Using the map

Views

The view switch sits in the bottom panel; the button is labelled with what it will switch to.

Structure — an anatomical layout: nodes sit by area and node size does not depend on the number of links.

Clusters — nodes are pulled together by link density, and node size reflects the number of links. This shows where the work has actually piled up.

The clustering is computed once per map snapshot and survives switching views.

Controls

Action	What happens
Hover	highlights the node (with a small delay so the map doesn’t flicker)
Single click	selects the node and highlights its links
Double click on a node	the camera flies in, the panel opens
Click on the background	clears the selection
Double click on the background	fits the whole map on screen
Esc	first closes the panel, then clears the selection
Rotation	a button: pause / resume

A selected node not only changes colour but grows — on a dense cloud colour alone is not enough.

While a node is selected its neighbourhood is pulled towards it and the links are drawn as that group’s tendrils: you can inspect the surroundings without losing the whole picture.

Search

The search field highlights nodes by title and shows the number of matches. Search highlighting outranks other colouring: what you are looking for is visible on top.

The node panel

Opened with a double click.

• Type and Area;
• Freshness — how recent the object is; without a timestamp it says so;
• Grouping — when a node stands for a group of mail threads;
• “Why it is in memory” — what put the object on the map;
• “Extracted from” — the source the agent took the fact from, plus “Other sources” when there are several;
• confidence as a percentage;
• Open — jump to the object in its own section.

While the panel loads it says “Reading links and sources…”; if that fails it says so plainly.

If the object has no links at all, the panel states it: the agent has not linked this to anything yet — the node stays in its area and is dimmed.