Doors as Edges

A door connects spaces and shapes how people move between them. Its location and properties affect circulation, privacy, accessibility and, in some cases, fire separation. But the first thing a door does is simpler than any of that: it decides whether two rooms that touch are rooms you can walk between. A plan full of shared walls tells you who the neighbors are. Only the doors tell you who can get to whom.

Adjacent is not connected

Four equal rooms on one plate (16 × 12 ft each at the base 32 × 24 ft) share four walls, so four doors are possible: D1 R1–R2, D2 R1–R3, D3 R2–R4, D4 R3–R4. R1 and R4 meet only at a corner, so no door can join them. R1 is the reference room. Routes run room center to room center through open doors only. A door with no backup is one whose closing cuts some room off from R1; a dead end is a room with one door, so no route passes through it. The example tests connectivity, not exit count, travel-distance limits, clear width, hardware, ratings, accessibility or life-safety compliance.

Adjacency is the offer; the door accepts it

Thesis 27, Walls as Edges, stored every shared wall as an edge between two room nodes and noted that a door is an attribute on one of those edges — this page is about what happens when you set it. Thesis 28, Adjacency Rules, asks which rooms should touch. This one asks which of those touches people can pass through. Every door needs a shared wall; most shared walls never get a door. So the access graph (rooms joined by doors) is always a subset of the adjacency graph (rooms joined by walls), and the choice of which subset is where circulation, privacy and egress actually get decided.

One plate, four possible doors

The drawing uses the smallest plan that shows it. A 32 × 24 ft plate split into four 16 × 12 ft rooms. R1 and R2 share a wall, so do R1 and R3, R2 and R4, R3 and R4: four shared walls, four possible doors, D1 through D4. R1 and R4 meet only at a corner, so no door can ever join them. Treat R1 as the reference room — the one with the way out — and measure every route center to center through open doors only: 16 ft across a vertical wall, 12 ft across a horizontal one.

• All four doors open. The access graph is a ring. R2 is 16 ft from R1, R3 is 12 ft, R4 is 28 ft by either of two routes. Close any one door and every room still reaches R1. No door is load-bearing, so to speak.
• Close D3 (R2–R4). Every distance stays exactly where it was: 16, 12, 28. A route-length check would say nothing happened. The graph says plenty did. R2 and R4 are now dead ends, and all three remaining doors are the only way into at least one room.
• Close D1 (R1–R2) instead. Same three doors, same count on the schedule. R2 goes from 16 ft to 40 ft (12 + 16 + 12), and the route to it runs through R3 and R4, which have just become corridors.
• Close D3 and D4. R4 still shares two walls with its neighbors and has no way in at all. Adjacent, not connected.
• Stretch the plate to 46 × 38 ft with D3 closed. R4 is now 42 ft out instead of 28, but the same rooms are dead ends and the same doors have no backup. Geometry changed the lengths; the doors still decide the structure.

The cards under the drawing count exactly those things: open doors, rooms connected to R1, the longest route to R1, and doors with no backup — the ones whose closing cuts some room off. The labels in each room say whether it is a dead end (one door) or a pass-through (two or more).

What the door graph decides

Paths. Route length depends on which doors, not how many. Three doors gave R2 a 16 ft route in one arrangement and a 40 ft route in another. A door count on a schedule or a cost estimate can’t see that difference. The graph can.

Privacy. A room with one door is a dead end: nobody walks through it to get somewhere else. A room with two or more doors is a pass-through, whether anyone meant it to be one or not. That’s the difference between a bedroom and a bedroom somebody crosses to reach the kitchen. In my view every door you add to a room is a small vote for turning it into a hallway, and the graph is the only place that vote gets counted.

Egress. Before anyone measures a travel distance, there are two reachability questions: does every room have a route out at all, and does that route survive one door being closed or blocked? The ring passes both. The tree with D3 closed passes the first and fails the second for every room behind a single door. That’s a graph fact, not a code finding: how many exits a space needs, how far apart, and how far people may travel all depend on occupancy and the adopted code, and clear widths, hardware and ratings must still be checked against the applicable requirements. The graph locates the questions; it doesn’t answer them.

The door as a connection

Thinking about a door as a connection helps relate its specification to both spaces it serves. Rating, hardware and clearance are properties of the passage between two particular rooms, not of the leaf in isolation, which is why the same slab can need a different hardware set on a different wall. It also explains what a door schedule really is: every row names two rooms, so the schedule is the building’s access list — the subset of its adjacencies people can walk through — written as a table.

Draw the door last, decide it first

A route through a building can be represented as a sequence of rooms and openings, and that representation turns a plan into relationships that can be reviewed systematically while leaving room for architectural judgment about how the route feels and works. The judgment I’d make: decide doors as graph edits, not as symbols. Before a door goes into a wall, ask which room it turns into a corridor, which room it leaves with only one way in, and whether any room is now one closed door away from being cut off. A door is an object, but its meaning comes from the passage it creates.