Reviewing & promoting knowledge

Available today

Everything the connector captures is born pending. Review is where a human turns pending evidence into trusted knowledge. It is the single most important gesture in the product, because it is the control on what the agent is allowed to treat as true.

Why the gate exists

Recall the trust model: only human-approved knowledge grounds injection and answers. Review is that approval. Without it, the agent’s guesses would flow straight into what the next session sees, and the whole guarantee (“every trusted claim was accepted by a human”) would be a lie. So the gate is not bureaucracy; it is the feature.

The review inbox

mla review opens the pending items for this workspace. You look at what the connector captured and decide, at the bundle level, what to accept and what to reject. The same review can be done in the Console for deep work. Accepting is a deliberate human act; rejecting is cheap, so you can be generous about what gets captured and strict about what gets trusted.

mla review          # open the pending review inbox
mla review <id>     # review a specific item

Two axes you are reviewing

Meetless Agent tracks two separate kinds of “reviewed,” and it is worth knowing the difference:

• Trust (should this be believed?). A captured document or claim is pending until you accept it. Accepting it lets the workspace ground on it.
• Relationships (how does this connect?). The typed edges between claims (SUPERSEDES, CONTRADICTS, REFINES, REFERENCES) are reviewed on their own track. mla graph pending lists relationship candidates and mla graph review records a verdict on them.

Note

Keeping the two axes separate is deliberate. “This document is trustworthy” and “this document supersedes that one” are different judgments, and Meetless Agent does not force you to make them together.

What changes when you accept

The moment you accept a piece of knowledge, it crosses from evidence into the trusted corpus. From then on:

• Injection can place it in the agent’s context.
• Ask can ground answers on it and cite it.
• It participates in contradiction detection against future changes.

Until you accept it, none of that happens. That is the whole point: the corpus grows only through your decision, so you can trust everything in it.

A note on personal knowledge

Some captured documents start as your own shadow knowledge before they are shared into the workspace. mla kb promote <doc-id> flips such a document from shadow to live so the workspace grounds on it; the reject form leaves it as your private draft. This lets you stage knowledge privately, then promote it when it is ready.