AI TRANSPARENCY

Know what an output is—and how far to trust it

Nolpop distinguishes the source and trust boundary of original, edited, generated, and rendered artifacts across a connected repaint project.

Effective
2026-09-03
Version
1.0
Last updated
2026-09-03

1. Artifact classes and trust boundaries

OriginalSource suppliedEditedHuman directedGeneratedAI generated3D renderRendered
SourceA file uploaded by an authorized user, or data entered or captured directly by that user.A supplied source changed through surface selection, masking, crop, notes, or color edits.Model output based on a prompt, sketch, reference, selected surface, or another artifact.A render based on user-supplied, inferred, contextual, or measured geometry plus material, light, and environment settings.
Appropriate useReference, record keeping, internal review, and traceable input for later artifacts.Design decisions, proposals, comparison, and client communication when the edit scope remains visible.Idea exploration, concept options, missing-view direction, and early visualization.Color and finish review, presentation, view consistency, and—in measured workflows only—supported quantitative work.
LimitationCamera processing, framing, lighting, occlusion, noise, and capture error can still affect what the source shows.Its reliability cannot exceed the source, and unmarked editing can hide what changed.It can invent or alter detail, material, geometry, lighting, scale, or surrounding context and is not evidence of actual condition.Trust depends on geometry provenance and settings. A realistic image does not prove that geometry or dimensions were measured.

2. Fidelity levels describe evidence, not beauty

Fidelity records where geometry and scene constraints came from. It is not a visual-quality score.

  1. 01Conceptual

    Generated or manually arranged without verified geometry.

    Use for visual direction and option discovery only.

  2. 02Contextual

    Approximate geometry placed within supplied or inferred surroundings.

    Use to discuss context and camera direction, not quantities.

  3. 03Supplied model

    Geometry comes from a client, designer, or another named source.

    Use within the accuracy and revision limits of that supplied model.

  4. 04Measured

    Geometry is tied to a recorded capture or survey method and its stated tolerance.

    Use for supported quantitative work only when method, unit, tolerance, and review state travel with the result.

3. Metadata through share and export

  • A shared or exported artifact should retain its class, source relationship, creation method, and fidelity level. Relevant model, rendering, color, material, and review records should travel with it.
  • Cropping a screenshot or copying an image into another document can separate it from those disclosures. The recipient should be able to reach the project record or an attached disclosure without relying on appearance alone.

4. Human checks and responsibility

  • The person preparing a result must verify the selected source, changed surfaces, color direction, fidelity level, and intended use before sharing. A reviewer must know whether they are approving an appearance direction, a specification, a quantity, or completed work.
  • Nolpop output does not replace site inspection, substrate assessment, manufacturer instructions, a qualified specification, measurement, estimating judgment, contract approval, or safety review.

5. Physical samples and site measurement

  • Paint appearance varies with illumination, camera processing, substrate, texture, gloss, coating system, weather, and display calibration. Confirm final color and finish on the property with the relevant physical sample and professional judgment.
  • Generated or contextual geometry must not be used as measured area or quantity. A measured label requires a named method, units, tolerance or uncertainty, capture date, and review state.

6. C2PA and Content Credentials boundary

  • Nolpop reserves fields for future standards-based provenance such as C2PA Content Credentials. The current frontend does not sign artifacts, validate a credential chain, or promise that metadata will survive third-party export and editing.
  • A future implementation must identify the signer, claim generator, ingredient relationships, transformations, validation result, and failure state. Until that infrastructure exists, the visible Nolpop disclosure is an application record—not a cryptographic authenticity certificate.

7. Versions and archived policy

  • Version 1.0 is the first published AI transparency policy in this frontend. There is no earlier archived version.
  • Material changes will create a new version and keep prior text available with its effective date. Wording changes that do not alter a trust boundary may update the last-updated date without changing the major version.