98% isn't much

Original Article ↗ Hacker News Discussion ↗

Context: “98%” Is Highly Domain‑Dependent

  • Many argue the key variable is cost of failure, not the raw percentage.
  • 98% can be excellent for “upside” metrics (market share, conversion lift), but unacceptable for “downside”/safety (planes, surgery, food safety, car AIs, uptime/SLA).
  • Repeated events compound: 1–2% failure per query, pane, or session means almost every user soon hits a bad case.

Browser Support, Old Devices, and Business Impact

  • Big debate over dropping support for old browsers (IE, ancient mobile, locked‑down corporate machines).
  • One camp: build for the 95–98% using modern, evergreen browsers; legacy support is costly, slows development, and users should upgrade.
  • Other camp: small percentages can represent high‑value office workers, B2B customers, or millions of government users; losing 2–6% of such traffic can be financially huge.
  • Some note real traffic still comes from IE9/11 and old Android, often during office hours. Others say very old UAs are mostly bots or edge cases not worth targeting.

Progressive Enhancement vs. All‑or‑Nothing Features

  • Many emphasize graceful degradation: use new CSS/JS for “nice to have” features, but keep core functionality working everywhere.
  • Examples: fallback images instead of AVIF/WebP‑only; simpler layouts when nesting or flexbox aren’t supported; feature detection via @supports/JS.
  • Counterpoint: some features (e.g., replacing Sass with native CSS nesting) are structurally all‑or‑nothing, making progressive enhancement hard.

Accessibility, Equity, and Ethics

  • Some see dropping old browsers as class discrimination or analogous to excluding people with disabilities; argue inclusivity is “the right thing,” especially for public services.
  • Others claim businesses can legitimately decide not to serve low‑value or extremely costly segments, just as luxury brands ignore budget shoppers.
  • Wheelchair access and digital access for essential services are cited as areas where regulation or public‑good logic should override pure profit.

Measurement, Statistics, and Perception

  • Several note that global browser stats (e.g., caniuse, analytics) may undercount privacy‑focused users and Firefox, and overcount stale Chrome UAs.
  • People recommend reframing 98% as “1 in 50 fails” or using odds/log‑odds; small changes near 100% can halve failure rates.
  • Thread repeatedly stresses that many overlapping 98% decisions can cumulatively exclude far more than 2% of users.

Skepticism of the Article’s Framing

  • Some think the article mixes unlike domains (CSS vs food poisoning) and ignores cost–benefit.
  • Others respond that its core message is modest: don’t be lulled by “98% supported” claims; check who your actual users are, what’s at stake, and whether you can degrade gracefully.