Skip to content
epok

From a change
to a decision
you can revisit.

Epok is the change-control and evidence layer for AI systems. It connects technical work to the reasoning, sources and review that make it useful to the next person.

Sign in to your workspace
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

01 / Context

Give the change
a starting point.

Record the system’s purpose, owners, deployment context and limitations. Keep an accepted baseline visible so a proposed improvement has a meaningful reference.

Example system

System
Digit recognition
Purpose
Handwritten digit classification
Accepted baseline
Compact MLP
Validation balanced accuracy
97.2%

Historical example

02 / Evidence

Keep the result
close to its source.

Inspect the proposed variation and the relevant parent or baseline. Link model, dataset and evaluation evidence inside the selected system, with source detail available for review.

Example sources

DatasetMNIST dataset
ModelSmall CNN
EvaluationEvaluation log

03 / Review

A useful decision
keeps its conditions.

Accept, reject or record an inconclusive finding. The outcome stays in the experiment lineage and research record. Acceptance for governance evidence is a separate decision.

Example decision

Accept the small CNN + BatchNorm model as the new baseline pending independent model review.

Internal decision recorded
Independent model review still required

04 / Handoff

Prepare the handoff.
Preserve the gaps.

Map evidence to controls in the Gap Workbench. Assemble Evidence Pack drafts from stored evidence and deterministic templates, with sources, blockers and reviewer states visible.

Source evidenceControl reviewEvidence Pack draft

Missing evidence remains a blocker.
An export is not a regulatory approval.

Evidence can come from research.
The decision stays with your team.

Epok includes research proposals, literature evidence and experiment lineage. Execution inside Epok is currently limited to a digit-recognition demonstration. Broader autonomous execution and automatic deployment are not offered.

Read the operating boundaries

Start with one AI system

Make your next change traceable.

Discuss beta access

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