Confirm the Datadog RUM SDK is initialized and Core Web Vitals collection is enabled; Datadog RUM automatically collects LCP, INP (replacing FID for users on supported browsers), CLS, FCP, and TTFB and attaches them as numeric attributes on view events.
Navigate to UX Monitoring > Performance Monitoring in Datadog; filter views by application and page group to see current p75 values for each Core Web Vital; apply the Google thresholds: LCP ≤ 2500ms = good, INP ≤ 200ms = good, CLS ≤ 0.1 = good.
Create RUM-based monitors for each Core Web Vital: use the RUM event query type, filtering on @type:view and the relevant metric attribute (e.g., @view.largest_contentful_paint); set the aggregation to percentile p75 and the alert threshold to the 'needs improvement' boundary.
Set the monitor evaluation window to at least 15 minutes to accumulate enough view samples for a statistically meaningful percentile; very short windows on low-traffic pages produce noisy alerts from small sample sizes.
Add a warning threshold at the 'good' boundary and a critical threshold at the 'poor' boundary for each metric to get tiered notifications; configure notify_no_data to avoid false alerts during deployment windows with low traffic.
Create a Datadog SLO for Core Web Vitals using a metric-based SLO where good events are page views with LCP ≤ 2500ms (and similarly for INP and CLS) to track the fraction of users experiencing good performance over a rolling 30-day window.
Known gotchas
INP is measured using the Event Timing API, which is only available in Chromium-based browsers; Safari and Firefox users will not contribute INP data points, so INP percentiles in Datadog reflect only the Chromium user segment.
CLS values in SPAs can be misleading if the RUM SDK is not configured for SPA routing; without SPA mode, CLS accumulates across virtual navigations rather than resetting per-view, producing artificially high values on pages with frequent in-page transitions.
Datadog RUM samples sessions, not individual views; if sessionSampleRate is set below 100, the p75 Core Web Vital values are computed from the sampled subset, which may not be representative of all users if sampling is not uniformly random across device types or geographies.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp