For compliance, risk and procurement
Know what your AI system was tested against, and who approved it.
Every finding traces back to the run that produced it, the rule it was held to and the accounts that approved the release. Read it in the browser, with nothing to install.
Requirements
Your requirements, checked against the evidence.
Choose the catalogues that apply where the system is used. Each requirement reads met, not met, partly met or unable to determine, with the evidence behind it.
Evidence
Each finding traced to its run
A finding names the rule it rests on, the run that decided it and the bar it was held to. Only runs of the system's current version, made within its review period, can meet a requirement.
Approval
Who approved the release
Where the policy asks for it, a passing release waits for approval by two accounts, one of them an Owner. The decision stays what the evidence said, and the record names who approved.
What it covers
Beyond the score: what a reviewer asks of a release.
Governance
Requirements and approvals
Catalogues selected by where the system is used, a finding for each requirement, and approval of a passing release where the policy asks for it. Governance guide
Safety and risk
Ceilings on what goes wrong
A ceiling bounds how often something goes wrong on the cases your suite holds. A hazard register names what could go wrong, and seeded attack cases run in a sandbox. Risk ceilings guide
Fairness
Gaps between groups
The largest gap in an error, false positive, false negative or selection rate between the groups you declare, decided on an interval, with small groups reported as not enough evidence. Fairness guide
Privacy
Personal data and memorization
Checks that answers carry no personal data and repeat nothing memorized, without storing the values they look for. Privacy guide
People and AI
Reviewers, and decisions with the system
Your reviewers label in the browser, blind. A study measures whether people decide better with the system's output than without it. People and AI guide
Continuous assurance
Still true after it ships
A decision says when the system it describes has changed since. Once it is live, people review a sample drawn from real traffic. Production review
Can you show why this AI system was allowed to ship?
Oloproof keeps the answer: what was measured, against which bar, what was decided, and who approved it. A team can share one run's evidence as a page anyone with the link can open, labelled as reported by the developer, with no case content and nothing to install.