IDENTITY LAYERS
Model identity, configuration identity and physical-instance identity are not the same thing.
Model identity
The commercial or canonical robot model, such as a Unitree G1 or a Comau Racer-5-0.80 COBOT.
Variant / configuration
A specific option set, edition or hardware configuration can change DoF, compute, sensors or payload.
Profile identity
An OpenRGD profile has its own identity and integrity commitment for the exact machine-readable configuration being evaluated.
Physical instance
A real deployed robot may additionally need serial, calibration, firmware and lifecycle identity beyond the generic model.
Source identity
Evidence must retain which document, manufacturer page, dataset or artifact produced the assertion.
Temporal identity
Claims can change over time, so valid-from, valid-to and verification dates matter.
PROVENANCE PIPELINE
Preserve the path from evidence to interpretation.
The critical rule is non-retroactivity: later interpretation should not silently rewrite the historical evidence that supported an earlier claim. OpenRGD’s reconciliation work applies the same principle to recovered artifacts and candidate contracts.
VERIFIED ≠ INDEPENDENTLY TESTED
Evidence quality needs more than one boolean.
A manufacturer-stated specification can be accurately captured and source-reviewed without being independently measured. OpenRGD.org therefore keeps evidence type, verification status and independent testing as separate fields. This prevents a reviewed product-page claim from being presented as laboratory validation.
CONTENT IDENTITY
A hash proves which content you have — not who approved it or whether it is correct.
The canonical OpenRGD repository uses OPENRGD_SOURCE_TREE_SHA256_V1 to commit the selected modular source tree. Repository guidance explicitly separates content identity from authorship, governance acceptance, compatibility and physical safety. The same distinction is useful for robot data: integrity is one layer of trust, not the whole trust model.
IMPORT PROVENANCE
Source-derived robot evidence should stay distinguishable from later enrichment.
The current OpenRGD importers record source filename, format, byte length and SHA-256 while extracting only supported Foundation evidence. Identity, safety policy, cognition and hardware authorization are not inferred from URDF or USDA. Later enrichment is therefore a separate operation with a separate provenance story.
OPENRGD.ORG EVIDENCE MODEL
Entity pages can expose claims instead of hiding them behind a polished specification table.
The OpenRGD.org catalog stores robot assertions separately from the canonical robot row. An assertion can carry its property, normalized value, evidence type, verification state, source URL, verification date, validity window and whether independent testing is recorded. The public robot pages now surface that distinction directly.