Performance

Raw readings in the library and the CLI; budgets in journeys, so transcripts stay stable.

var t = await page.TimingsAsync();
foreach (var (name, value) in t.Named)
    Console.WriteLine($"{name}: {value}");

ttfb, first-paint, fcp, dom-interactive, dom-content-loaded and load, in milliseconds from the start of the navigation, from the page's navigation and paint entries. Chrome reports a paint once its frame is rendered, which on a machine without a GPU can come after load; TimingsAsync waits for it only when the page has loaded and reported no paint yet.

await page.RecordInteractionsAsync();
await page.GotoAsync(url);
// … interact …
var v = await page.VitalsAsync();
Console.WriteLine($"LCP {v.LargestContentfulPaint} ms on {v.LargestContentfulPaintElement}, CLS {v.CumulativeLayoutShift}, INP {v.InteractionToNextPaint}");

1. before the page loads, for INP

Chrome keeps only interactions of about 104 ms or more. RecordInteractionsAsync adds a small script before any page script that records them from 16 ms, as Google's web-vitals library does. It is opt-in; a journey turns it on only when it uses vitals. Vitals wait two frames for entries to arrive, so read them a moment after the change you want measured.

var metrics = await page.MetricsAsync();

1. Nodes, JSEventListeners, LayoutCount, ScriptDuration, JSHeapUsedSize, …

Chrome's Performance.getMetrics, as Chrome names and scales them.

timings
vitals --budget lcp=2500 cls=0.1 inp=200
metrics --budget Nodes=1500

1. raw: changes every run

With budgets the checkpoint prints only lcp ok (<= 2500) for each one, and a reading over its limit, or with no value yet, fails the step with the actual value. agent-browser has vitals but not timings or metrics; the export keeps vitals without budgets.