98% browser support sounds safe. The thread did the math and it isn't

Why 98% still means 150 million people

The trigger was ordinary: the author was deciding whether to drop Sass for native CSS nesting, a feature roughly 98% of browsers now support. That "98%," he argues, is a trick of scale. Two percent of the web is on the order of 150 million people, and whether that is a rounding error or a catastrophe depends entirely on what you are counting.

To make the point he lines up mismatched yardsticks. A shop with 98% market share is dominant. A restaurant whose food only poisons you 2% of the time is making someone sick every week. A venue that turned away 2% of returning guests would be a scandal, and a babysitter who keeps a baby alive 98% of the time is a horror. Same number, opposite verdicts. His conclusion reads as a rule:

"Truly robust engineering isn't about what works for most; it's about gracefully handling the edge cases."

"It depends on the cost of failure, not the percentage"

The most common reply was that the essay smuggles in its conclusion by cherry-picking domains. Whether 98% is fine, several argued, is set by the cost of a miss, not the digits.

"A 98% success rate for a parachute is criminal, but if I could achieve 98% of my goals, I couldn't be happier." — a web engineer

A developer building the migration tool tern.sh sorted work into two piles, upside and downside:

"Most feature work is upside… Infra work is downside-based. 98% secure, 98% available, 98% acceptable performance… Winning means the thing works as expected and nobody notices." — a developer building tern.sh

Where 98% quietly compounds into a coin flip

A sharper camp accepted the premise and pushed it further than the author had: in real systems the number rarely stands alone, it multiplies.

"99% of dashboard panes load normally (so a dashboard with 20 panes is broken in nearly 1 in 5 loads)… If your LLM gets 99% of tool calls right, nearly every session will contain a malformed tool call. Probabilities are hard for humans." — a software engineer at Yandex

The most vivid version came from GPU code, where the exponent is explicit:

"…each thread does the same thing 98% of the time… you might expect the probability that any thread in the warp diverges to be .98^32, or 50% of the time… 98% isn't enough in this case." — a commenter working in GPU programming

The cleaner fix, one reader suggested, is to stop using percentages near the edges and switch to odds: "1 in 50" instead of 98% makes the jump to 99% ("1 in 100") read as what it is, a doubling of reliability.

The rebuttal the essay skips: does it break, or just degrade?

Against all of this sat the pragmatists. If your plan needs more than 98% of the market you have already lost, went one line; force the stragglers to update, went another. The strongest version pointed at what the author had quietly chosen:

"The author picked a CSS feature (nesting) that is basically all-or-nothing… Most features aren't like that; maybe the site won't look pixel-perfect on old browsers… but by and large it will still be functional." — a developer at Felt, formerly on the X-Plane flight simulator

That is the question the 98% framing hides. A percentage tells you how many users are affected; it cannot tell you whether they hit a blank page or a slightly uglier one. That distinction, not the digit, is where the real decision lives.