Skip to content

Change control & evidence for AI systems

Know what changed. See the evidence.
Keep the reasoning behind the decision.

Explore the product
Internal classifierIllustrative example · not live results
Accepted baseline0.841

model-v3

Candidate result0.867

model-v4

Reference Macro F1 · +0.026

DatasetModelEvaluation log

Technical and governance impact reviewed in one Change.

01 / Technical change

model-v3 → model-v4

Upgrade the internal classifier from its production state.

Starting state retained

01 / 05

The model moves on.
The reasoning
should come with it.

A new dataset. A different model. A revised requirement. AI systems change long after their first release.

The result might survive in a dashboard. The rationale sits in a thread. The reviewer’s decision disappears into a meeting.

Epok connects the change, its evidence and the human decision in one system record. That connected record is the decision trace.

A number tells you what happened.
A trace tells you why it matters.

01

Compared with what?

Keep the accepted baseline and parent comparison beside the result.

02

Supported by what?

Follow findings back to source evidence, evaluations and recorded context.

03

Decided by whom?

Keep reviewer state and rationale attached to the finding, including rejected and inconclusive outcomes.

Follow a change through Epok

Technical detail.
Human judgement.
Same record.

Engineering needs to understand the result. Product needs to understand the tradeoff. Reviewers need the source and the decision.

Bring that evidence into the Gap Workbench and source-mapped Evidence Pack drafts. Missing evidence and further review stay visible.

How the record supports EU AI Act work

Start with one AI system

Make your next change traceable.

Discuss beta access

Private beta. A focused scope, agreed with your team.