System connections

The screen is built over one table of connections — server-to-server edges collected from the dependency-monitoring source — and shows it five ways. The tabs stand in reading order: from the wide picture to the single case.

Tab	What it answers
Estate map	What the whole estate looks like at once — islands of systems, who is central and who is isolated
Infrastructure map	One system as an infrastructure diagram: its servers, its databases, the firewall and the equipment around it
System connections	One system and its neighbors
System servers	The same system at the individual server level
Collisions	Overlapping maintenance windows on connected systems

Where the connections come from

A daily sync pulls the server-to-server connections from the dependency-monitoring source.

The system name does not come from that source. Each server is resolved to its system through the inventory — the same authority the server list works against — so the two answers cannot diverge. A server with no inventory match is still recorded: the edge exists, simply without a system name beside it.

Estate map

Every system in one picture. Three selectors at the top:

Selector	Options
Circle size	Connection strength · connected systems · servers · overdue servers
Color	Cluster (the islands of systems that talk to each other) · update status · business criticality
Layout	Categories · structure · forces

In the header: how many systems, how many connections, how many clusters, and how many systems have no connection at all — with a switch that shows or hides them. Clicking a system focuses it and the focus holds; "Route" marks how one system reaches another, by the fewest hops, preferring a route whose weakest link is stronger.

Infrastructure map

One system drawn as a layered infrastructure diagram. Each layer's heading is written on its box — so a component's position is a claim a reader can check, not just an arrangement.

• The firewall is drawn as a wall, full width, between what is outside the system and what is inside it. A firewall drawn as one more circle in a row of circles says nothing.
• The compute layer splits into an external zone and an internal zone by the same Tier/environment rule that decides the update cadence — not by a new classification.
• Identity, monitoring and backup are drawn quietly and pinned to the last layer. Every system reaches the identity directory; at full weight that would turn a map about this system into a picture about the identity directory. They can be hidden entirely.
• Every component's panel carries the rule that classified it and the evidence — not only the classification.
• A map that was truncated says how many components are not shown. A system that is not in the register gets an empty map with the sentence "that is an answer, not a fault".

System connections

One system's neighborhood, in three presentations: radial, matrix and table. Above roughly 20 neighbors the default is the matrix and below it the radial — not for looks, but because a controlled experiment found node-link readability degrades sharply with node count while a matrix is barely affected, with one exception: path following, which is not the question asked here. You can switch at any time.

There is a visible neighbors shown control. Nothing is capped silently.

Two decisions that apply to every graph

Position is derived from structure, not from a random seed. A force layout drew the same estate differently on every load, so position only appeared to carry a fact. Even the "forces" mode is seeded from the structure, so it reproduces.

Every graph has a table twin. That is not a convenience — it is how the data is read with a keyboard and with a screen reader:

Alongside that: every graph exports to PNG, and pan and zoom reset whenever the view is rebuilt — including on a focus click.

Free-form mapping

Beside the maps the product computes there is a surface you draw yourself: Free-form mapping. Drag entities from the inventory onto the canvas, arrange them, connect them, add text and shapes — and save.

• The save is explicit, with an unsaved marker, and the camera (position and zoom) is stored with the document.
• Two editors at once: saving over a map that changed meanwhile stops and offers a reload. There is no silent overwrite.
• A direct link to a saved map, PNG export, and adding it to a dashboard.