Should Rovo expose reasoning confidence instead of only providing answers?

[AGASTYA SHARMA]

Rovo currently returns responses without clearly communicating its confidence level.

Would it be valuable if every response included confidence signals such as:

Retrieval confidence

Evidence completeness

Knowledge freshness

Reasoning confidence

Missing information score

Could confidence calibration improve enterprise trust and reduce AI overconfidence?

[Arkadiusz Wroblewski]

Hi @AGASTYA SHARMA 

I completely agree that Rovo needs to communicate uncertainty better, but I’d be cautious about relying on raw confidence percentages.

Self-assessed model scores can easily create false reassurance without rigorous calibration. Verifiable signals like evidence coverage, source freshness, missing context, and clean abstention offer much more value.

Atlassian is already tracking related trust features under ROVO-431 for Slack thresholds, ROVO-731 for unsupported-claim detection, and ROVO-730 for knowledge freshness.

Specific source claims and missing context are more actionable for enterprise trust than a generic percentage; we need clear attribution indicators, not an unexplained reasoning score.

Best,

Arek 🤠