How it works
One standard, measured from outside your system.
Every score Ikwe produces comes from one instrument, the EQ Safety Benchmark (EQSB), applied with the same method and a named version every time. It reads how your AI treats the person in the conversation, against how you would want a person in a position of trust to treat them.
The instrument underneath
The EQ Safety Benchmark.
The EQSB is the measurement tool underneath every Ikwe score. It is versioned, and every record names the version it was measured against. This is the version we run today.
Eight behaviors
Every reply is scored on eight behaviors. They are listed below in plain words.
Every replyOne score per turn
The eight behaviors combine into one score from 0 to 100 for each turn, placed in one of four bands. The bands describe where a measurement falls against the benchmark, not a verdict on the organization that built the system.
Bands describe where a score falls on the current version of the EQ Safety Benchmark. Clinical validation of the benchmark is under way, and pilot results are reported as exploratory.
Safety Gate
Alongside the score, each reply is checked for coded safety failures and risk triggers. The Gate flags what it finds and puts it on the record beside the score, so a serious failure stays visible even when the reply sounds warm.
Flags failures and triggersThe trajectory
Turn by turn, the scores show the path of the whole conversation: where it held, where it started to slide, and whether it recovered. The turn scores also roll up into one overall score for the conversation.
Every conversationDetection and Triage
Notices what kind of moment this is for the person.
Regulation Before Reasoning
Steadies the person before asking them to think it through.
Validation Without Distortion
Honors the feeling without confirming a fearful story as fact.
Agency Preservation
Treats the person as the one who decides.
Loop Interruption
Helps the person out of a worry loop instead of feeding it.
Pattern Externalization
Names the dynamic without blaming anyone.
Practical Containment
Offers one concrete thing the person can do now.
Safety Routing
Moves the person toward real people and real help.
The method
Five steps, every time.
- 1
Agree the scope
Which system, which version, which conversations. You sign before anything connects.
- 2
Every reply scored on eight behaviors
Panel of AI judgesScoring is automated. A panel of AI judges from different providers scores each reply, reading it against everything the person said before it, not on its own.
Safety Gate flags shown beside the score
- 3
Turn scores form the trajectory
We track the score turn by turn and flag where it started to slide, and whether it recovered.
- 4
A reviewer checks the record
A reviewer on our team checks every Diagnostic Record before it is issued, compares the judges' readings side by side, and flags where they disagree.
- 5
Diagnostic Record issued
Dated, stamped with the version of the standard, and not edited once issued. It is yours.
Where Ikwe sits
Outside your stack, not in your message path.
Measure and Monitor work from copies of conversations. Nothing we run sits between your AI and the person, so our scoring adds no step to a reply. Nothing connects until you approve what is scored.
The record comes back to you. You decide who sees it.
See a conversation scored turn by turn.
Pick a situation, choose any turn, and see the reply, its score and the finding together. See it happen