Why Provenance Beats Detection

Detection asks a model to guess whether an artifact is authentic. Provenance records where the artifact came from and how it changed. The two age very differently.

A detector produces a probabilistic score from patterns it learned. That score can be useful evidence, but its reliability changes as generation techniques and training data change. The receiver must also trust the model, its version and the party operating it.

Cryptographic provenance answers a narrower, more durable question: does this file match the artifact that was recorded, and does the custody history connect it to the asserted source? Content hashes, signed custody events and anchored Merkle roots remain checkable as generation quality improves.

Provenance is not a magic authenticity label. Missing provenance does not prove that content is synthetic, and a valid chain does not prove every claim made about the content. It provides verifiable origin and integrity signals that reviewers can combine with source evidence and forensic analysis.

Noesea therefore treats detection output as evidence rather than authority. The protocol prioritizes provenance because a durable proof record should not depend on permanently winning an escalating model-versus-model contest.