Skip to content

Knotest by HephaKnot

An evaluation workspace from HephaKnot. Check data, test models, and keep the evidence behind your decisions together.

Open workspace ↗Documentation ↗

Inside Knotest

Check your data before testing your model

A good model starts with data you can trust. Find the checks that need attention before they affect your results.

6 passing2 failing
IntegrityCompletenessValidityConsistencyUniquenessRepresent.LeakageRedundancy
  • Data Integrity: 88/100
  • Completeness: 91/100
  • Validity: 28/100
  • Consistency: 22/100
  • Uniqueness: 84/100
  • Representativeness: 76/100
  • Leakage Risk: 90/100
  • Redundancy: 82/100

Complete data can still be inconsistent.

In this example, completeness scores well. Validity and consistency fall behind. Check formats and conflicting values before the next model run.

Explore all data checks
Data Integrity88/100Pass
Completeness91/100Pass
Validity28/100Fail
Consistency22/100Fail
Uniqueness84/100Pass
Representativeness76/100Pass
Leakage Risk90/100Pass
Redundancy82/100Pass

Test how your model handles real conditions

Choose a test to see what it can reveal, and what your team could do next.

Does it work on the data you actually use?

Original input

Baseline test

Know your baseline.

Compare it with your acceptance criteria.

Explore scenarios, metrics and supported models

Start with an unchanged input

Establish how the model performs before you introduce harder conditions.

Test scenarios

  • Baseline

Test metrics

  • Accuracy
  • F1
  • AUROC / AUPR
  • mAP 50:95
  • mIoU
  • RMSE / R2

Supported models

  • TabularClassification · Regression
  • VisionClassification · Detection · Segmentation · OBB

Track risks and how you address them

Keep the issue, the response, and the supporting records together. Give the next reviewer a clear path through the evidence.

  1. Fairness changes across groups

    Record the affected groups and the test that exposed the difference.

  2. Assign the next step

    Name an owner, document the response, and agree when to test again.

  3. Keep the evaluation in reach

    Link the test results and source records so the decision can be reviewed.

Explore the risk register

No owner assigned to AI management

R-42001-014 · Level: High

Inherent

Fairness changes across groups

R-23894-021 · Level: Medium

Mitigated

Missing evaluation records

R-TS42119-006 · Level: Medium

Residual

Sensitive to corrupted data

R-QA-088 · Level: Low

Mitigated

Access & enquiries

For current access, pricing, or help with an evaluation, contact the team. The documentation describes how the platform works.

Contact the team ↗

Release notes

Product updates will be published here.