A brand-new error is one line
in ten thousand.
By the time a new exception surfaces in a dashboard or a user report, it has been firing for hours. Epok flags a genuinely new error within minutes of its first occurrence — grouped, counted, and scoped to the services it hit.
New failure modes hide in volume. A fresh exception starts at a handful per minute, lost among the normal error noise. Counting total errors doesn't surface it — the count barely moves. Searching for it requires knowing the string, which you don't, because it's new.
So it runs until it's loud enough to notice or a customer reports it — and by then it's been degrading the experience for hours.
Fingerprinted by shape
Each error is normalized to its structural template — IDs, timestamps, and values abstracted out — so the same failure across thousands of unique strings is one pattern.
Compared against your normal
Every pattern is checked against the rolling 7-day baseline. A shape that has never appeared before is surfaced as new, not drowned in the total count.
Minutes, not hours
Detection runs continuously, so a new error is flagged within minutes of its first occurrence — long before it’s loud enough to notice by eye.
Grouped, not a storm
A thousand variants of the same new error collapse into one alert with a fire count — the pager rings once, with the full picture.
- ✓An alert the moment a genuinely new error pattern appears in production.
- ✓The normalized pattern, the count, the rate, and the services it showed up in.
- ✓Example raw lines for the pattern, so you can jump straight to the evidence.
- ✓Severity scaled to volume — a new error firing hundreds of times pages louder than a one-off.
Resurfaced errors — patterns that went quiet for a day and came back — are flagged too, so a regression you thought you fixed doesn’t sneak back in.
Watch it catch this one. No signup.
The live demo runs detection on real-shape data — every alert cited back to the line behind it. Or send your own logs in the 14-day trial.