Know when your
model knows.

Typed uncertainty for any classifier. Every prediction returns CERTAIN, PARTIAL, or UNCERTAIN.

Live demo

Loan application assessment — 3 sample predictions:

Sample 1CERTAIN

approve

margin: 7 · support: 12commit
Sample 2PARTIAL

approve / deny

margin: 0 · support: 4narrow
Sample 3UNCERTAIN

insufficient evidence

margin: 0 · support: 0abstain

The problem

Confidence scores lie.

Your model says 90% confident. It's wrong. You approved the loan, shipped the part, cleared the patient. The confidence score told you what to believe. It didn't tell you what to do.

Traditional classifier

90% confidence

Wrong. Approved a bad loan.
Cost: $47,000.

Tercet

PARTIALapprove / deny

Flagged for review. Analyst catches it.
Cost: $0.

How it works

Three types. Not a score.

Every prediction tells you what to do with it — not just what it thinks the answer is.

CERTAIN

Commit

The model is certain. One class has decisive evidence. Auto-decide.

67% of predictions · 99.2% accuracy

PARTIAL

Narrow

Evidence supports 2–3 classes. The model narrows it down. Human picks from a shortlist.

26% of predictions

UNCERTAIN

Abstain

Not enough evidence. The model says so before you waste time. Route to a specialist.

7% of predictions

Case study

The numbers.

Credit risk assessment. 1,000 loan applications. 5-fold cross-validation.

0

risky loans caught

that the baseline approved

$0K

in prevented losses

on 1,000 applications

0%

auto-decided

zero human review needed

0%

committed accuracy

when the model says CERTAIN

Four lines to typed uncertainty.

Scikit-learn compatible. Fits any dataset. No dependencies beyond NumPy.

1from tercet import Classifier
2
3clf = Classifier()
4clf.fit(X_train, y_train)
5
6result = clf.predict(sample)
7# → {'type': 'CERTAIN', 'class': 'approve', 'margin': 7, 'action': 'commit'}

Also available as a REST API. See the docs →

Pricing

Free to start. No credit card required.

Free

$0

  • 50,000 predictions / month
  • 5 models
  • Full API access
  • Community support

Team

$49 / month

  • 500,000 predictions / month
  • 20 models
  • Multiple API keys
  • Usage dashboard
  • Email support

Business

$199 / month

  • 2,000,000 predictions / month
  • 100 models
  • Audit logs
  • Batch SLA
  • Priority support

Enterprise

Custom

  • Unlimited predictions
  • Unlimited models
  • SSO / SAML
  • Dedicated instance
  • Compliance docs