MIT Network Audit docs
How it works

What the audit log can and cannot show

4 min read Odoo 18.0

This page lists the design rules of MIT Network Audit, what each rule gives you, and where each rule stops. Read it before you use the log as evidence.

Who this is for: security officers, auditors and administrators who decide how far to rely on the log. You need: nothing beyond the earlier pages. The statements are checked against the Odoo 18 copy of the module.

Before you begin

  1. Know the short answer. The log can show that the rows it holds were not edited, and which module made each recorded call. It cannot show that every connection was recorded, and it cannot show what a call contained.
  2. Read Security and capture limits for the list of connections that are not captured.
  3. Know who you must trust. The module is open source, but you still run it inside your own Odoo and database.

What the log can show

You can use the log to... Because... Limit
Read the module's source before you install it. The module is not obfuscated. Reading the source does not check the copy you installed. Compare it with the copy you reviewed.
See metadata about the connections that were captured. Events have no body field in normal auditing. They hold destination, sizes, times and the module. Path and non-sensitive header values are stored. Strict redaction removes the path. Forensic body capture, if enabled, stores bodies.
Detect edits, inserted rows and removed rows inside the stored chain. Each row's hash includes the previous row's hash. Verify chain recomputes them. A person with full control of the database could rebuild the chain.
Detect that the newest rows were cut off. A daily signed head records how far the chain had reached. The signing key is stored in the same database. The head is signed once a day. Odoo's Verify chain does not compare it. Use the standalone verifier with the key.
Keep the module from changing rows in normal use. The model refuses write and unlink. A database trigger blocks changes to sealed rows and gates deletes. A person who controls the database role can lift the delete gate.

What the log cannot show

Limits

  • Completeness. Some traffic is never captured: WebSocket connections, connections to the same machine (including PostgreSQL), unix sockets, other processes on the server, and traffic while auditing or a channel is off. Events can be dropped when the buffer is full. A log that passes Verify chain can still be incomplete.
  • Content. The log has sizes, not content. It cannot tell you what a call carried.
  • Intent. The module name on a row is the module found on the call stack. It does not prove who intended the call.
  • Absence of calls to MIT. MIT-bound means "matches the Egress Components list". A MIT host missing from the list shows as third-party.
  • Past gaps. If auditing was off, or the module was not installed, there is no row, and no marker for the gap.

The design rules

1. Open source, not obfuscated

The module's code is readable. The project's build checks for obfuscation before release. This lets you review what the module does. It does not replace a review of your own installation.

2. No body field in normal auditing

The event type that the module records has no field for a body. The log keeps a salted fingerprint of the method, host and path, not of the content. The optional forensic body capture is a separate store with its own group, settings and logging. See Change audit settings.

3. Append-only, hash-chained, signed head

The log is built so that changes are blocked in normal use and visible afterwards. The details and limits are on Verify the log and export evidence.

The retention job is the only code that deletes rows. It never deletes at or after the latest signed head.

4. Errors in the auditor are caught

The module catches errors in capture, attribution, redaction and saving, so that they do not stop the call that is being observed. The code is written to do this, and you can read it. The effect on speed has not been measured. If something fails, the usual result is a missing row.

If installation fails, the process runs without auditing. The only sign is a warning in the Odoo log. See Install MIT Network Audit.

5. Self-egress check

The module is designed to make no outbound calls of its own, and the Trust Report counts rows that are attributed to it. The attribution code skips the module's own frames, so this count is a weak signal. Do not use it as proof that the module makes no calls.

6. The interceptor core has no Odoo imports

The capture logic in interceptor/ imports nothing from Odoo, so you can review it alone. The Odoo layer still decides what is saved, how modules are named and how classification works.

If it does not work

Next step

Security and capture limits.