Extraction procedure

How a document type is parsed into structured facts.

Part of Platform in the register catalog.

Register ID

extraction_procedureUse this name to ask for the register. The API calls it a shelf ID.

Each record is

A procedureA rule or method Demiton follows.

Status

AvailableFilled today from connected systems, uploads or public data.

Data

OrganisationPrivate to each organisation's own workspace.

Version

v1Current version.

Download JSON SchemaSee an example record

Standards

This register holds records these standards and laws ask for. Holding them does not by itself certify compliance.

• ISO 19650Information management using building information modellingInternational

Connectors

Connected today

Demiton keeps this register current from these systems now.

No system fills this register yet.

On the roadmap

Systems in the planned connector rollout that can fill this register. Partial means the system holds only some of its fields.

None planned yet.

Fields

11 fields: 4 required, 5 optional, 2 defined by each organisation.

Required

Every record carries these fields.

org_entity_idWhole number

The org spine entity the current procedure is anchored to; makes the row per-org rather than per-tenant-blind.

Example: 501

methodsList

The per-doc-type extraction methods; the payload the shelf exists to carry.

doc_typeText

Which document type this method configures.

Example: programme

source_systemText

Producing system of the fact (AUTHORED for this shelf, DEMITON on the import path)

Example: DEMITON

Optional

Present when the source system records them.

llm.modelText

The model chosen for the LLM extraction pass, e.g. a hosted chat model name.

Example: gpt-5.1

parse.engineText

The primary text-lift engine, default pdfplumber

parse.ocr_fallbackTrue or false

Whether to OCR a page when the text-lift yields nothing.

statusText

Whether the operator has confirmed the procedure or it is still a candidate.

Example: completed

authored_byDateYYYY-MM-DD

Operator and timestamp recorded in _authored at write time.

Defined by each organisation

Their contents come from the organisation's own forms or systems, so they differ between organisations.

parse.keyword_signalsList

Extra page-flag keywords added to the parser's built-ins.

llm.promptText

The per-doc-type extraction system prompt.

Example record

A sample from Bridgeworks Group, Demiton's demonstration company. The values are illustrative; the shape is exactly what the contract describes.

extraction_procedure.example.json

{
  "doc_type": "programme",
  "llm.model": "gpt-5.1",
  "methods": [
    "regex",
    "llm"
  ],
  "org_entity_id": 501,
  "source_system": "DEMITON",
  "status": "completed"
}

Use this contract

The contract is a standard JSON Schema (draft 2020-12) document, so any JSON Schema validator can check a record against it. Keywords starting with x-demiton- add Demiton's own meaning, such as units and requiredness, and validators ignore them.

Schema URL

https://docs.demiton.io/registers/extraction_procedure/v1.json

Show the full JSON Schema

Loading the schema…