01 / Intake
What the sender is asked
Sending a tip is a conversation. Leed asks one question at a time and the sender answers or skips. No account is required and no name is requested up front.
The order is the point. A blank text box returns two sentences nobody can check. Asking what happened, then what evidence exists, then what it risks, returns a lead that already knows its own weaknesses.
- The story
- What happened, a headline, why it matters now, who is affected, where.
- The proof
- What evidence exists, what a reporter could independently check, whether anyone will speak on record, and whether it has been shared elsewhere.
- The risk
- What is sensitive about it, anything a reporter should know before making contact, and how the sender wants to be reached, if at all.
02 / The read
What the AI does, and what it does not
Claude reads each tip and returns a structured brief: the claim, the way to verify it, and the exposure it creates for the sender. It assigns a signal tier in words, never a numeric score, because a number invites false precision about a stranger’s account.
- What’s claimed
- The assertion, stated plainly.
- How to verify
- The public record, dataset, or witness that would settle it.
- Risk to source
- What telling you may cost the person who did.
The tip is saved the moment it is sent. The read runs afterwards, in the background, so a slow or failing model can never cost you a lead. If the read fails entirely the lead is still there and still findable.
Anything the read flags as risky, along with spam and PR, waits for a person before it reaches any journalist.
03 / Clusters
How separate tips become one story
A single tip is an anecdote. Leed files each lead into a signal cluster: the group of tips describing the same underlying story. Assignment runs after the read, using the beat and region the brief identified.
This is the part a newsroom cannot do by hand. Nine people noticing the same thing over six weeks, arriving through four different inboxes, look like nine unrelated messages. Grouped, they are a story with corroboration already attached, visible before it breaks.
- Tier
- Hot, Strong, or Watch. A word, never a number.
- Weight
- How many stories the cluster holds, and how many arrived this week.
- Placement
- The beats and region it belongs to, so it reaches the right reporter.
04 / Access
Who can see what
Nothing derived from a real tip is public. Verified journalists and admins see leads, clusters, and the person behind each one. Everyone else, including signed-in readers, learns only that a pattern exists.
There are no anonymous read grants on leads or clusters. The boundary lives in the schema, not only in the interface, so a bug in a page cannot leak what the database refuses to hand out.
- Automatic
- Journalists at recognised newsroom domains, verified on sign-in.
- Reviewed
- Freelance and independent reporters, checked by a person first.
- Never
- Freemail addresses, for a press account.
Known limitation, stated plainly: visibility today is platform-wide among verified press. Org-scoped workspaces, where each newsroom sees only its own intake, are the next thing we build for a committed pilot.
05 / Consent
How contact happens
A journalist who wants to talk sends a contact request. It lands on the source’s own dashboard, where they accept or decline. Nothing reaches them before that point.
A decline tells the reporter the request was not accepted and nothing else. It reveals no name, no reason, and no signal about who the sender is.
This is the trade the whole product is built around. A source who assumes contact is automatic sends less and hedges what they do send. A source who knows the yes belongs to them sends the document.
06 / Desk Agent
The optional agent
Desk Agent is a premium add-on for verified journalists. It walks the clusters on your subjects, ranks what moved since yesterday, and leaves one brief: what to open first, why it matters this morning, and what is still forming.
It never contacts a source, never acts on your behalf, and never changes platform state. Its model input excludes source identity entirely, so it could not reveal a name if asked to.
07 / Questions
Questions people actually ask
- Do you publish anything?
- No. Leed is a private desk. The one exception is deliberate: a sender polling the status of their own tip receives progress booleans and a tier word, nothing else.
- Does the AI decide what gets seen?
- It structures and groups; it does not gatekeep. Every lead a journalist can reach stays reachable through search and the board regardless of tier.
- What if the read is wrong about a tip?
- The tip is stored exactly as sent and the brief sits beside it rather than replacing it. A journalist always reads the original.
- Can a source change their mind?
- Yes. Contact preferences travel with the lead, and a pending request can be declined at any point.
- What does a newsroom actually get?
- One queue instead of four inboxes, every tip pre-read and structured, patterns surfaced before they are obvious, and a source-safety story you can put in front of the people you are asking to take a risk.