Reviewing & accepting proposals

The curator does most of its filing silently. The exceptions — anything that would throw away or restructure what you know — it brings to you as a proposal. Working through that queue is the one recurring job The Librarian asks of you, and it usually takes a minute or two. This guide explains how to do it well.

Why some changes need you and others don’t

The curator follows one simple rule. Safe, additive operations — creating a memory, updating one, or merging near-duplicates — it applies on its own when it is confident enough. The two operations that lose or reshape information — archiving a memory and splitting one apart — are never done automatically; they always become proposals. So does any change to a memory you have marked as needing approval. The queue, in other words, is exactly the set of changes worth a human glance, and nothing else.

The workflow

1. Open Proposals in the dashboard. Each card is one suggested change.
2. Read the reasoning and the diff. The card tells you what action it is (Update, Merge, Split, Archive, New), why the curator suggests it, and shows the before-and-after. For a merge you see every source and the combined result; for a split, the original and its pieces.
3. Decide. If the change is right, press Approve — the button names the exact consequence (Approve & archive, Approve & merge, …), so there is no guessing. If it is wrong, press Reject and the memory is left untouched.
4. Tidy up a split. When a split has produced good replacement memories, use the Archive original button beneath them to retire the source.

How to judge a proposal

• Archive proposals — ask “is this really stale or wrong?” Archiving is reversible (the memory moves to the Archive, not oblivion), so you can approve with low risk and undo later.
• Merge proposals — check the merged text did not drop a nuance from one of the sources. If it reads worse than the originals, reject.
• Split proposals — confirm each piece stands on its own and nothing was lost in the division.
• When unsure, reject. Rejecting is always safe — it changes nothing. A fact you reject today can be proposed again later, or you can fix it yourself on the Memories page.

Teaching the curator from what you see

If you keep rejecting the same kind of suggestion, that is a signal. Open the Curator page and add a line of guidance (“don’t merge security notes”, say). The curator’s next run respects it. Tuning by reacting to real proposals — not by chasing a metric — is exactly how the curator is meant to improve; see Configuring the curator.

Related

• Proposals page — the screen itself.
• Flagged — a related queue, for memories an agent reported as wrong.