Editorial review
The review queue, the approve / reject flow, and the lifecycle states a draft moves through.
Editorial review is where reviewers — editors-in-chief, managing editors, desk editors with review permission — do the call.
Who can review
Review-capable roles
- Owner and Admin — can review and approve everything.
- Member with editorial assignment — can review drafts they own or are routed.
- Member without editorial assignment — can submit drafts but can't approve.
- Viewer — read-only, no review actions.
See Roles overview for the full matrix.
Where review happens
Two surfaces:
Editorial board
`/app/editorial` — the queue of drafts in_review across the newsroom.
Review workspace
`/app/editorial/[storyId]` — the per-draft review page.
The review flow
Open the editorial board
Navigate to /app/editorial. Filter by priority, owner, sensitivity, freshness, or category.
Pick a draft
Click into a draft to open its review workspace. Read the localised article, the source list, the conflict markers, the confidence read.
Inspect provenance
Click any claim to see which source supports it. Sensitive-flagged drafts always show the reasoning for the flag.
Edit if needed
You can edit headline and body inline — useful for tightening a lede or fixing a phrase.
Approve, reject, or annotate
- Approve moves the draft to
published. - Reject sends it back to the author with notes.
- Annotate without state change leaves a comment for the author.
Lifecycle states
Prop
Type
Allowed transitions:
draft → in_review(author submits)in_review → published(reviewer approves)in_review → rejected(reviewer rejects)rejected → in_review(author revises and resubmits)
No reverting from published
A published draft cannot be moved back to in_review. Approve carefully — if a correction is needed, treat it as a new revision rather than reverting state.
Sensitive-topic review
Sensitive coverage always reviews
Sensitive-flagged drafts cannot bypass review, regardless of confidence. The flag is set when the cluster involves:
- deaths or casualties,
- political figures,
- legal proceedings,
- medical or health claims,
- armed conflict.
There is no override toggle. This is by design. See Sensitive coverage.
Reading the signals as a reviewer
Confidence
Weighted from source count, source tier, agreement, and recency. Low confidence = slow down, look at conflicts.
Conflict markers
When sources disagree, both versions are visible. Decide which to publish; the AI does not decide.
Source tier
Wire / Tier 1 / Tier 2 / Tier 3. Heavier tiers carry more weight in the confidence read.
Provenance per claim
Each substantive claim ties back to a source. If a claim has no provenance, that is itself a signal.
Audit trail
Every review action is recorded:
- who approved or rejected,
- what they changed (if anything),
- annotations they added,
- timestamps.
Visible in the timeline rail of the review workspace. Useful when something goes wrong and you need to reconstruct the call.