Skip to content

tscfbench

A benchmark-and-workflow package for time-series counterfactual inference.

python -m tscfbench helps you turn a counterfactual question into a reproducible study, a readable report, and a reusable workflow. It is not only a model package: it also provides benchmark protocols, canonical studies, teaching surfaces, and agent-friendly artifacts.

What it is

  • A stable schema for impact and panel counterfactual tasks.
  • A benchmark layer for single studies, canonical studies, and model sweeps.
  • A workflow layer for reports, notebooks, docs, CI, and coding-agent use.

What it is not

  • It is not a claim that one built-in baseline is the last word in methodology.
  • It is not a giant all-in-one causal inference framework.
  • It is not only a demo notebook; it is meant to survive in real research workflows.

Why people adopt it

  • It starts from recognizable research jobs instead of source files.
  • It tells users why each API exists, where it works best, and what it returns.
  • It ships canonical studies, benchmark cards, tutorials, and release-facing docs.
  • It is also designed for token-aware, agent-driven research workflows.

First commands to run

python -m tscfbench package-story
python -m tscfbench capability-map
python -m tscfbench api-atlas
python -m tscfbench scenario-matrix
python -m tscfbench tutorial-index