Did the editor preserve the provenance? Test the export.
A repeatable before-and-after protocol for Premiere, CapCut and other editing workflows without unsupported version claims.
The short answer
To check whether editing preserves C2PA, keep the original untouched and inspect an export made with the actual delivery preset. Compare manifests, actions and ingredient references. A new declaration may legitimately change the record; a missing one is an evidence gap to investigate, not proof that the export is deceptive.
Source context: Official c2pa-web documentation. The workflow recommendations below are editorial analysis.
Test the path your team really uses
“Supports Content Credentials” is too broad to be an export acceptance test. A desktop application, mobile application, cloud workflow and batch preset can behave differently. Instead of assuming a universal Premiere or CapCut switch, write down your actual application version, operating system, source format, export format and preset. Then run a controlled experiment on that path.
Choose a small file with a known embedded manifest and keep an untouched copy. Create one export without visual edits and one with the same ordinary edit your team performs every day. Avoid changing several settings at once. A simple matrix makes a disappearing or altered provenance record much easier to investigate.
Compare structure, not visual similarity
Inspect every output and save its JSON. Compare the active manifest, recorded actions and ingredient references. A successful preserving workflow may create a new active manifest instead of leaving the old JSON unchanged. Look for an understandable link back to the source rather than requiring every string to match.
Also review the file that reaches the client. A transfer service may deliver the original while a messaging app may create a derivative. If you only inspect the editor’s local output, you cannot conclude that the client’s copy carries the same evidence. Include the transfer step as its own row in the matrix.
Write a useful failure report
If a record disappears, state the last step at which you observed it. Include exact formats and settings, the reader result and whether you inspected the source independently. If validation fails, preserve the returned code; do not simplify it into “metadata removed.” A missing record and a broken binding are different failure modes.
Build a tiny regression pack for the studio. Re-run it when upgrading the editor, changing presets or switching delivery tools. That pack should include a permitted signed asset and a known unsigned one. Your result is then a dated statement about a reproducible workflow, not a permanent claim about every product with the same brand name.
A comparison for your review
| Copy | Purpose | Compare |
|---|---|---|
| Original source | Baseline evidence | Initial active manifest |
| No-edit export | Export-path isolation | Presence and new declaration |
| Edited export | Normal production change | Actions and source relationship |
| Client download | Transfer-path isolation | Evidence in delivered copy |
Put it into practice
- Record app, version, OS and preset.
- Change one variable per test.
- Inspect every file independently.
- Check source relationships rather than JSON equality.
- Repeat after upgrades and delivery changes.
Bring your own evidence.
Inspect a local file and see what it actually declares.