SPECTER LABS

Paper Workflow Runbook

This is the operator path for the current wonton-soup paper workflow.

Decision

The shared runtime lake is the dataset.

Do not treat small lake-job slices as alternate sources of truth. They may exist as optional convenience surfaces, but the paper should be analyzed against the shared lake.

Inputs

  • lake DB: LAKE_DB_PATH or the shared-lake default resolved by build_figures.py
  • figure builder: dossiers/wonton-soup/paper/build_figures.py
  • manuscript: dossiers/wonton-soup/paper/main.typ

Steps

1. Refresh the contributing lake roots if needed

Run postprocess + lake reconcile for any root whose runs should feed the shared lake analysis. Iterate on data collection until the lake contains the evidence the paper needs.

2. Build figures directly from the shared lake

build_figures.py reads the shared lake and then applies the paper’s internal cell filter: March p2-paired for lesion / structural figures, and March p4-basin-deep for the basin panel.

LAKE_DB_PATH=/path/to/shared/lake.duckdb \
uv run python dossiers/wonton-soup/paper/build_figures.py \
  --out-dir dossiers/wonton-soup/paper/artifacts

Expected artifacts:

  • paper/artifacts/fig17-followup-provider-splits.svg
  • paper/artifacts/fig16-ged-bimodality.svg
  • paper/artifacts/fig18-followup-basins.svg

3. Compile the paper

typst compile \
  --root . \
  dossiers/wonton-soup/paper/main.typ \
  dossiers/wonton-soup/paper/artifacts/main.pdf \
  --font-path addenda/typst-field-manual/assets/fonts

Optional convenience slices

If a narrow materialized cohort is useful for debugging or release prep, these optional presets exist:

  • dossiers/wonton-soup/analysis/lake/presets/79_wonton_paper_closeout_v1.json
  • dossiers/wonton-soup/analysis/lake/presets/80_wonton_paper_program_v1.json

They are convenience slices, not the dataset.