On this page
Quick answer
Every RubixKube root cause analysis states the cause, how sure Rubix is, and the evidence behind each step: the log line, the Kubernetes event, the metric, the deploy. Quotes are checked in code against what the tools returned, confidence is capped and held down when a claim is not cited, and your team rates each RCA Right, Partly right or Wrong.
An answer you cannot check is a guess. This one shows its working.
What an RCA contains
- Root cause, in one or two sentences, with a verdict: Root cause found, Probable cause or Cause not determined.
- Timeline, step by step, from the first change to the failure your users saw.
- Evidence for each claim, numbered, with the tool output it came from.
- Impact, current state and contributing factors.
- What was ruled out, so you do not check the same things again.
- Tasks: a Fix for the cause, and Follow-ups so it does not happen again.
The side panel shows Confidence, Completeness and how long the investigation took. The Full report tab has the numbered citations and charts, and Evidence has the raw material.
How Rubix keeps it honest
- Investigations only read. In every mode, an RCA can look but not touch.
- Several angles at once. Rubix reads the incident and everything grouped with it, then checks logs, events, metrics and recent changes across your environments and connected tools.
- Quotes are checked in code. Every quoted line is matched against what a tool actually returned. A quote that does not match is labelled Not verified word for word.
- Confidence follows the evidence. It is capped at 0.95, and held at 0.5 when the cause is not determined or the claims are not cited.
- It says when it does not know. An inconclusive RCA reads Inconclusive, lists its Limits, what it could not see, and proposes no Tasks rather than invent a fix.
A claim you can verify
A useful RCA is specific enough to prove wrong: "orders-api v2.4 raised DB_POOL_SIZE from 10 to 50, and orders-db reached its connection limit," with the deploy and the connection metric cited beside it. You can open each source and check it in seconds.
Your team closes the loop
- Was this the cause? Mark it Right in one click, or Partly right or Wrong with a note on what it really was. Rubix reads that note the next time a similar incident comes in.
- Ratings move confidence. A Right raises it, never above 0.95. A Wrong from anyone stops it rising.
- Run the RCA again from the page menu after you add a comment with new context. It skips every wait and writes a new report, and duplicate Tasks fold together.
- Take it with you. Export as Markdown for your post-mortem, or Ask Rubix about this to keep digging in chat.
Start free and read your first RCA, or see how root cause analysis works and reading an RCA.
Frequently asked questions
Can an RCA change anything in my systems?
No. Investigations only read, in every mode. Changes come later, from Tasks, and go through your mode and approvals.
What happens when Rubix cannot find the cause?
The RCA reads Inconclusive, says what it could not see in a Limits section, holds confidence at 0.5 and proposes no Tasks.
How is confidence set?
From the evidence. It is capped at 0.95, held at 0.5 when claims are not cited or the cause is not determined, and moved by your team's ratings.
Does Rubix learn from our ratings?
Yes. A Wrong or Partly right rating asks what the cause really was, and Rubix reads that note when a similar incident comes in.