Architecture
A map nobody had to draw.
Every architecture diagram in a company is out of date, because it was drawn by hand and the code moved. This one is read out of the repository — services, modules, files and the calls between them — so it is as current as your last scan, and it says which parts of itself it could not follow.
What it draws, and from what
Not a file tree with arrows on it. The scan resolves calls between symbols and lifts them into the levels a person actually thinks in — Project → Services → Features → Modules → Files → Functions → APIs → Dependencies — so a request going handler → service → database is visible as that path rather than as four boxes somebody arranged.
- Every box and every line comes from a resolved call in your code. Nothing on the map is a guess from a folder name.
- Open a box and it names the file and the line. The map is a way into the code, not a picture beside it.
- The server half of a request path is drawn today. Screen → API is Release 2, and the screen says not measured yet where that line would be, rather than leaving a gap that reads as “nothing calls this”.
- A request seen travelling, live, is not built. It is a named phase in our own plan and it has never been attempted; this page will say so until it has.
What a change would reach
The question worth asking before a merge is not “what imports this file” — it is what depends on it through those importers, which is the number somebody schedules an afternoon around. The map walks the closure and gives you that.
And it is a floor, in three separate ways that travel with the number rather than sitting in a footnote:
- Anything reached only through a call the scan could not follow is not in it.
- The walk is file to file, so it over-counts inside a file and under-counts across one it could not read. The two errors do not cancel, and neither is quietly dropped.
- A pathological file hits a ceiling, and a walk that crossed one is reported as incomplete rather than as a total.
Three things it will tell you about the shape
These feed the Architecture index, calibrated at 0.11 across 20 repositories. That is above the 11 this product requires before it stops calling a scale thin — so unlike our Optimization index, this one is set by a distribution rather than by its single worst member, and it is not labelled as provisional.
How much of it was read
A map is only worth what the scan could resolve, so the figure is on the screen beside it. It is a share of resolvable call sites, it floors, and with anything unresolved it reaches 99 and never 100. The calls no static tool can resolve are drawn on the same scale outside that figure — neither counted in it nor dropped.
A region the scan could not read is never drawn as an empty one. That is the difference between a map and a diagram: a diagram is finished because somebody stopped drawing.
Draw it from your own checkout
One command, in the repository you want mapped. Your source never leaves the machine on this path.
Or on every pull request
The GitHub App maps the branch and leaves a neutral check run. No permission on the installation could commit to your repository.
Traffic is shown beside the map, never on it
Observed requests are drawn in their own panel rather than as lines on the architecture. A measured thing and a traced thing look different on purpose — joining them would make one picture that is half fact and half inference.
The other two questions
The same scan answers both, and each has its own page.
Nothing here is a grade of your codebase, and there is no screen that names who wrote what — permanently, by decision. Privacy says what we keep; Security names the build gate behind each claim.