Which interface should I use to run the pipeline?¶

CartLoader offers three levels of control. They produce the same kind of CartoScope assets — they differ in how much the tool orchestrates for you.

Interface	What it is	Best for
`run_together`	One command runs the whole pipeline (ingest → FICTURE → cells → packaging → images → optional publish) across platforms and samples, as a single resumable Makefile	Multi-platform or multi-sample runs; a joint model across sections; one-command convenience; batch processing
`run_xenium` / `run_visiumhd`	Per-platform orchestrators for a single platform and sample, with explicit action flags and per-tool path options	A single Xenium/Visium HD sample when you want fine-grained control over each action and tool path
Individual modules	Call `sge_convert`, `run_ficture2`, `run_cartload2`, `import_`, `upload_` yourself	Custom pipelines, unsupported/custom input formats, debugging a single step, tight HPC integration

Quick decision rule¶

Several samples, or want a joint model, or just want it to run end-to-end? → run_together. Start with the single-sample or multi-sample tutorial.
One Xenium/Visium HD sample and you want explicit control per step and per tool? → run_xenium / run_visiumhd.
A platform without a built-in profile, a custom format, or you're debugging one step? → the individual modules. See Supported Platforms & Inputs and the platform starter tutorials.

They interoperate

All three call the same underlying modules and produce the same catalog.yaml + PMTiles. You can prototype with run_together, then drop to modules for a step that needs customization — or vice versa.