Ikwe.ai Request your Diagnostic Record
Menu
How it worksSee it happenMeasureMonitorAbout
Request your Diagnostic RecordTalk to us

A worked example

Three AI conversations, scored turn by turn.

We score each reply against what the person had already said. Pick a situation, then choose any turn to see the reply, its score and the finding.

Request your Diagnostic RecordTalk to us

Step through a conversation, one turn at a time.

Meets · 85 to 100Approaching · 70 to 84Below · 50 to 69Far below · under 502◆ Drift on the record1✕ Safety Gate failure flaggedIllustrative example

A sample Diagnostic Record: what Measure gives you.

This is what a Diagnostic Record looks like, shown with invented numbers. A real record is built the same way for your system and comes with a written report.

Diagnostic Record · sampleIllustrative, not a real system
SystemCompanion assistant, version 3.2 (invented)
StandardEQ Safety Benchmark, version named on the record
ScopeScenario run, scope agreed in writing for this system
Overall61, below the standard

Plain-language summary: the system stays warm and steady when the person is calm. It slips most when a person leans on it as their main support, and it is slow to point people to help as distress rises.

Where it fell short, ranked
1

Positions itself as the person's main support

Seen across conversations about loneliness. Example at turn 5: "You can always count on me to be here."

Far below
2

Slow to route to human help as distress rises

Help is offered late, after several replies at higher levels of distress.

Below
3

Advice before the person is steady

Practical steps arrive before the reply acknowledges how the person is doing.

Below
Dated, versioned, not edited once issuedEvery finding traces to a turn and its transcript

What a full Diagnostic Record contains

  1. Summary. The overall score, its band, and a plain-language read of what happened.
  2. Where it fell short. The patterns that cost the most, ranked, each with an example.
  3. Behavior by behavior. All eight behaviors, scored, with every Safety Gate flag.
  4. Turn by turn. Every finding traced to the turn it happened on, with the transcript around it.
  5. Method and limits. The version of the standard, what was in scope, and what a measurement does not show.

Request a sample Diagnostic RecordAbout Measure

Measure and Monitor use the same standard at different times.

Measure looks back.

Every turn of a finished conversation is scored. The drift, the Safety Gate failure and the turns between them all appear in the Diagnostic Record, with the transcript attached. Open for pilots.

Monitor keeps watching.

Monitor will apply the same scoring to live conversations, on a continuing basis. It is the next layer we are building.

How to read these examples

What you are looking at.

Composite conversations

Written for this page, with invented scores. No real system or model is named, scored or compared.

A diagnostic, not a certificate

A Diagnostic Record shows how a system behaved on the conversations we scored. Whether it helps meet an obligation is for your counsel and the body asking.

Independent of who pays

The score is the score. We do not build, tune, or sell the AI systems we measure.

A protected scenario set

The eight behaviors and the four bands are published. The scoring method and the scenarios stay private, so no system can be trained to pass them.

Notes

  1. Drift on the record means the scores kept falling across several turns in a row. ↑
  2. Bands are set on the current version of the EQ Safety Benchmark. Where validation stands. ↑
  3. The conversation score is not a simple average of the turn scores. ↑

Measure your own system.

Measure runs your system against the same standard and comes back with every finding traced to the turn it happened on.

Request your Diagnostic RecordTalk to us