A perf budget you can actually enforce in CI
Most perf budgets don't survive contact with reality. The one that does: bundle baselines on PR + a single LCP alert from real users.
Performance budgets fail because they're either too granular (everyone ignores them) or too forgiving (they catch nothing). The version that survives: two checks, one in CI, one in production.
In CI: bundle size against a stored baseline per route. Not a flat number — a per-route delta with a 5% threshold. If a PR adds more than that to a route bundle, the PR comments with the offending modules. No approval, no merge.
In production: a single LCP alert wired to Datadog RUM, set at p75 above target for 10 minutes. That's it. Don't alert on TTFB, don't alert on CLS spikes — those have too much noise and too little signal at the team scale where you're trying to ship.
On M&M'S we ran this exact setup. The CI check caught a third-party tag library that would've added 180kb to the homepage. The LCP alert caught a CDN regression two minutes after deploy. That's the whole game.