Skip to main content
Accuracy tells you whether an investigation got the cause right. Time to diagnosis looks at the incident as a whole: how long it took, from the moment the investigation started, for anyone on the incident to work out what was wrong. That might be the investigation, or it might be one of your responders.

​
When the root cause is diagnosed

Once an incident closes with a known cause, we look back through its record to find the first time someone stated that cause. We search the channel messages, the incident call (what was said and the in-call chat), and every version of the investigation’s hypothesis. The earliest match is when the root cause was diagnosed, whether it came from a responder or from the investigation. It’s the step marked Root cause diagnosed on the investigation’s timeline.

​
An example

1

The investigation starts

Checkout starts returning 500s and an incident is declared. The investigation starts, and this starts the clock.
2

The investigation narrows it down

Within a minute, the investigation traces the errors to the payments service and posts a heads-up. It then finds a spike of database connection-pool timeouts in that service’s logs. These clues tell you what’s failing, but not why, so the root cause isn’t diagnosed yet.
3

A responder names the cause

Twelve minutes in, a responder spots that the timeouts began right after a deploy that cut the connection-pool size, and says so in the channel. The root cause is diagnosed here, so the time to diagnosis is 12 minutes.
4

Fixed

Responders restore the pool size. Fixing the incident comes after the diagnosis, so it doesn’t change the time to diagnosis.

​
Seeing it in your dashboard

You can see time to diagnosis under Speed on the Investigations page in your dashboard. It shows the median time to diagnosis across the timed incidents in the selected period, as well as the incidents grouped by how long the time to diagnosis took. Like the rest of the Investigations page, it appears once at least 10 of your incidents have a known cause that we could trace back through the incident’s record.

​
FAQs

We can only time an incident when we know when the investigation started and when the cause was first stated. An incident is left out if its cause was never established, if we couldn’t find where the cause was first stated, or if the cause was stated before the investigation started.
The investigation starting is the earliest point it could help with the diagnosis, so it’s the fairest place to start measuring from.

Measuring accuracy

How we grade investigations against what really caused your incidents.

Measuring engagement

How responders read, steer, and act on what an investigation finds.

How investigations work

The process behind a result, and how an investigation builds conviction in real time.

Incident channel experience

How investigations show up for responders while an incident is live.