Performance Pulse
Scheduled re-scans that only interrupt you when a metric has genuinely moved.
What Pulse watches
Each due site gets a fresh CrUX and PageSpeed pull. The new field value for LCP, CLS and INP is compared against the immediately preceding stored value for that metric — not a rolling average, so a single bad deploy cannot hide inside a smoothed line.
How an alert is decided
Two gates sit between a change and your inbox:
- 1
The noise floor. Any movement under 0.5% is ignored outright. Field data wobbles; that is not news.
- 2
Your threshold. A worsening change must also meet the site’s alert threshold — 10% by default, adjustable per site.
The delta is computed as (current − previous) / previous, so a positive number is a
regression. Improvements are recorded too, but they never raise an alert.
Every qualifying change is logged either way: regressions get a root-cause analysis with the top Lighthouse savings opportunities attached, improvements get a note without one.
Choosing a threshold
| Setting | Good for | Trade-off |
|---|---|---|
| 5% | A checkout or pricing page where small regressions cost money | More alerts, some of them noise on low-traffic pages |
| 10% (default) | Most pages | Balanced — catches real regressions, ignores wobble |
| 20–25% | Low-traffic pages, or ones you know are noisy | Slow drift can pass unnoticed — review the diagnostics weekly |
If a site alerts more than about once a fortnight without anything having shipped, its threshold is too low for its traffic volume rather than its performance being unstable.
Email, and silence by default
Alerts only go out when a notification address is configured and passes validation. With none set, Pulse keeps monitoring and logging and says nothing — it will not guess at where to write. If you expected an alert and got none, check the address on the site before assuming nothing regressed.
What it costs you
A scheduled re-check spends one scan from your monthly allowance, exactly like one you run by hand, and each device profile counts separately. If an account reaches its limit, its scheduled sites pause and resume next cycle rather than failing — nothing already saved is touched.
Monitoring ten sites on both devices is roughly 600 scans a month. Worth checking against your plan before turning Pulse on across the board.
Getting good results from it
- Monitor your highest-traffic templates, not your most interesting pages. One product page stands in for ten thousand.
- Turn Pulse on before a risky deploy, so the regression has a baseline to be measured against.
- Read the diagnostics weekly even when nothing alerted — slow drift never trips a threshold.
- Keep release notes next to it. Half of root-cause analysis is knowing what shipped on Tuesday.
Scheduled scans use your monthly allowance
A Pulse re-check costs the same as a scan you run by hand. If an account runs out, its scheduled sites pause and resume next cycle rather than failing.