Skip to content

Paper workflow

The publication workflow is intentionally split between tracked library support and local-only benchmark assets.

Tracked in the repository:

  • core metrics, reporting, LaTeX table export, and figure generation;
  • smoke and shared suite manifests under configs/suites/;
  • documentation and handoff notes.

Local-only by design:

  • paper_support/;
  • configs/suites/paper/;
  • tests/integration/test_paper_workflow.py;
  • generated reports/;
  • .lrdbench_cache/.

These paths are ignored by Git so draft paper runs, generated artefacts, and machine-specific cache state do not enter the remote repository.

Expected local layout

When the paper kit is available on a workstation, a typical setup is:

paper_support/
configs/suites/paper/
  canonical_ground_truth.yaml
  stress_contamination.yaml
  null_false_positive.yaml
  sensitivity_disagreement.yaml

Run paper suites from the repository root:

python -m paper_support.run_paper_suites configs/suites/paper/canonical_ground_truth.yaml

The local runner should write normal reports under reports/paper/<run_id>/, stage selected LaTeX tables and figures under paper_support/artefacts/, and append a run index when configured.

For clean clones without the paper kit, use the tracked manifests under configs/suites/ and the standard CLI:

lrdbench run configs/suites/smoke_ground_truth.yaml

See the development handoff for current execution notes and phase planning.