Interveil Labs
Frameworks · Method

Provenance & Integrity.

The discipline under everything we build — and the layer we offer others. Verify the provenance, grade the claim, anchor the record, keep the human the author. Governance you can deploy, not just admire.

The principle

Restore the signal. Never falsify the record.

Our method is older than our company. It comes from forensic audio — the craft of pulling voice and music out of a damaged, noisy recording without inventing anything that wasn't there. Clean the noise; keep the truth. If a sound is gone, you say it's gone; you don't fabricate it and call the fake a restoration.

That one rule, carried into AI, becomes a whole posture. A system can enrich, clarify, and contextualize — but it may never quietly substitute a convincing fabrication for the real thing. Plausibility is not provenance. Fluency is not truth. The record is sovereign.

We can verify the provenance. We cannot manufacture the experience — and saying the second part is what makes the first part believable.

The four practices

What integrity looks like, operationally

01 · Provenance by default

Show the source

Every claim carries where it came from and how certain it is. Interpretation is labeled as interpretation. Nothing arrives as an unsourced verdict.

02 · Grade the claim

State the confidence

A published Risk & Credibility Index — Established · Attested · Contested · Speculative — so a reader knows how much weight a statement can bear before building on it.

03 · Anchor the record

Tamper-evident, correction-friendly

A record can be fingerprinted so alteration is detectable. Corrections aren't hidden — a new anchor is made and the old one kept. The scar is part of the provenance.

04 · The human stays the author

AI proposes; people ratify

Autonomy is deny-by-default. The system offers; a person, seeing the evidence and its grade, decides. The instrument never quietly becomes the author.

EstablishedAttestedContestedSpeculative

Every empirical claim we publish carries one of these. You'll see the same four grades on our essays and papers — the method applied to ourselves first.


The test

Never "trust me." Always "here, compare."

A claim without a comparison the audience can make for themselves is just an advertisement. Wherever we can, we show the before and the after — the degraded next to the restored — and let the trained eye or ear be the proof. Verification you can perform beats authority you're asked to accept.


What we refuse

The lines that aren't for sale

We will not fabricate a source or a certainty; we will not let a metaphor be upgraded to a mechanism; we will not profile anyone covertly or dress engagement-harvesting as help; we will not publish an unverified claim as settled truth. These are constitutional, not commercial — the perimeter that lets the rest of the work be trusted.

The standards we align to are public: content provenance via C2PA, and risk practice via the NIST AI Risk Management Framework. Where our discipline goes further than the standard, we say so — and show our work.


Further reading in the Journal: AI and Us, Between Veil and Vision, and Forensic Provenance for Synthetic Worlds. The framework this method serves: Chronoaesthesis.