Performance intelligence
Your Site Got 2x Slower. Your Uptime Tool Never Noticed.
Your client's site responded in 400ms for months. After last week's deployment, it's consistently at 900ms. It's not "slow" by any absolute standard — but it's more than twice as slow as its own baseline. Sitewatch maintains a rolling p95 response time baseline for every monitored site and alerts when performance regresses beyond its own normal. No manual thresholds. No tuning.
- Rolling p95 baseline maintained automatically per site
- Alerts when response time exceeds 2x baseline for 3 consecutive checks
- 3-check confirmation eliminates false positives from transient blips
Performance regression detected
Current vs baseline
- PERFORMANCE_REGRESSION: response 2.1x above p95 baseline9:15 AM
- Baseline p95: 380ms — current: 810ms9:15 AM
- Confirmed across 3 consecutive checks9:15 AM
- All assets OK — isolated server regression9:15 AM
01The detection gap
Absolute thresholds vs rolling baseline regression
| Scenario | Absolute threshold | Sitewatch baseline |
|---|---|---|
| Fast site (200ms normal) slows to 600ms | Not detected — under 2s threshold | PERFORMANCE_REGRESSION — 3x baseline |
| Slow site (1200ms normal) spikes to 2100ms | Not detected — under 3s threshold | PERFORMANCE_REGRESSION — 1.75x baseline |
| Transient blip (single slow check) | May alert (false positive) | Ignored — 3 consecutive checks required |
| Sustained regression after deploy | Only if absolute threshold exceeded | Always detected — relative to site's own normal |
| Threshold maintenance | Manual — requires tuning per site | Automatic — self-updating per site |
| Per-site calibration | Same threshold for all sites | Unique p95 baseline per monitored site |
Fast site (200ms normal) slows to 600ms
- Absolute threshold:
- Not detected — under 2s threshold
- Sitewatch baseline:
- PERFORMANCE_REGRESSION — 3x baseline
Slow site (1200ms normal) spikes to 2100ms
- Absolute threshold:
- Not detected — under 3s threshold
- Sitewatch baseline:
- PERFORMANCE_REGRESSION — 1.75x baseline
Transient blip (single slow check)
- Absolute threshold:
- May alert (false positive)
- Sitewatch baseline:
- Ignored — 3 consecutive checks required
Sustained regression after deploy
- Absolute threshold:
- Only if absolute threshold exceeded
- Sitewatch baseline:
- Always detected — relative to site's own normal
Threshold maintenance
- Absolute threshold:
- Manual — requires tuning per site
- Sitewatch baseline:
- Automatic — self-updating per site
Per-site calibration
- Absolute threshold:
- Same threshold for all sites
- Sitewatch baseline:
- Unique p95 baseline per monitored site
02Why it matters
Absolute thresholds are crude. Your site deserves better.
Relative detection, not arbitrary limits
A site that normally responds in 200ms and starts responding in 600ms is degraded — even though a 2-second threshold wouldn't catch it. Sitewatch detects a 2x deviation from the site's own rolling p95 baseline, regardless of absolute numbers.
Catches deployment-caused regressions
An unoptimised database query, a missing cache header, a heavier JavaScript bundle — these are the changes that slip through QA because the site "still works." Sitewatch catches the performance cost of every deploy.
3-check confirmation, no noise
A single slow check could be network variance or a cold start. Sitewatch requires 3 consecutive checks above 2x baseline before creating a high-severity incident. Only sustained regressions fire alerts.
Automatically maintained
No threshold configuration required. Sitewatch builds each site's p95 baseline from its own check history and updates it continuously. Zero setup — monitoring starts from the first check.
Correlate with page weight
Did the site get heavier AND slower at the same time? Sitewatch tracks both page weight and response time baselines on every check — so you can confirm whether a deploy shipped unoptimised assets alongside a performance regression.
High-severity alerting
Performance regression is classified as high severity — same as a site being down. Your team receives an immediate alert across configured channels: Slack, email, PagerDuty, Opsgenie, or SMS.
03Under the hood
How performance baseline regression detection works
- 01
Build the baseline
Sitewatch tracks the response time of every full-page check and maintains a rolling p95 baseline per site. The baseline reflects what "normal" looks like for that specific site — not a generic threshold.
- 02
Detect deviation
On each check, the current response time is compared to the rolling p95 baseline. If it exceeds 2x the baseline, the check is flagged as a potential regression.
- 03
Confirm with consecutive checks
A single slow check is discarded — transient latency is normal. Only when 3 consecutive checks show the same deviation is a PERFORMANCE_REGRESSION incident created.
- 04
Alert and auto-resolve
A high-severity incident is created and your team is alerted. When response time drops back below 2x baseline in subsequent checks, the incident resolves automatically.
04What you get
Outcomes, not a feature list.
Measure against your site, not somebody else’s threshold.
A fixed speed limit is wrong for every site it is applied to — it either never fires or fires constantly. Sitewatch learns what normal looks like for each page and alerts when a release moves it, which is the thing you actually want to know.
p95
Rolling baseline per site
2x
Deviation threshold
3 checks
Confirmation required
Your CI passed. Is your site as fast as it was yesterday?
Free plan. No credit card. 2-minute setup.
05FAQ
Frequently asked questions
The existing SLOW_RESPONSE detection fires when a site exceeds an absolute response time threshold (e.g., 3 seconds). Performance baseline regression detection fires when a site gets significantly slower than its own established normal — regardless of absolute numbers. A site that normally responds in 300ms and starts responding in 700ms will trigger baseline regression but not the absolute threshold. Both checks run simultaneously.
Sitewatch begins accumulating baseline data from the first check. The baseline stabilises over the first few days of monitoring as more p95 data points are collected. Regression detection is most reliable after a site has an established check history.
Three things must be true: the current response time exceeds 2x the rolling p95 baseline, this condition is observed across 3 consecutive checks, and the site is not already in a SLOW_RESPONSE incident state. This combination eliminates false positives from transient spikes while catching sustained regressions reliably.
Baseline regression is tracked per monitored page. Each page in your site configuration has its own p95 baseline. A regression on your checkout page fires independently of your homepage response times.
Performance baseline regression detection is included on all plans — Free, Starter, and Pro. No configuration or add-on required.
06Explore more