wardley-maps.sgit.ai / Method / Maps are graphs with position
Maps are graphs with position
Original contribution 7 February 2026 W5 · glossary No implementation
a Wardley Map is not a separate artifact from the graph. It is a graph with positional metadata.… The graph is primary; the map is a projection… Map queries are graph queries with positional filters.
The claim
Every Wardley map is already a graph. It has nodes — the anchor and the components — and directed edges — the dependency links. The two coordinates on each node are attributes of the node, not a separate kind of information. The picture is a rendering of the graph with those two attributes used as x and y.
Taken seriously, that has a consequence that is bigger than it first sounds: the map stops being the artefact you maintain. The graph is what you maintain. The map is one view of it, generated on demand — and it is not even the most useful view for many questions.
If the graph is primary, then a question like "what depends on something we have placed at Genesis?" is not a mapping question at all. It is a graph traversal with a filter on a node attribute. So is "which components moved right since March?", and "what is the least evolved thing on the critical path to this user need?" — which is the question most maps are drawn to answer, and which nobody can currently ask a map because a map is a PNG.
What follows if it holds
| If a map is a graph… | …then |
|---|---|
| Position is a node attribute | Alternative axes are just other attributes. Relabelling the axis becomes a rendering choice rather than a schema change — which is exactly what the corpus proposed when it said to store "axis metadata in the graph alongside everything else." |
| The map is a projection | Two maps of the same system are two queries, not two documents that can disagree. The paired before/after technique becomes a diff over one graph rather than two hand-drawn pictures somebody has to keep in sync. |
| Maps are queryable | A map diff — "a before/after showing what moved" — is computable rather than eyeballed. That is specified in the corpus as a standing requirement and has exactly one dated pair to show for it. |
| Evidence attaches to nodes | The doctrine assessment's discipline — every rating names the artefact it rests on — becomes available for placements too. Which is Q5, and currently no map in the corpus attaches evidence to a single placement. |
And it does not exist
.owm or map-JSON export anywhere in
the corpus. What exists is this argument, four rendered SVGs, and a handful of Mermaid sources —
all of which are pictures with the graph implicit in the text, which is the situation the claim
is complaining about.Publishing an architectural idea as method is legitimate. Publishing it without saying it is unbuilt would not be, and it would be a particularly poor thing to do on a page arguing that maps should be checkable. So: this is design. /shipped/ says so too.
The nearest thing to a demonstration on this site is small and worth being precise about: the four maps in the gallery ship their source next to their render, so the graph is recoverable — by parsing a text file, which is not the same as querying a graph, but is the difference between a claim you can check and one you cannot. And graphs.sgit.ai holds the methodology for maps-as-graphs rendering; this page holds the argument. The two should be read together.
What would falsify it
- A map question that is not a graph question. The claim is strong: map queries are graph queries with positional filters. One useful, common mapping question that cannot be expressed that way — anything involving the shape of a region rather than its nodes, for instance, like the broken middle — weakens it considerably. That example is not hypothetical; it is on this site.
- Implementations that exist and do not work like this. OnlineWardleyMaps, wardleyToGo and cli-owm all parse a text DSL into an internal structure. If none of them finds a graph representation worth exposing, that is evidence the projection framing buys less than it promises.
- Building it and finding the queries are not interesting. The cheapest falsifier available, and the one nobody has run.
Related
- /agents/ — the coordinate contract and the parse rules, which are what a machine needs to get the graph out of the text.
- graphs.sgit.ai/v1/maps/ — the methodology-and-inventory surface for the same idea. Cross-linked deliberately rather than duplicated.