Story workspace
Read the localized draft, refine it, and route it for review.
The story workspace at /app/stories/[storyId] is where you read, refine, and route a draft. The localised article is the centerpiece — everything else supports it.
Layout
Article body (centre)
The localised headline and body HermesAI drafted. This is what you're working on.
Provenance rail
Source list, claim-level attribution, conflict markers, and confidence read.
Timeline rail
When sources arrived, when clustering happened, when compose ran, when reviews fired.
Compose / refine
Modes inside the same workspace — not a separate area. You stay on the story while editing.
What you can do
The default mode. The localised draft is in front of you, with provenance and timeline visible. Read first — the AI does some work; the editor does the call.
If the draft is still placeholder text (the cluster is fresh and synthesis hasn't run yet), use Compose to generate the first localised article.
Compose options include format (Quick Brief, Full Story, Broadcast Script, Digest Block) — choose the one that fits where the story is going.
Once you have a draft, Refinement chat lets you ask for changes in plain language: shorten, sharpen the lede, add a paragraph on a specific angle, swap an attribution, fix tone.
Refinement keeps source grounding — it can't ask you to invent facts.
When the draft is ready, Submit for review sends it to the editorial board with optional priority and owner. See Editorial review.
The compose flow
Choose a format
Quick Brief, Full Story, Broadcast Script, or Digest Block. Format shapes the length and structure HermesAI generates.
Optional: tone, length, voice
Tone hints (formal, plain, breaking-news clipped). Length bounds. Voice notes. These are guidance — the AI still respects source grounding.
Run compose
HermesAI generates the localised article. Sources come in attached at the claim level; confidence reads update.
Iterate or submit
Refine via chat, run compose again with different options, or submit for review.
The refinement chat
What the chat will and won't do
It will: rewrite paragraphs, sharpen ledes, change tone, add sourced angles, fix attribution, suggest cuts.
It won't: invent facts, claim things sources don't support, smooth over conflicts that the cluster shows, override sensitive-topic review gating.
Provenance — what's attached to each draft
Source list
Every outlet that covered the cluster, sorted by tier and recency.
Claim-level attribution
Which source supports which claim. Surfaced inline so review is fast.
Conflict markers
Where sources disagree, with both versions visible.
Confidence read
Updated as compose / refine runs. Shows what changed.
Sensitive flag
If set, you can read and refine — but not publish from here. Submit for review.
Submitting for review
Click Submit for review
The draft moves to in_review status. It appears in the editorial board for reviewers.
Add owner + priority (optional)
Hand off to a specific reviewer or flag urgency. Useful when the desk is busy.
Add notes for reviewers (optional)
Context that the source list doesn't capture — local angle, breaking confirmation, why timing matters.
Lifecycle states
Prop
Type
Published is a one-way door
A published draft cannot be moved back to review. If something is wrong, treat the next iteration as a new revision rather than reverting.