课程大纲
学院模块 3 · Entities up close
Entity profiles: one thing, fully known
你将学到
- You can open any entity's 360° page from the table, search, chat links, other profiles, or a case
- You can read the header at a glance: note count, first-observed date, type badge, description, and AI summary
- You can check any connection against the exact sentence in your notes, then retract a wrong one for good
- You can split a bad alias merge and hand the entity off to the graph canvas or a case file
Six months of notes mention Meridian. Is it a project? A client? It shows up in meeting notes, code reviews, and a budget draft. A new teammate asks you to brief them on it in one minute. Could you? The answer is already in your workspace, just scattered across dozens of files.
The entity 360° page pulls it all into one tab: what kind of thing it is, an AI one-line summary, its connections ranked by importance and backed by the exact sentences that state them, its aliases, and every note that mentions it. This lesson walks the page top to bottom, then shows the two hand-offs that turn it into a starting point.
Six doors into one page
- Click any row in the entity table (the tab titled "Entities").
- Click a connection chip on another entity's page. Profiles chain right into each other.
- Click an entity result under "Concepts" in global search (⌘⇧F).
- Click an entity reference in a chat answer.
- Click a new-entity chip on the "While you were away" digest card on the graph canvas.
- Click a pinned entity chip inside a Case file (a container for one investigation).
Profiles are ordinary tabs, so you can line several up side by side. Reopen the same entity and it just focuses the tab you already have. Pages dedupe by normalized label, and every alias resolves to the same canonical entity: open "Bob Smith" or "Robert Smith" and you land on one page.
Reading the header
| Element | What it tells you |
|---|---|
| "Appears in {n} notes" | The note count across the whole merge cluster, every spelling folded into this one entity. It's this entity's footprint in your workspace. |
| "First observed about {n} days ago" | When the change ledger, the app's running record of graph changes, first saw the entity. Older entities read "Observed before change tracking began". |
| Type badge (colored pill) | One of the 11 entity kinds. Click it to set your own. An untyped entity shows a dashed "Set type…" pill instead of guessing. |
| Description | A one-line description the extractor pulled from your notes, when it caught one. |
| One-line summary | One AI-written sentence that sums the entity up. It reads "Summarizing…" while it generates. |
- 360° page
- The full-tab profile of one entity: kind, summary, connections, aliases, and source notes in one place.
- merge cluster
- Near-duplicate labels that entity resolution folded into one entity. The other spellings show up under "Also known as".
- source sentence
- The exact sentence in your note that states a relation, shown when you hover a connection chip.
- retraction
- A wrong edge removed for good, recorded so re-extraction can't bring it back.
- first observed
- When the change ledger first recorded the entity, i.e. the first time it was seen. That can fall long after the file was created.
Connections and their sources
The "Connections" section groups neighbors by direction and relation: a caption like → founded or ← broader, then one chip per neighbor. Once the full profile loads, each group ranks its neighbors by how central they are, so the ones that matter most rise to the top. A group shows up to 6 chips, then "more…", and clicking a chip opens that entity's page.
Hover any chip and it fetches the source sentence. The tooltip shows "Reading the note…", then the exact sentence in quotes with a caption like "{note title} · line {n}".
It matches against each note's current text, never a stale copy from extraction day. The quote you see is the sentence that stands in your workspace right now. If the note changed and no sentence states the relation anymore, the tooltip says it couldn't find one.
Retract a wrong connection
Hover the chip
An unlink icon appears on the connection chip. Its tooltip flags the connection as wrong and offers to remove it.
Confirm inline
Click the icon. The chip turns into an inline confirm: "Remove this connection?", with a Remove and a Cancel button.
Remove for good
Click Remove. The edge is retracted and the profile refreshes. The retraction is on the record, so re-extraction won't bring the edge back.
快速自测
You remove a wrong connection from an entity page, then later re-extract your notes. What happens to the removed edge?
Aliases, source notes, and related code
- "Also known as" lists the merge cluster's other spellings as alias pills. Each one has an unlink button, tooltip "Not the same as {name}? Split them apart", that rejects the merge and rebuilds entity resolution.
- "Found in" shows up to 8 notes that mention the entity. Click one to open it in the editor.
- "Related code" shows up only when the code index and AI code features are on. It lists up to 8 source-code symbols that relate to the entity in meaning, and each one opens its file.
- "In your words" appears when statement extraction has captured your own decisions, questions, and claims about this entity, word for word. The statements lesson covers it in depth.
Hand-offs: to the graph, into a case
Two header actions turn a profile into a starting point. "Open on the graph" jumps to the graph canvas with this entity's node in isolation. It's the same knowledge, drawn as a picture.
"Add to case" opens a menu. "New case from this entity" creates a case named after the entity and opens it. Pick an existing case instead and the entity is pinned there, with an "Added {label} to the case" toast to confirm.
快速自测
You open the page for "Bob Smith". Later you click a connection chip labeled "Robert Smith", a name that entity resolution had already merged into the same entity. What happens?
The type badge you clicked in the header gets the next lesson to itself: the 11 entity kinds and the collections they unlock.