How to prove you made your music — and what actually counts as evidence

This question used to be rare and is now routine: a distributor asks you to support a release, a sync client asks about AI use, a collaborator disputes authorship, or a platform flags a track. The instinct is to look for the one document that settles it. There isn't one, and understanding why makes the rest of this straightforward.

Why the finished file tells almost nothing

An exported track is the end of a process with the process removed. Two files that sound identical — one played and mixed over three weeks, one generated in twenty seconds — are, as artifacts, nearly indistinguishable. Everything that distinguished them happened before the export and was never recorded anywhere.

File metadata does not close that gap either. Creation and modification timestamps are trivially editable, they are rewritten by ordinary operations like copying and re-encoding, and everyone evaluating these disputes knows it. Metadata is worth attaching, but it carries little weight on its own.

What actually counts, roughly in order

Evidence in this area is not binary; it is a gradient of how hard something would have been to fabricate after the fact.

1. Project files with genuine edit history. A session with hundreds of small edits, undo history, dead ends, muted experiments and abandoned tracks is very hard to manufacture convincingly, because faking it means actually doing the work. This is usually the strongest thing you have.
2. Independently timestamped backups. Cloud sync version history, Time Machine, or a git-style versioned folder. The value is that a third party recorded the time, not you. Dated e-mails or messages where you sent a rough mix to someone have the same property and are frequently overlooked.
3. A sequence of drafts. Rough bounces at successive stages, showing the arrangement arriving in pieces. A progression is far more convincing than any single file, because a generated track has no plausible reason to have five earlier versions.
4. Human witnesses. A collaborator, engineer, session player, or anyone who heard it in progress. Slower to obtain, but carries real weight, particularly when a dispute escalates past an automated queue.
5. Stems and multitracks. Useful supporting material — they demonstrate structure consistent with how you say it was made — though on their own they are weaker than people assume.
6. Raw source recordings. Room tone, count-ins, false starts, talkback. Small, unglamorous, and quite persuasive, because generated audio rarely comes with the debris of a real take.

The property that matters is timing

Notice what the strongest items share: they were created while the work was happening, by systems that were not thinking about a future dispute. That is the whole game. Evidence assembled after a challenge arrives is weaker in proportion to how obviously it was assembled — and reconstructing files to look older is the single worst move available, because it converts a resolvable factual question into a credibility problem you cannot recover from.

This is also why the honest advice for someone already in a dispute is different from the advice for everyone else. If you are mid-flag right now, work with what already exists. Nothing you make today improves the record of a session that ended months ago.

What a signed record adds

The weakness of the list above is not persuasiveness — it is effort. Each item lives somewhere different, has to be gathered by hand, and has to be interpreted by whoever you send it to. That is a bad position when the person reading it has a queue and a checklist.

This is the problem Chronatum works on, and we build it, so weigh this section accordingly. It observes a DAW session while you work, structures the production activity it can see into a record, and binds that record by hash to the exact file you export — so the thing you hand over is one file with a verifiable signature rather than a folder of assorted material. The container is C2PA / Content Credentials, which is worth understanding on its own terms, including what it does not establish.

Its limits are worth stating as plainly as its use. A record covers only sessions it was actually running for; it cannot describe work that happened before it was installed. It reports what was observable inside one DAW session, which means audio brought in from elsewhere carries no history of its own, and the record says so rather than quietly counting it. And it separates what was directly observed from what was inferred, from what you declared, from what could not be determined — leaving unknowns visible instead of resolving them into a verdict. It is a better-organised, harder-to-dispute version of the evidence above, not a different kind of thing.

Anyone can check a proof for free without an account, which matters mainly because it means the person evaluating your claim does not have to trust us or install anything.

A setup that costs almost nothing

Whatever tools you use, these habits do most of the work and take no ongoing effort once configured:

• Put your projects folder somewhere with automatic version history and leave it alone.
• Save incrementally rather than overwriting one session file.
• Bounce a rough mix at each real milestone and keep it, even when it is bad — especially when it is bad.
• Keep receipts and licences for samples, presets, and any AI tool you legitimately used.
• Send rough mixes to someone. A dated message elsewhere is independent corroboration you get for free.
• Declare AI components accurately at release — what that involves is narrower than most people expect.

None of this is about anticipating an accusation. It is the same discipline that lets you recover a session you broke, or find the version that was better two weeks ago. The evidentiary value is a side effect of working carefully, which is why it is worth doing before you have a reason to.