Where are outputs written, and which files should I keep?

With run_together

Everything goes under --out-dir:

<out-dir>/
├── run_together.mk            # the generated Makefile
├── run_together.resolved.json # the fully resolved configuration (provenance)
├── mk/                        # per-stage done flags
├── tsv/<id>/                  # converted SGE
├── fic/                       # FICTURE models, decoding, and cell analyses
└── cartl/<dir>/               # catalog.yaml + PMTiles  ← the deployable result

The packaged directory <dir> is <out-dir name>-<id>, shortened to <id> when the two are equal. A joint multi-sample run writes one such directory per sample plus cartl/multi-catalog.yaml.

Keep cartl/ (that is what CartoScope reads; upload it as a unit) and run_together.resolved.json (it records exactly what was run). tsv/ and fic/ are needed to rerun or extend the analysis, e.g. to add a model with a different --n-factor without repeating SGE conversion; delete them only when you are done.

With the individual commands

Each command writes into its own --out-dir, plus a manifest that records its key artifacts:

• sge_convert: unified SGE files + sge_assets.json
• run_ficture2: models and decoding outputs + ficture.params.json (run_ficture2_multi writes ficture.multi.params.json)
• run_cartload2: PMTiles layers + catalog.yaml (and ficture_assets.json)

Keep the final manifests and catalog.yaml, the PMTiles they reference, and any SGE you plan to reuse. Remove later, if needed, intermediate files that the pipeline regenerates, and diagnostic plots or logs you no longer need. Before deleting, confirm your next step does not depend on those files.