The brief
A map is very good at telling you where something is and hopeless at telling you what it is like to be there. A GPS trace records speed and elevation and nothing about the underpass that felt unsafe, the view that was worth the climb, or the traffic noise that made conversation impossible.
That gap is the whole problem for anyone studying why people walk where they walk. The qualities that decide a route are experiential, and the standard toolchain treats them as unrecordable. In practice researchers work around it by keeping a video file open in one window and a map in another, reconciling the two by hand and by memory.
WalkGIS closes that window gap. First-person video, think-aloud narration and GPS are bound to one timeline, so the question of what somebody said, where they said it, and what they were looking at is a single query rather than three separate acts of manual cross-referencing.
The insight
Two ideas do the work, and the second one is the more uncomfortable.
The first is that synchronisation is the whole product. Once video, map and transcript address the same timeline bi-directionally, the analysis changes character. Speech becomes a spatial layer, because every timestamp is also a coordinate. Asking where participants mentioned traffic stops being a read-through and becomes a map operation. That single binding is what took the analysis of one walk from two or three weeks of manual reconciliation down to about a week.
The second is that precision can be a form of dishonesty. GIS offers polygons, so qualitative spatial research draws polygons, and a crisp boundary gets placed around a perception that never had one. Nobody says they feel safe at a coordinate; they say an area feels friendly, with edges they could not point to.
The spraycan and the gradient contours exist to refuse that. Regions accumulate from overlapping translucent marks, so density carries confidence and the edge stays soft. A boundary is drawn as bands of decreasing certainty rather than a line. The output is less tidy and considerably more truthful, and it lets two participants disagree about where a place begins without one of them having to be wrong.
What it does not do
It scales to studies, not to cities. A week of analysis per walk is a large improvement on three and it is still a week, so this is a method for tens of walks rather than thousands. The volume problem is what WalkGrid went on to address from the other direction.
The fuzzy tools capture perception and do not resolve disagreement between participants. Two sprayed regions that overlap are two accounts, not a consensus, and the framework deliberately declines to average them into one. Anyone wanting a single authoritative boundary will have to make that judgement themselves.
Local-first protects participants and limits collaboration. Because nothing uploads, there is no shared workspace, no multi-coder reliability workflow, and no way for a second researcher to annotate the same walk without passing the files across. That was the right trade for footage recorded inside people’s neighbourhoods, and it is a real constraint on the method.
