Review

You write documents directly. Knowledge is often distilled from somewhere else: a conclusion in a long session, a point repeatedly referenced in a document, or a convention found in a repository.

Before distilled knowledge becomes official, it enters Review as a candidate.

Why review matters

A knowledge base is useful only while you trust it. If unverified claims enter freely, both you and your assistants start second-guessing every result.

Distillation can be automated; adoption still requires your decision.

The review desk

Each candidate shows:

• Content — the proposed knowledge;
• Source — where it came from, linked back to the original;
• Rationale — why the item appears worth keeping.

You can:

Approve — make it official, searchable knowledge available to assistants.

Edit and approve — correct wording before adopting it.

Reject — keep it out of the knowledge base.

Defer — leave it pending when you cannot decide yet.

After approval

Approved knowledge appears in search results and becomes available to connected assistants.

If an approved item later proves wrong, withdraw it. It disappears from active use but remains recorded and can be restored if needed.

Pending review indicators

When candidates are waiting, Deerlight shows their count in navigation and the page header. The count stays until the candidates are dealt with.