Audit Cockpit¶
tibet-audit is the local reconstruction cockpit for an AInternet node.
It does not score an actor. It reads evidence that already exists on your machine and shows which route, policy, material and state can be reconstructed.
What It Answers¶
Run the cockpit when you need to know:
what happened?
which actor or runtime was involved?
which route posture was proven?
which policy allowed, denied or darkened the action?
which object, model, tool or package was touched?
which evidence is missing?
The output is meant for operators first: a terminal view you can read during an incident, a lab run, a compliance check or a local node setup.
Run It¶
For a repo or node that carries a profile map:
The command should not phone a central service by default. If a report endpoint is used, it must be explicitly configured by the operator.
The Six Panes¶
| Pane | Meaning |
|---|---|
| Pulse | latest events, tail-style flow and warnings |
| Machine and Hardware Evidence | local compute evidence such as CPU capability and runtime posture |
| Surfaces and Enclaves | SSM cards, sealed carriers, suspicious surfaces and enclave evidence |
| Evidence Chains | correlated story from parent event to policy, route, result and follow-up |
| Stack | packages present for the selected profile, such as ainternet[node] |
| Next Actions | missing packages, policy gaps or operator follow-up |
The header folds evidence lanes into a single route posture:
That is not a grade. It is the weakest proven route across the current evidence tree.
What It Reads¶
tibet-audit sits above the evidence spine:
| Source | Role |
|---|---|
tibet-tail |
live and recent event flow |
tibet-pol |
policy verdicts, drift and runtime checks |
| TIBET receipts | causal action chain |
| MUX | route posture, named surfaces and null routes |
| SNAFT / Cortex | consent and policy decisions |
| Airlock | quarantine, run, release and darkening evidence |
| Wayback | state seal, diff and restore point |
| SBOM | software materials |
| AI-SBOM | model, prompt, tool and runtime materials |
| CBOM | capability and authority inventory |
| SSM | surface hints checked against sealed manifest truth |
Magic Bytes, Not Filenames¶
The cockpit follows the same rule as transfer carriers:
A file named .tza is not trusted because of its extension. A sealed carrier is recognized by structure, magic bytes, signature and manifest. If the name and the manifest disagree, the manifest wins and the surface drift is evidence.
Good Output¶
Good output is not always green. A useful cockpit can say:
route darkened
consent expired
external tool call denied
zombie .tza surface rejected
hardware cadence not proven
package missing for this profile
That is the point. Audit is not a decorative report. It is how a local actor knows what it can prove.
Related¶
Machine-Readable¶
https://ainternet.org/api.json— audit, route and carrier inspection verbshttps://ainternet.org/resources.json— cockpit, SSM and evidence references