> ## Documentation Index > Fetch the complete documentation index at: https://docs.getvector.me/llms.txt > Use this file to discover all available pages before exploring further. # Candidates > Review what research found before it becomes a feature: attach each candidate to an existing feature, turn it into a new one, or dismiss it. Research finds far more than you'd want to keep: the same capability under five different names, marketing phrases that aren't features at all, and genuinely new things your catalog doesn't have yet. None of it goes straight into your [catalog](/features/features-in-domain). It lands in **Candidates** first and waits for you to decide — so nothing becomes a feature without you knowing. A candidate is what a source said, not something you own yet. It has no status of yours and no place in the comparison. Once you decide what it is, it's gone from the inbox and its content moves wherever you sent it. The goal is an empty inbox. Open it from **Features** → **Candidates** in the left menu. ## Know where candidates come from Vector creates candidates in the background from two sources: * **✦ Feature scan** results in a [competitor's dossier](/competitors/competitor-dossier) — each competitor's own feature list, as they present it * [Meeting analysis](/features/meeting-analysis) insights that didn't match any existing feature When several sources mention the same thing, they become **mentions** inside one candidate rather than separate cards. Each card shows the candidate's name, how strong the evidence behind it is, how many mentions it has, and where they came from — vendor docs, a changelog, a pricing page, a public review, or something a customer said. **▸ N more mentions** expands the rest. ## Decide what a candidate is Every card asks **What do you want to do with it?** There are three answers. Choose that feature and click **Attach to**. The candidate's mentions become evidence on it — competitor evidence, or customer voice if it came from a meeting. No new feature is created. Click **Create a new feature**. Correct the name first if you need to — candidates from meetings often arrive as a raw quote. The new feature gets its own [dossier](/features/feature-dossier), and the candidate's mentions become its first evidence. A slogan, a pricing term, something that isn't a capability: click **Not a feature**. It's dismissed and won't come back. ## Let AI find the matching feature Most candidates turn out to be something you already have under another name. **Suggest matches** looks for the existing feature each candidate might belong to: the card then says **looks like** that feature, and **Attach to** points at it. A suggestion only pre-fills the choice — you still confirm it or pick something else. ## Work through the inbox faster * **Take one part of your product at a time.** The inbox is grouped by cluster, largest group first. If a candidate sits in the wrong cluster, move it with the **Cluster** picker without resolving it. * **Narrow it down.** Filter by cluster or by competitor, or search the names, notes and quotes. * **Work from the product map.** A cluster's page on the [product map](/features/product-map) lists the candidates waiting in it, with the same actions. ## Merge candidates that say the same thing Two candidates can describe the same thing when they came from different sources at different times. [Possible duplicates](/features/duplicates-and-merging) finds them, and merging two candidates combines their mentions into one card — so you decide once instead of twice. ## Where to go next What a candidate becomes. Where features and candidates are sorted by cluster.