The review queue

The queue is where an LLM's suggestion becomes your organization's rule, or does not. It is the single most consequential screen in the product, and it takes about thirty seconds per item once you know what to look at.

What a suggestion is

A candidate decision: a statement Memnox believes your organization settled, extracted from a window of source events.

It carries:

• the statement itself, in one sentence;
• the source events it came from, each with a permalink;
• the source type and the author of each;
• whether any of that evidence was tainted.

How to read one

1. 1

   Read the evidence, not the summary

   Open the source events and read the original thread. The summary is a compression, and compressions lose the qualifier that made the decision conditional.

2. 2

   Check the taint badge

   A suggestion drawn from tainted evidence, a forwarded document, a comment from outside the workspace, is not necessarily wrong, but it is not something your team said either. It deserves a higher bar.

3. 3

   Ask whether it is still true

   Extraction runs over a window of history. A decision that was correct in March may have been superseded in June by a thread the window did not cover.

4. 4

   Decide

   Approve, and it becomes a constraint with your name on it. Reject, and it costs nothing but a click.

What approving does

1. The decision is recorded with its provenance, the source type, the source reference, and your name as the reviewer.
2. It reaches the runtime's decision memory, where it can escalate any future action that contradicts it.
3. It becomes a node in the organizational graph, with an evidenced_by edge to the source event.

Rejecting is not a failure

A rejected suggestion is signal, not waste. It tells the system what your organization does not treat as a decision, a preference, a one-off, somebody thinking out loud.

A queue where everything gets approved is a queue nobody is reading.

Running extraction

Open Decisions and choose Run extraction. It reads a window of source events and fills the queue with candidates; on Pro and above it can also run on a schedule, so the queue is there when you sit down rather than when you remember.

Each run is counted against your plan's extraction quota. Extraction is also BYOK: without a model credential it answers 503, and every other part of the product carries on unaffected.

When suggestions look wrong

Three usual causes, in order of frequency:

Notification

A suggestion.queued webhook subscription fires when something lands in the queue, so reviewing does not depend on somebody remembering to look. See Notifications.