Command-line reference

The installed commands are chromantis-data, chromantis-dataset, chromantis-experiment, chromantis-select, and chromantis-explore. The checkout scripts shown below are equivalent.

Data repository

# Print the complete catalog.
python -m chromantis.data catalog

# Initialize data/catalog.json and repository directories.
python -m chromantis.data --data-root data init

# Preview or execute groups.
python -m chromantis.data --data-root data pull --group core --dry-run
python -m chromantis.data --data-root data pull --group all

# Inspect and verify immutable snapshots.
python -m chromantis.data --data-root data show delphi_nhsn
python -m chromantis.data --data-root data verify delphi_nhsn

Groups are all, core, cdc, delphi, hubverse, and pophive. Dataset keys can be given instead of a group.

Delphi selection and resume

python -m chromantis.data --data-root data pull delphi_nssp \
  --mode archive \
  --signal pct_ed_visits_influenza \
  --geo-type state \
  --workers 4

python -m chromantis.data --data-root data pull delphi_nssp \
  --signal pct_ed_visits_influenza --geo-type state \
  --resume-from data/.staging/delphi_nssp/<staging-id>

Relevant options include --mode, --snapshot-date, --report-time, repeatable --signal and --geo-type, optional --fill-method, and bounded --workers. --report-time accepts a date or comparison such as '<2026-09-01' and is valid only in archive mode. --snapshot-date is valid only in snapshot mode. Both modes retain reference_time observation dates and full report_time timestamps. Requests select the full observation history and all published fill variants by default. Use --fill-method source only for a source that publishes that variant; claims return blank labels.

The delphi group contains delphi_nhsn, delphi_nssp, delphi_nwss, delphi_claims_inpatient, delphi_claims_outpatient, delphi_fluview_ilinet, delphi_fluview_clinical, and delphi_flusurv.

python -m chromantis.data pull delphi_claims_inpatient \
  --signal claims_inpatient_adm_pct_claims_flu --geo-type nation \
  --report-time 2026-09-11

Resume requires identical query selectors and a matching saved request.json. Use the original command with --resume-from added; --workers can change. Staging directories without a query record must be restarted.

Hub historical state

python -m chromantis.data --data-root data pull hub_flusight_current \
  --hub-as-of 2025-01-15

Use --hub-ref for an explicit branch, tag, or commit.

Explorer

# Build only; reuse a current index.
python -m chromantis.explorer --data-root data index

# Force a complete index rebuild.
python -m chromantis.explorer --data-root data index --force

# Ensure the index is current, then serve it.
python -m chromantis.explorer --data-root data serve

# Serve the existing index immediately.
python -m chromantis.explorer --data-root data serve --no-index

# Bind another local port without opening a browser.
python -m chromantis.explorer --data-root data serve \
  --port 8877 --no-browser

# Write the thinned static copy published as the live explorer on GitHub Pages.
python -m chromantis.explorer --data-root data export --out docs/explorer/data

# Index, export, and optionally preview the published copy in one step.
scripts/update_published_explorer.sh --preview

See Published explorer for what the export keeps and how it is committed.

Shared selection inventory

PYTHONPATH=src python -m chromantis.data.selection --data-root data
# After installing the package:
chromantis-select --data-root data

This read-only command reports policy version, source grouping, the NHSN measure allowlist size, Delphi/CDC crosswalks, and missing downloads. See Shared selection for the downstream interface.

Experiments, evaluation and production

Command	What it does
python -m chromantis.experiment.planner plan -e NAME --problem problems/PROBLEM.json --study experiments/NAME.json	Plan a study (or -s SCENARIO ...)
sbatch --job-name=NAME --array=0-3 scripts/jlessler.sbatch NAME	Fit on the patron GPUs (shared queue)
python -m chromantis.experiment.planner run -e NAME	Fit locally
python -m chromantis.experiment.planner status -e NAME	Completion and the exact resubmission command
python -m chromantis.experiment.planner rank -e NAME	Score, rank and write the report
python -m chromantis.experiment.planner replay -e NAME --inputs reported	Evaluate completed fits on other inputs, no refit
python -m chromantis.evaluation.ensembles GROUPS.json --out DIR	Combine and rank saved forecasts, no refit
python -m chromantis.production run --release R --dataset P --issuance D	Weekly forecast from a release: replay, export, plots
python -m chromantis.production intervals\|peers\|diff ...	Submission check plots and comparisons
python -m chromantis.dataset.ili	Rebuild the pre-2022 historical ILI archive

The workflow describes each step; the production README the weekly submission.