A portable version of pipelines/roads-transform-demo.json (docs/TRANSFORMERS.md, sections 3-4): drops the text warning layer "Disclaimer" and fills in a default where no feature carries the field at all. Both step parameters (the dataset name "Disclaimer", the field name "purpose") are specific to THIS dataset, so applies_to closes the schema with a fingerprint of a known dataset, not with classes: the rule itself isn't portable to a stranger's drawing, and applies_to says so honestly.
We read DWG, DXF, MapInfo TAB, Shapefile, GPKG, GeoJSON, and a ZIP that holds them: we take the file and its sidecars out ourselves, and say exactly what we took. Upload limit, 300 MB.
How much we take on at all is counted from the memory of the container that will READ the file: DWG up to 345 MB, DXF and tables up to 1035 MB. A bigger file is refused right away, with a number, not left to die of memory in half an hour.
And separately, a limit by the NUMBER of features: we take up to 434110 of them. It matters more than size: a 20 MB drawing with a hundred thousand features is heavier than a 200 MB drawing with a thousand. We find out how many you have in the first seconds of reading and, if we cannot take it on, say so as a number, before the work, not after half an hour.
The file is not read without consent: this is checked here, on our server, not by a checkbox in the browser.
work plan, stages, queues, jobs and a diagram. Drawn from a file in the repository, there is no other source for it.