than strictly necessary — lower temperatures held for longer achieve the same bacterial reduction. The FDA picked a number that works instantly because instant is the only thing a busy kitchen will actually do.

That's the hidden variable in every threshold: compliance. A perfect limit nobody follows is worse than an imperfect limit everyone does. Regulators aren't just modeling bacteria. They're modeling human laziness, and honestly, they're right to.

When the Number Was Written by the Lobby

We'd be lying if we said every threshold was a good-faith guess. Some were written by the industries they were supposed to constrain.

Occupational exposure limits are the classic example. For decades, OSHA's permissible exposure limits for hundreds of chemicals were simply adopted from lists drawn up in the 1960s — some based on studies with sample sizes you'd reject from a middle school science fair. A handful of those limits still stand today, protecting workers against hazards we can measure far better now, using numbers nobody has revisited because revisiting is expensive and litigation is expensive-er.

The pattern repeats across agencies: a threshold gets set once, under time pressure, with whatever evidence was on hand. Then it ossifies. Updating a federal limit can take a decade, cost millions in rulemaking, and survive exactly one change of administration. So the number from 1971 stays on the books, not because it's right, but because changing it is harder than keeping it.

What a Data-Driven Limit Would Look Like

Here's where the sample-size people get to have their day. A threshold built the way we'd build it now would look nothing like the ones we inherited.

It would come with a confidence interval, not just a point estimate — a range, with the uncertainty printed right on the label. It would be versioned, with a scheduled review date the way software gets security patches. It would draw from continuous monitoring rather than a single study: real-time sensor networks measuring air and water in thousands of locations, feeding models that update as the data flows in. We have the sensors. They're cheap now. What we don't have is the institutional habit of letting the threshold move when the evidence does.

Most importantly, it would separate the two questions we currently smush together: how much harm does this cause, and how much harm will we accept? The first is a science question with a real answer. The second is a political one, and pretending it's science is how you end up with numbers that are simultaneously too vague to defend and too rigid to fix.

Respect for the Guesswork

None of this is an argument that the numbers are fake or that regulation is a sham. Run the counterfactual: an America with no thresholds, where every chemical is innocent until your funeral. The guesswork saved lives. The clean water acts, the food inspections, the air quality standards — built on wobbly foundations, and still among the best public health interventions of the twentieth century.

But it is an argument for humility about what a regulation is. When the inspector shuts down the kitchen or the label warns about the chemical, you're not witnessing pure science in action. You're witnessing a decision — made decades ago, by people with less data than your phone generates in a week, balancing evidence against feasibility against politics — that we've all agreed to stop questioning.

The next generation of thresholds doesn't need to be guessed. It needs to be measured, monitored, and revised. The data's already out there. It's the paperwork that's stuck in 1971.