Glossary

The vocabulary the product, the API and these docs all use. Where a word means something specific here, this is the definition that wins.

Access profile: A definition of what a person may see, independent of queue assignment. Effective visibility is queue ∩ profile.
Boundary: Where one document ends and the next begins inside a multi-document file. Split records a confidence per boundary.
BYOK: Bring your own key. Routing IDPForge's model calls through your own LLM provider account, so the provider bills you for the tokens and IDPForge waives its LLM surcharge on those calls.
Class: A document type from a taxonomy, such as an invoice or a packing list. A document can legitimately carry more than one.
Connector instance: A configured, credentialed integration, set up in Settings → Integrations → Connectors. Pipelines bind one by id; they never re-declare credentials.
Coordinate resolution: The Extract step that matches extracted values to their position on the page, so a reviewer sees each one highlighted where it was read.
Coverage: The percentage of an operator's corrections that are sampled for a second pass. Set per person as a list of review-rule/percentage rows.
Credit: The unit IDPForge meters work in. Reserved when a job is submitted, settled to actual cost when it finishes, and released if unused — in full if the job fails.
Destination: Where a finished record is delivered. A pipeline may deliver to several; its output format is set once for the pipeline.
Document: The unit of extracted data. One uploaded file may produce many documents, or none.
Exception: A stage succeeded but its output needs a human. Distinct from a stage error, which is a failure.
Generic extraction: Extraction with no schema bound: the model returns what it can find rather than a validated field set.
Job: One submission to one pipeline. Owns processing status, including waiting for review; documents inside it own their own review status.
Lineage: The relationship between a document and the documents it was split from or rerouted into.
Operator: A person who clears exceptions — a workspace role. Their access profile and verification settings are assigned separately.
Overage: Optional paid credit used after your free and paid credits run out, up to a per-cycle dollar limit you set; billed at cycle close.
Park: Setting a document aside for a supervisor. Parking releases the claim and moves the document into the Parked queue's view.
Pipeline: A versioned set of stages that turns documents into delivered records.
Provider pool: A group of your LLM keys for one provider, balanced by a shared strategy, that pipelines route through under BYOK.
Reservation: The credits held for a job when it is submitted. The final charge can never exceed it.
Schema: The field definition applied at Extract, bound to a class by a routing rule.
Segment: A child document that Split carves out of an uploaded file; each segment is classified and extracted on its own.
Source: Where a job came from — an upload, the API, or a connector such as Google Drive. Shown in the Jobs list's Source column.
Straight-through: A document delivered without any human touching it: no exception, no correction.
Taxonomy: A tree of classes. A routing taxonomy maps classes to schemas; a label taxonomy only annotates.
Verifier: A member eligible to give a sampled correction its second pass. Coverage decides which corrections are sampled; verifiers pick them up from the Verifier queue.
Work queue: A saved live query over the document catalogue. Documents match a queue; they are never moved into one.
Workstream: The focused correction surface an operator works inside, one document at a time.

Was this page helpful?

Last updated 11 Sept 2026