Glossary

Definitions for Document AI terms, features, statuses, and workflows.

Use this glossary to decode the terms used across Document AI.

A–M

Account

The top-level tenant for Document AI settings, usage, users, API keys, webhook URLs, and billing.

Active runs

In-progress jobs shown on Workspace Home.

Admin

A role with broad operational access. Admin can manage users and API keys, but some account-level controls remain Owner-only.

Agent instructions

Optional guidance passed into extraction to shape how the model interprets a document.

API key

A credential for programmatic access to the platform.

Assertion

A field-level comparison rule in Test Dashboard that checks an extraction result against ground truth.

Audit log

A record of account-level changes, including Zero Data Retention updates.

Batch run

A multi-file processing job, such as ZIP-based extraction or classification.

Blob storage

External storage used when extraction results are too large to store inline.

Category

A grouping applied to document types or fraud rules, depending on context.

Certification status

A form-level tracking state used in Document Types.

Classification

The process of identifying a document's type and category before extraction.

Collection

A named container in Gallery used to organize documents.

Confidence score

A numeric score that shows how certain the system is about a classification or extracted field.

Continuation Of

A link between segments of a multi-document PDF when later sections continue the same document type.

Custom document types

A caller-supplied subset of document types used to constrain classification.

Custom extraction prompt

Optional prompt text supplied at extraction time to shape the output.

DOCX

The output format used by Document Generation.

Document type

A supported document definition in the shared registry used by classification and extraction.

Extraction

The process of converting a document into structured data using OCR, layout analysis, and LLMs.

Extraction group

A grouped extraction workload, often tied to multi-document or batch processing.

Field schema

The list and structure of fields expected for a given document type.

Fraud Detection

The rules-based and model-assisted analysis that checks documents for fraud signals.

Gallery

The Workspace area used to organize documents into collections.

Generation

The process of turning structured data into a finished report.

Ground truth

The expected result used in Test Dashboard to benchmark extraction accuracy.

Home

The Workspace dashboard for metrics, quick actions, active runs, and pipeline health.

Idempotency

A delivery pattern where a repeated webhook event can be safely ignored by checking its event ID.

Judge accuracy

A Workspace metric tied to LLM-based validation quality.

Layout analysis

The step that identifies structure such as tables, regions, and positions within a document.

LLM

A large language model used in classification, extraction, generation, and AI rule review.

Metadata

Optional caller-supplied context stored with a processing request.

Model

The selected AI engine used for a job, such as GPT-4.1, GPT-5.1, or Gemini 3.

N–Z

OCR

Optical character recognition. The step that reads text from PDFs and images.

Operator template

A reusable assertion pattern in Test Dashboard for comparing expected and actual values.

Owner

The highest-permission account role.

Pagination

The UI pattern for splitting long lists into pages.

Pending

A job status that means work has been accepted but not yet started.

Post-extraction analysis

The fraud detection phase that evaluates extracted values against configured rules.

Pre-OCR analysis

The fraud detection phase that checks the raw uploaded file before OCR runs.

Processed

A terminal job status that means processing finished successfully.

Processing

A live job status that means work is actively running.

Processor

The extraction or classification pipeline variant used to run a job.

Quick extract

A faster extraction mode intended for speed-sensitive workflows.

Quick links

Shortcuts on Workspace Home to common operational screens.

Raw-extraction-only mode

An extraction mode that returns raw extraction output without the full review experience.

Registry

The shared catalog of supported document types used by classification and extraction.

Retry with backoff

Automatic webhook redelivery using increasing delays after failure.

Risk level

The overall fraud outcome, such as CLEAN, LOW, MEDIUM, HIGH, or CRITICAL.

Rule category

A fraud rule grouping such as math_logic, visual_forensic, or compliance.

Rule weight

The value that determines how much a fraud rule contributes to the total risk score.

Search box

A UI search control used on several pages. Some documented instances are currently non-functional.

Signed URL

A time-limited link used to access a generated or source document securely.

Status filter

A list control that narrows records by state, such as Pending, Processing, Processed, or Failed.

System-supported categories

The built-in category list shown in the classification upload flow.

Template

A reusable starting point for fraud rules, report structure, or test configuration, depending on context.

Test Dashboard

The benchmarking workspace used to measure extraction accuracy against expected results.

Type restriction

A classification option that limits results to a caller-specified subset of document types.

Usage by Member

An account usage breakdown visible in usage reporting for supported roles.

Viewer

A read-only role with limited operational access.

Visual forensic

A fraud detection method that inspects raw files for visual signs of tampering.

Webhook

An HTTP callback sent when classification, extraction, or generation finishes.

Webhook signing key

The account secret used to verify signed webhook payloads.

Workflow

The ordered sequence of steps a user follows to complete a task.

Zero Data Retention

An account setting that processes documents without storing them after results are returned.

ZIP batch

A batch upload packaged as a ZIP file for multi-file processing.

Related pages

• Workspace covers navigation and operations.
• Document Types defines the shared document registry.
• User Management explains roles, account settings, and API keys.