Explore: Following Connections
After the first scans, you have assets, and whatever lineage, links and duplicates came with them. Explore is where you follow them. It answers what a list cannot:
- Where does this table’s data come from, and what is built on it?
- Is this the same document as that one, or does it only share a name with it?
- What else is this connected to, and is any of it already in a case?
- Nothing links this file yet: what is it like?
Explore never draws your whole workspace. A picture of ten thousand documents is a cloud of dots that answers nothing. It starts from one asset, a few, or a source’s most connected ones, and shows what is connected to them, one view at a time. You meet it where you already are: the Connections tab of every source and every asset, the overview of every glossary term, the Explore section of duplicate review, and the Explore card on the Overview.
On a source’s page
A source’s Connections tab opens on its ten most connected assets, and what they are connected to: the lineage between its tables, the pages that link its issues, the duplicates inside it and in other sources. The Find panel searches this source. Worth adding next lists the neighbours reached by the most kinds of connection.
The dropdown at the top right says which connections to keep:
| Choice | Keeps |
|---|---|
| All | Every connection |
| Same source | Only connections between assets of this source |
| Other sources | Only connections that leave it: to another source’s assets, or to objects no source has scanned yet |
An asset from another source carries that source’s name under it, so a wiki page linking a ticket reads as such at a glance. On the source’s page, Other sources starts from the assets with the most connections out of it.
Diving in, and back out
Double-click an asset, or select it and press Dive in: the canvas re-centres on it, with its own connections around it, traced the way you had chosen. A trail at the top right says where you are: the level you came from and this one, HD-214 › HD-214 runbook; the … before them lists every level further back, All of Service desk first. Click any of them to go back out to it. Open asset opens the asset’s own page.
On an asset’s page
Every asset page has a Connections tab: the same canvas, starting from that asset, with the same counters, switch, views, trace and panels, and the same way of diving in. Select the asset to see its meaning (the concepts it is about, by binding, declaration, mention or link by hand) and to link a concept by hand. For a table or a view, the column lineage panel sits under the canvas.
On a glossary term’s page
A concept’s or an entity’s overview opens on Where it is connected: the ten assets that carry it most (the most severe findings first, then the most evidence), the term itself drawn between them and ringed, the other concepts they share, and the documents those bring in. The term stays on the canvas in every view, so switching to Lineage shows how the documents about it feed each other without losing why they are there. Double-click a document to dive in; the trail leads back to the term.
In duplicate review
Duplicate review’s Explore section draws its pairs on the canvas instead of listing them: the documents in the most pairs are the starting points, each with what is the same as it around it, every line with the reason the pair screen gives (same content, repeated text · 97%, likely the same · person ×2, confirmed the same). Filters above the canvas pick which pairs:
| Filter | Picks |
|---|---|
| Pattern | One pattern; a pattern’s page links straight here with Explore |
| Kind of match | Shared values, similar spelling, identical content, repeated text |
| Score | Any, from the worth a look cutoff, or from the likely the same one |
| Review | Every pair not ruled apart, pairs not reviewed yet, or pairs confirmed the same |
| Source | Pairs with a document in one source |
The dropdown at the top right keeps pairs within one source or across two. Pairs someone ruled apart are never drawn as the same. Lineage and links between the drawn documents come along, so a copy that is explained by lineage shows it. Double-click a document to dive into everything it is connected to; the queue is still where pairs are decided.
Explore on the Overview
The Explore card on the Overview opens Explore on its own page, for questions that start from nowhere in particular. It opens on a search box, your saved views, and one table of assets worth starting from, a tab per reason:
| Tab | Lists |
|---|---|
| Results | What the search found, by meaning and by the exact words, as on the Assets page, each with the passage that matched |
| Most critical | Assets with open critical and high findings, the most critical first |
| Most lineage | Assets with the most lineage links |
| Duplicates | The biggest duplicate patterns, each with its strongest pair |
| Drawn links | The links people drew last |
Press Explore on a row, or tick several and Explore them together. Once something is on the canvas, search in the Find panel; + adds a result next to what is already there.
The view lives in the address bar, so back and forward walk your exploration, and a copied link opens exactly what you were looking at.
Saved views
When a canvas tells you something, keep it: Save view in the top bar gives it a name and notes. A saved view keeps its starting points, the view, how far and which way it traces, and where you put things: drag assets where they make sense, and Save changes keeps the layout. The menu next to it saves a copy as a new view, renames it, resets the layout, or deletes it.
Saved views belong to that page, and are listed on its start and on the Overview card. They are kept by the identity of the assets, so re-scans leave them whole; an asset deleted at its source is not drawn, and comes back if the asset does.
The views, and the guide
Explore shows one view at a time, and opens on the one your asset has most to show in:
| View | Shows | Opens when |
|---|---|---|
| Lineage | Where the data comes from and where it goes, traced as far as you ask | The asset has any lineage |
| Same and alike values | The same content, likely the same thing, and shared detected values | It has duplicates or shares values |
| All connected | Lineage, links, duplicates and shared values together | It only has plain links |
| By meaning | The concepts and people it shares, and what reads alike | Nothing links it yet |
A line at the top of the canvas, the guide, says what the view shows (Lineage: 7 upstream, 7 downstream) and which other views have something for this asset (Also: By meaning 56). One click switches.
An asset nothing links yet is not a dead end. Explore opens it by meaning, and the guide says so: these are the documents it shares concepts with or reads like, so you can link what belongs together.
Tracing lineage
The View button in the bottom left corner chooses the view, and how lineage and links are traced:
- Upstream, Both, Downstream: where the data came from, where it went, or both.
- How far: one step at a time up to ∞, as far as the lineage goes. Duplicates and shared values go at most two steps; concepts and likeness are always one, because two steps of “alike” is no longer alike.
- Kinds of connection: switch individual kinds on or off when no view fits.
On the canvas, lineage reads left to right: what feeds your starting points stands to their left, one column per step, what they feed stands to their right. A small ↑2 or ↓1 says how many steps away it is.
Reading the canvas
The canvas works as the case board’s does: the wheel zooms, a drag pans, and the row in the bottom right zooms out and in, goes back to 100%, fits everything, fits what is selected, and shows a mini-map. Right-click anything for all it offers: an asset (details, dive in, open it, open it in a new tab, add it to a case), a term (what carries it, open it in the glossary), a line (why it is there, review the pair, remove a link you drew), or the empty canvas (fit to view).
The counters in the top bar count what is on the canvas, one per kind of connection, each with the line it is drawn with. Press one to light that kind and bring it into view; the rest steps back. Press it again to see everything.
- Your starting points are ringed.
- Above them: what is the same (duplicates) and what shares values.
- Below them: the concepts and entities they share, drawn as labelled boxes between the documents, then other documents about the same things, then what reads alike, then plain references.
- Every line carries its reason. Zoom in to read it on the line; point at a line to read it at any zoom. A fainter line is a weaker score; an amber line was drawn by a person.
| Kind | Means | The line says |
|---|---|---|
| Lineage | Data flowed from one into the other, reported by the systems themselves | loaded into, transform, copy of |
| Links | References between documents a scan found (a page that links an issue, a message and its attachment), usage, and links people drew | references, or what the person wrote |
| Same as | The same content, very likely the same thing, or the same text read passage for passage | same content, likely the same · person ×2, repeated text · 97% |
| Shared values | Both contain the same detected values: a person, an IBAN, an email | shares person ×2, phone number ×1 |
| Shared concepts | Both are about the same glossary concept or entity | the concept itself, as a box |
| Alike by meaning | They read alike, compared by meaning rather than by words | 84% alike, with the closest passage |
Same as and shared values come from the same evidence as duplicate review, and carry what review decided: a pair a reviewer confirmed reads confirmed the same, one marked unsure says so, and a pair ruled apart is never drawn. Select one of those lines to see its review state, and Review this pair opens it there. Alike by meaning needs embeddings switched on.
What is already in a case
An asset that is evidence in a case wears a CASE badge; one whose findings answer a watch wears WATCH. The top bar counts both, and the Cases and watches panel lists them with links to each, so you see at once what is already being looked at and what nobody has picked up yet.
The panels
One rail on the right opens the panels, as on the case board:
| Panel | Holds |
|---|---|
| Find | One search box. What it matches on the canvas glows there and is listed first: by name, by the words of a passage, or closest by meaning (the few that stand out from the rest of the canvas). What it matches elsewhere in the workspace can be brought in. Then your starting points |
| Worth adding next | The neighbours reached by the most kinds of connection, the strongest first |
| Details | What you selected: an asset’s links with their reasons and the cases and watches it is part of, a line’s reason and score, a term’s documents |
| Cases and watches | The cases holding these assets and the watches they answer |
Click an asset in a panel’s list and its details open in that panel, with an arrow back to the list, so you never lose your place.
On a source’s or an asset’s page, double-click an asset to dive in (see above). On Explore’s own page, a double click or Add as a starting point walks on instead: its own connections come in around it, next to what is there, and Start over from this makes it the only starting point. All to a case sends everything on the canvas to a case as evidence; Add to a case sends one asset.
Your own links
You often know something the data does not: that the photos were taken for this complaint, that this works order answers that survey. Pull a line from the edge of one asset to another, and say what the link means:
| Meaning | Where it shows |
|---|---|
| Related to | As a plain link, no direction |
| Same as | With the duplicates |
| Feeds | As lineage, in the lineage view too |
| References | As a link, one way |
| Something else | As a link, named by you |
Add a note saying why. Links people drew are amber everywhere, so they are never mistaken for something a system reported.
What happens to your link later. A link is kept by the identity of the two documents, not by a copy of them. Re-scanning either source does not touch it, and neither does recomputing duplicates. If one of the documents is deleted at its source, the link disappears from view, and comes back if the document does. Deleting a source removes the links of its documents with it.
Built to stay fast
Explore never reads the whole workspace. Each kind of connection is read separately and with its own limit, so a document with a thousand look-alikes cannot crowd out its lineage, and a workspace of millions of documents answers as quickly as one of hundreds. When a limit is reached, the guide says so; add a neighbour as a starting point to see its own connections.
For AI agents
The same view is available to agents over MCP:
explore_connections returns what assets are connected to and why, traced as far
and in the direction asked, with the cases and watches each asset is part of; given
a source instead of assets, it starts from the source’s most connected ones, and it
keeps all connections, those inside the source or those leaving it;
find_similar_assets returns the assets that read like a given one.
Reading next
- Lineage & Relationships: how lineage is collected
- Connections & Duplicates: the signals behind same as and shared values
- Glossary & Shared Vocabulary: where concepts and entities come from
- Cases: where what you found becomes an investigation