Provenance Audit Trail

An append-only audit trail records every decision the AI stack informed and the data behind it, giving the fabric a defensible history.

An immutable record

A safety-relevant control system must be able to reconstruct, after the fact, what it knew and why it acted. The provenance audit trail is an append-only record linking each decision the AI stack informed — a twin forecast, an anomaly alert, an MPC actuation recommendation, a human override — to the exact features, lineage, and model versions behind it.

What is recorded

• The decision or recommendation and its timestamp.
• The feature values and their lineage at decision time.
• The model versions that produced the recommendation.
• The human-in-the-loop action, if any, and who took it.
• The machine and configuration context.

Append-only and tamper-evident

The trail is append-only and tamper-evident, so a record cannot be altered after an event without leaving a trace. This is the data-side complement to the security layer's zero-trust and unidirectional-diode design: the fabric can prove not only what happened but that the record of it is intact.

Uses of the trail

The audit trail supports forensic reconstruction of any event on the breeder or the burner, regulatory and safety review, and MLOps post-mortems that ask whether a model behaved as trained. It rests on lineage and governance, and it is the reason the fabric can stand behind every number the twin ever acted on. It is a design and operational discipline for machines whose FOAK operation is expected near 2030.