Validation

The model reads; code decides what passes

Arithmetic, formats and duplicates are checked deterministically, outside the model — so a confident-sounding wrong answer still fails.

Overview

Deterministic hard gate

The failure mode of any extraction model is being confidently wrong, and a confidence score cannot catch an invoice whose lines do not sum to its total. Clio checks arithmetic, formats and duplicates in ordinary deterministic code, outside the model, so those errors are caught by something that cannot be persuaded.

What it does

Checks the model cannot talk its way past

Arithmetic is verified, not trusted.

Formats are validated deterministically.

Duplicates are detected before anything is written.

Checks run outside the model, so bad data cannot auto-clear.

Includes the VÖEN alongside the usual fields.

How it works

How the gate works

1Extraction produces field values with confidence
2Deterministic code re-checks arithmetic and formats
3Duplicates are detected against what already exists
4Anything that fails is held for a human
FAQ

Common questions

Why not rely on the model’s confidence?

Confidence does not catch an invoice whose lines do not sum. Arithmetic is checked in code, which cannot be persuaded.

What is actually checked?

Arithmetic, formats and duplicates — deterministically, outside the model.

Can a bad record clear automatically?

No. That is the point of putting the gate outside the model.

Explore more

More of what Clio does

Get started

See Clio read one of your documents

Request a demo to watch an invoice or contract become a validated, human-approved CRM record.