Throw your user feedback in.
It comes back sorted.

A backlog is a plan: it holds work that is committed and ordered. An idea board holds everything you have heard but not arbitrated yet. Mixing the two is what makes backlogs unreadable, because a board where most lines will never be built stops being a plan. The Radar is the idea board, your backlog stays the plan, and a single action moves an item from one to the other in either direction.

Yes. The Radar is an extra tab on the public board you already have, on the same slug and the same host, so your users do not learn a second address. They submit ideas, upvote the ones they want, comment on them and follow them. The three public tabs are Ideas, In progress and Shipped, with no project management jargon on the visitor's side.

Two different operations. When a signal arrives, a short model call attaches it to the closest existing idea or opens a new one, and attaches that idea to a theme. That is cheap and runs once per signal. Separately, a recluster pass reads the whole corpus and redraws the themes, which is the expensive call and the one you trigger on purpose. A theme you pinned is never dissolved by a recluster, so your manual tidying survives.

Each idea's summary gets an embedding, and similarity is computed between them. Pairs above the threshold are surfaced as duplicate candidates. They are never merged automatically: a merge is always a human action or an agent proposal you accept. Automatic clustering is a proposed arrangement, not a decision.

Yes, one vote per person per idea, with the button on the card and on the idea's own page. Votes and signal counts answer different questions: the signal count is how many people took the trouble to describe the problem to you, the vote count is how many recognised themselves in it once it was written down. Both are shown.

It gives you the honest version of one. Visitors see the ideas you have qualified, what is in progress and what has shipped, on one page. What they do not see is your unfiltered inbox, because an idea posted by a visitor stays private to its author and to you until you qualify it.

It answers a question the theme view cannot: what is worth doing right now. Dragging an idea inside the 2x2 writes its impact and effort onto the record, so the position is data other views and the agent can read, not a drawing that lives in one screen.

It hangs your themes off the objective you actually care about this quarter, and ideas off the themes. Its real value is the unattached column: a theme connected to no objective is visible, so it gets connected, parked on purpose, or dropped. That column is what keeps an idea backlog from quietly becoming a cemetery.

Yes, and that is half the point. The ticket leaves the board, the idea stays visible and votable, and its reporter and followers become signals on the idea so nobody is dropped. Everyone concerned is told what happened and where their request now lives.

Yes. Select several tickets, or tick the whole visible column, and send the selection to the Radar. Because each public ticket has a real person behind it, the confirmation screen counts people rather than tickets, everyone gets one message instead of one per ticket, and the whole batch stays cancellable for ten minutes before anything is sent.

No. It reads the entire corpus, reclusters, proposes merges, scores impact and effort and argues for the next ticket with the verbatims that justify it. Creating a ticket and replying to a reporter are outbound actions towards real people, so they stay yours. You can also run it as a Brainstormer, a UX Researcher or a business analyst instead.

Yes to both. Clustering runs on your own key if you bring one, and on the included allowance otherwise. On mobile you get the same data and the same actions: inbox, themes sorted by heat, an idea's verbatims, merge, promote, vote and comment. The matrix and the tree become a sorted list, because a 2x2 is not something you drag with a thumb on a 390 pixel screen.