causal-worlds 0.7.0

Generate fictional-but-coherent causal operations worlds (executable sim + ground-truth answer-key) from a natural-language description, for benchmarking causal-discovery agents.

causal-worlds

Turn a plain-language description of an operation into a fictional causal world with a declared, ground-truth causal graph — then benchmark whether a causal-discovery method can recover it.

Because the structure is declared (not learned from data), it's an answer key: run any discovery method on the generated data and score how well it recovered the world. The worlds are fiction-first — plausible and internally consistent, not models of any real system — so there is no data to leak and nothing to memorize, which is exactly what makes a causal benchmark trustworthy.

from causal_worlds import worlds, grade_spec, InterventionalCiDiscoverer

spec = worlds.get("coffee")                          # a hidden confounder + a regime sign-flip
report = grade_spec(spec, InterventionalCiDiscoverer())
print(report)   # directed_shd=0  skeleton_shd=0  f1=1.0  confounded_reported=0
#                ^ swap in YOUR discoverer to benchmark it against a known truth

Status — v0.6, beta. The full loop works: natural language → an admitted causal world, persisted with provenance. A Claude author proposes the world; an independent Gemini judge (a different model family) plus statistical gates admit only worlds that are valid, recoverable, and not guessable from variable names. The deterministic engine (specify → sample → grade → score) and all grading run with no API key; only authoring needs keys. Worlds are currently tabular SCMs with do() interventions — a Gymnasium env, temporal lags, and counterfactual replay are on the roadmap. See the CHANGELOG.

Install

uv add causal-worlds            # or: pip install causal-worlds   (once published)
uv add 'causal-worlds[discover]'  # + the baseline discovery stack (PC/GES/FCI/GIES)
uv add 'causal-worlds[llm]'       # + natural-language authoring (Claude + Gemini)

The base install (engine, grading, built-in worlds, CLI) needs only typer, pydantic, numpy.

60-second quickstart (no API key)

causal-worlds worlds                     # list built-in worlds: coffee, ecommerce
causal-worlds gate coffee                # run the validity gates -> admitted=True
causal-worlds grade coffee               # grade the reference discoverer -> directed_shd=0 ...
causal-worlds score benchmark/v0.5/world_01   # grade the reference on a shipped benchmark world

New to it? Walk through the getting-started guide or run the examples.

Benchmark your own discoverer

Implement one method — recover(substrate, *, seed) -> set[(src, dst)] — and grade it against any world's answer key:

from causal_worlds import grade_spec, worlds

class MyDiscoverer:
    def recover(self, substrate, *, seed):
        sample = substrate.sample(2000, seed=seed)        # observational data...
        flows = substrate.sample(2000, seed=seed, do={"price": 1.0})  # ...or interventional
        return {("price", "demand")}                       # your recovered edges

print(grade_spec(worlds.get("coffee"), MyDiscoverer()))

Or from the CLI on a persisted world: causal-worlds score <bundle> --discoverer your_pkg:YourClass.

Author a world from a description (needs [llm] + keys)

Set ANTHROPIC_API_KEY and GEMINI_API_KEY in your environment, then:

causal-worlds generate "a coffee chain with weekend swings and variable lead times" ./my-world

from causal_worlds import generate
from causal_worlds.author import build_claude_author
from causal_worlds.judge import build_gemini_judge

world = generate(
    "a hospital ED with triage staffing and bed pressure",
    author=build_claude_author(complexity="hard"),   # easy | standard | hard
    judge=build_gemini_judge(),                       # independent model family
)
print(world.report.difficulty, world.report.grade)

Does it actually defeat the standard toolbox?

Measured across the 35-world benchmark/v0.5 set (3 seeds each):

method	mean skeleton-SHD ↓	directed F1 ↑	confounded pair kept as causal ↓
interventional-ci (reference)	1.47	0.91	0
PC	2.81	0.67	13
FCI	2.67	0.71	8
GIES	6.66	0.68	17

Observational/score-based methods keep the hidden-confounded pair as a causal edge in most worlds; the interventional grader never does. And structural difficulty (confounders + regime sign-flips) predicts the observational collapse (corr +0.62) where name-guessability does not (+0.14) — so the difficulty score is a real instrument. Reproduce: evals/.

What you get per world

1. An executable SCM — sample observational data and do()-intervene, deterministically by seed.
2. A time-series dataset — the observed variables (the input to a discovery method).
3. An answer key — the declared causal edges + the hidden-confounded pairs, derived from the spec.
4. A manifest — full provenance (models, grader version, seed, difficulty) and an honesty label.

Concepts

• Spec / IR — variables (with roles, incl. hidden), linear-Gaussian mechanisms, regime sign-flips.
• Answer key — directed edges over observed variables + the hidden-confounded pairs; derived from the spec, never stored separately, so they can't disagree.
• Gates — T1 validity · T2 sample-sanity · T3 non-triviality vs a random-graph null · T4 anti-cliché (the judge can't guess it from names). A world is admitted only if all pass.
• Reference grader — an interventional-CI discoverer that uses do() data to tell confounding from causation, where PC/GES/GIES/FCI (which assume causal sufficiency) cannot.

Depth: docs/scope.md · docs/hld.md · docs/lld.md · docs/architecture.md · docs/validation.md.

Roadmap

Shipped: NL authoring, independent judge + anti-cliché gate, artifact persistence, the baseline crossover, a structural-difficulty axis, a 35-world benchmark, and temporal worlds (lagged edges + autoregression — see the built-in supply). Next: time-series grading for the temporal worlds (PCMCI+, VARLiNGAM), a Gymnasium env with perturbations + counterfactual replay, scaling to 100+ worlds, and conversational elicitation. Tracked as issues.

Why this is the unoccupied intersection

Today's tools each own one corner — natural-language authoring × executable causal simulator × ground-truth answer-key for discovery is the gap:

Tool	Corner it owns	What it lacks (for this job)
G-Sim	LLM authors a sim + calibrates to data	needs real data; aimed at fidelity, not a declared answer-key
DEVS-Gen	NL → executable discrete-event ops sim	no declared causal-graph answer-key
SD-SCM	LLM fills mechanisms → counterfactuals	needs a user-supplied DAG; tabular, not an executable sim
TimeGraph	known-graph time-series for discovery	parametric/templated; no natural-language authoring

Built on the shoulders of pgmpy, DoWhy, CausalPlayground, causal-learn, and Gymnasium.

Contributing

Issues and PRs welcome. The bar: make validate green (ruff select=ALL, mypy strict, pytest with a coverage floor) — see docs/engineering.md. Atomic, conventional commits.

License

MIT. An open-source project from Noumenal.