Documenting 3D model provenance means recording where the model came from and every decision that shaped it: the source object, who captured it and when, the equipment and software, the processing settings, and crucially the interpretive choices — what was reconstructed, filled, or smoothed. This record (metadata plus paradata) is what separates a scholarly 3D model from a pretty picture, because it lets a researcher tell measured surface from educated guess. Capture it as you work, not from memory afterwards.

What is the difference between metadata and paradata?

Both travel with the model, but they answer different questions:

• Metadata describes what the model is: title, accession number, capture date, format, file size, rights. It is factual and structured.
• Paradata records how and why the model came to be this way: why a hole was filled by interpolation rather than left open, why one alignment was chosen, which areas are reconstructed versus measured.

Paradata is the heart of trustworthy 3D heritage. A model is the product of dozens of human decisions, and a future researcher needs to know which surfaces are real measurement and which are interpretation.

Why does provenance matter more for 3D than for a photo?

A photograph is a fairly direct capture; a 3D model is heavily mediated. Between the object and the published mesh sit alignment, dense reconstruction, meshing, hole-filling, decimation, and texturing — each a decision point. If a corbel had a missing corner that you interpolated, the model now contains geometry that never existed on the object. Without provenance, a viewer measuring that corner is unknowingly measuring your guess. Provenance makes the model honest and reusable.

What exactly should I record, and when?

Capture provenance at three moments: at the object, during processing, and at publication. A reusable template:

object:
  identifier:    HART-2024-031
  name:          Carved stone corbel, c.1480
  location:      St Mary's, north aisle, in situ
  custodian:     Parish of St Mary
capture:
  operator:      E. Reed
  date:          2025-03-18
  camera:        Sony A7 III, 35mm
  lighting:      diffuse daylight, overcast
  scale:         two 200mm coded scale bars, CHI targets
  n_photos:      214
processing:
  software:      Agisoft Metashape 2.1.1
  alignment:     High accuracy, 60000 key / 10000 tie points
  dense_cloud:   High quality
  decimation:    5.1M -> 250k faces (web copy)
interventions:           # the paradata that matters most
  - filled hole on lower-left, ~30mm, interpolated (NOT measured)
  - removed scaffolding pole reflection from texture
  - top face occluded; geometry incomplete, flagged
rights:
  licence:       CC BY 4.0
  credit:        E. Reed / Digital Relics

The interventions block is where most teams fail — it is also the most valuable thing in the record.

Which standards should I follow?

You do not need a heavyweight schema, but you should align with recognised principles:

Framework	What it asks of you
London Charter	Document sources, methods, and reasoning for any heritage visualisation
Seville Principles	Apply the London Charter specifically to archaeology and 3D
PREMIS	Record preservation events (capture, migration) for the digital object
CIDOC CRM	Model the object–event–actor relationships if publishing as linked data

In practice, implement the London Charter as a structured paradata document, and add PREMIS-style event records when the model enters a preservation repository.

How do I keep provenance attached to the model?

Documentation that drifts away from its model is worthless. Two complementary tactics:

1. Sidecar document — a human-readable paradata file (YAML, Markdown, or PDF/A) inside the same archival package as the mesh and source photos.
2. Embedded fields — write key provenance into the file itself, for example glTF asset.extras, so a model copied without its folder still carries its core record.

Use both: embedding survives careless copying; the sidecar carries the full reasoning the embedded fields cannot hold.

How does provenance support reuse and citation?

A well-documented model can be cited and built on. Provenance gives you the raw material for a data citation (creator, title, date, version, repository, persistent identifier) and tells the next user what they can safely measure or reconstruct from. When you publish, surface a short provenance statement next to the viewer — operator, date, scale source, and a clear flag for any reconstructed geometry — so trust is visible at a glance, not buried in a file.

Key Takeaways

• Provenance records the model's origin and every decision that shaped it; paradata records the reasoning.
• 3D needs provenance more than photos do because it is heavily processed and full of interpretation.
• Capture at three points: the object, processing, and publication.
• The most valuable record is the interventions log — what was filled, reconstructed, or removed.
• Align with the London Charter and Seville Principles; add PREMIS events in a repository.
• Keep a sidecar document in the package and embed key fields in the file (glTF extras).
• Good provenance enables citation, reuse, and visible trust at the point of access.

Frequently Asked Questions

What is provenance for a 3D model?

Provenance records where a 3D model came from and how it was made: the source object, who captured it and when, the equipment and software, the processing steps, and every decision that shaped the result. It is what lets a researcher trust and reuse the model.

What is paradata and how is it different from metadata?

Metadata describes the model (title, date, format); paradata records the reasoning and choices behind its creation — why you filled a hole, why you chose one alignment over another. Paradata captures interpretation, which is essential for scholarly 3D.

Why does 3D provenance matter more than for a photograph?

A 3D model is heavily processed and full of human decisions: hole-filling, decimation, texture choices, and reconstruction of missing parts. Without provenance a viewer cannot tell measured surface from interpolation, which undermines scholarly use.

What should I record while capturing and processing?

Record the object and its location, capture date, camera and lens, lighting, scale reference, software and versions, key settings, and every significant edit such as hole-filling or reconstruction. Note who did what and when.

Is there a standard for 3D paradata?

The London Charter and the Seville Principles set the framework for documenting computer-based heritage visualisation, requiring transparency about sources and methods. They are principles rather than a strict schema, so you implement them in a structured document.

How do I store provenance so it stays with the model?

Keep a human-readable paradata document in the archival package alongside the model and photos, and embed key fields in the file (glTF asset extras) where possible. Never rely on memory or on a separate system that can drift apart from the files.

Related reading

• Best practices to archive 3D models
• How to scale 3D models
• Documenting imaging paradata
• Provenance metadata
• PREMIS preservation metadata