The Review Queue
Approve or reject AI-proposed wiki updates before they go live.
The Review Queue is a human-in-the-loop checkpoint. When Panacea's AI agent identifies that a wiki entry might be outdated or incorrect — based on new information encountered in customer queries — it proposes an update. That proposal lands in the Review Queue rather than being applied automatically.
Opening the Review Queue
Click Review in the left sidebar. The page lists all pending proposals filtered by status.
Understanding review items
Each item shows:
- Entry path — Which wiki entry the AI wants to change
- Status —
pending,approved,rejected, ormodified - Review type —
quality(AI-detected quality issue) orpublic_approval(entry needs approval before becoming public) - Confidence — How confident the AI is in its proposed change (colour-coded)
- Proposed content — A snippet of the proposed new text
- Created — When the proposal was submitted
Filtering by status
Use the status tabs at the top of the page:
| Tab | Shows |
|---|---|
| Pending | Items awaiting human review (default view) |
| Approved | Items you have approved and applied to the wiki |
| Rejected | Items you have rejected and dismissed |
| Modified | Items you approved after editing the AI's suggested content |
| All | Every item regardless of status |
Approving a proposal
- Click an item to open its detail view.
- Review the proposed change — the diff shows what the AI wants to add, change, or remove.
- If you agree, click Approve. The wiki entry is updated immediately.
Tip: You can edit the proposed content before approving if you want to refine the AI's suggestion.
Rejecting a proposal
If the proposed change is wrong, irrelevant, or based on bad information:
- Click Reject on the item.
- The item moves to the Rejected tab and the wiki entry remains unchanged.
Rejected proposals help train the system — the AI learns not to propose similar changes for that entry in the future.
When are proposals created?
Proposals are created automatically when:
- A customer query exposes new information that contradicts an existing entry
- The agent encounters a pattern of questions it couldn't answer well (wiki gap)
- A connected Source sync brings in updated content that partially overlaps with existing entries
Best practices
- Review daily — The queue is most useful when actioned promptly. Stale proposals become less relevant over time.
- Don't approve blindly — Read the proposed change carefully. The AI can hallucinate. Your approval is the safety net.
- Use rejections as feedback — If you see many proposals for the same entry, the underlying entry may need a full rewrite.
- Prioritise high-confidence proposals — Items above 80% confidence are usually safe to approve quickly. Focus manual review on lower-confidence items.