C2PA for documents

Most C2PA discussion concerns images, video, and audio: a camera or a generative model produces an asset, a manifest is embedded in the file, and downstream tools display the credential. Documents share the goal and almost none of the mechanics.

Three ways documents differ

• They are generated, not captured. There is no sensor in the lineage of a compliance report. The appropriate inception signal is data-driven media, not captured media, and getting this wrong misrepresents the asset's origin.
• Embedding support is uneven. The specification contemplates embedding manifests in PDF, but implementation support in the common open-source tooling has lagged. Detached sidecar manifests are frequently the only practical mode.
• Documents are assembled from records. A generated report's provenance question is not only "who signed this file" but "what evidence is this file asserting, and can that be checked independently?"

Embedded versus sidecar manifests

An embedded manifest travels inside the file, which is ideal: the credential cannot be separated from the asset by accident. A detached sidecar is a separate .c2pa file that accompanies the asset and is matched to it by filename.

Sidecars work, and validators support them, but they introduce a handling burden: the pair must stay together, and naming conventions differ between tools. Where a document format lacks working embed support, a sidecar is the honest choice rather than claiming embedding that the tooling does not perform.

Manifest stripping is the deeper issue

Whether embedded or detached, a manifest can be removed. Platforms re-encode, pipelines rewrite, and a stripped file looks exactly like a file that never carried provenance at all. For media, that is a well-known limitation. For documents retained as evidence, it is a serious one.

Anchoring the manifest hash to a public ledger addresses the gap: even if the credential is stripped, the fact that a specific document was credentialed at a specific time remains provable from public infrastructure.

Conformance

Rubric Protocol signs the compliance evidence documents it generates, and only those documents. It ingests no third-party media and records no ingredients, which keeps the product's provenance claims narrow and checkable.