Uptime monitoring has a blind spot, and every one of us has been bitten by it: the site that is up — answering every check, keeping its 100% badge — while quietly running four times slower than it should. Customers feel it immediately. Your monitoring says everything is fine.
The traditional fix is a response-time threshold: “alert me if it’s over 800ms.” But what number do you pick? Your site answers in 120ms on a quiet Sunday morning and 400ms during Monday’s rush. Set the threshold low and you drown in false alarms every peak hour. Set it high and it never fires until things are already on fire. Most people set one once, get burned, and turn it off.
We built Anomaly Detection to skip the guessing entirely.
Your Monitor Already Knows What Normal Is
FlareWarden has been checking your site around the clock — for some of you, for a very long time. All that history describes your site’s normal better than any threshold you could invent.
So that’s what we use. Anomaly Detection learns a response-time baseline for every hour of the week from your monitor’s own checks. Tuesday at 2 PM is compared against other Tuesdays at 2 PM, not against a global average. When checks land far outside that learned range, the chart marks them — amber while we’re watching, red once the deviation persists long enough to be confirmed. Confirmed anomalies go into a log with the plain facts: when it happened, what we observed, what we expected, and how far off it was. Something like: observed 412–589ms, expected 118–176ms, 4.1× above normal.
If you want a nudge instead of a chart, opt into email alerts or subscribe a webhook to the new anomaly.detected event. Both are off by default — this feature is designed to inform, not to page you at 3 AM.
Two things anomalies will never do: they never touch your uptime percentage, and they never open incidents. “Slow” and “down” are different problems, and we keep them in different lanes.
No Thresholds, One Knob
There is exactly one setting: a three-position sensitivity knob. Low flags only the dramatic stuff. Normal is a balanced default tuned to bias against false positives. High speaks up about smaller deviations sooner. That’s the whole configuration surface.
Single slow checks don’t fire anything. A confirmed anomaly requires the deviation to hold across consecutive checks, so a one-off network blip stays a blip.
Cron Jobs Drift Too
The same idea applies to background jobs, where the failure mode is even sneakier. A nightly backup that used to take 4 minutes starts taking 40. It still completes. It still pings “success.” Nothing is technically broken — until it collides with your morning traffic, or the disk fills, or it silently stops finishing inside its window.
For cron monitors, Anomaly Detection baselines run duration from your job’s last 50 completed runs, using the median and a robust measure of deviation so one weird run doesn’t skew things. There are sensible floors built in: a 2-second job that takes 3 seconds will never be flagged. But when a run lands far outside the job’s own normal, you’ll know — while it’s still a curiosity, not an outage.
As far as we can tell, no other tool in this space covers learned baselines for uptime and cron duration drift in one product.
Honest Words About the Learning Period
Baselines come from real data, and we’d rather tell you exactly where the model stands than pretend it’s instant magic.
New monitors show a basic expected range after about 3 days, and reach full weekday-versus-weekend accuracy after about 21 days of history and enough checks — the chart shows a clear learning indicator until then. But if you’re already monitoring with FlareWarden, you skip the wait entirely: baselines are computed from the history we’ve already collected, so long-running monitors get value the moment you flip the toggle.
On Every Plan, With Simple Pricing
Features like this are usually where the enterprise gate slams down — the one comparable implementation in our corner of the market starts at over $625/mo. We think that’s silly. A freelancer’s client site deserves a learned baseline too.
Anomaly Detection is available on every FlareWarden plan, including free. Enabling it on a monitor uses one additional slot from your plan’s unified monitor pool — the same pool everything else draws from. Turn it off any time and the slot frees immediately (we keep the baseline, so re-enabling is instant). If you run out of room, upgrade a tier or add 10 extra monitors for $5/mo.
Try It
If you’re already using FlareWarden, open any uptime or cron monitor and look for the Anomaly Detection panel — or check the box when adding your next monitor. If you’re new, create a free account and let the baseline build while you get on with your week.
For the details — baselines, sensitivity levels, the learning ladder, webhook payloads — see the Anomaly Detection documentation.
Your site has a normal speed. Now you’ll know the moment it loses it.