Failure taxonomy

Your bug, or the agent’s confusion?

Every failure is classified — product defect, agent shortcoming, environment problem or assertion mismatch — so a confused agent never arrives as a bug report.

Overview

Honest failure classification

An autonomous tester that reports every failure as a defect is worse than no tester, because someone has to triage the noise. Argus classifies each failure by cause and has the agent declare why it gave up, so a suspected product bug is flagged the moment it happens and an agent that simply got lost is recorded as exactly that.

What it does

Triage that has already been done

Failures are classified by cause, not just recorded.

Product defect, agent failure, environment failure and assertion failure are distinguished.

Loop, step-limit and timeout protections are separate outcomes.

The agent declares why it stopped.

An agent getting lost is never filed as a software defect.

How it works

How a failure is classified

1A run stops short of its objective
2The agent states why it gave up
3The failure is classified by cause
4Only suspected product defects are flagged as findings
FAQ

Common questions

What stops it filing false bugs?

Classification plus a rule: the AI never asserts a defect on its own authority. A human reviews and files it.

What does a suspected bug come with?

A ready-to-file report — environment, build stamp, the agent’s diagnosis, assertion results and step-by-step repro.

Are infrastructure problems separated out?

Yes, environment failures are their own class, and a failed preflight aborts before any model call.

Explore more

More of what Argus QA does

Get started

See agents test your own application

Request a demo to watch a multi-role regression suite run through your real interface.