Closes every method-coverage gap listed in the v1.2 roadmap. 16 runnable demos, 68 numeric-check assertions, 46 Crossref-verified references, 51 validated tool bindings, 246 unit tests — all CI-enforced.
Five new numeric-contract demos
| Demo | The point it pins numerically |
|---|---|
randomization-inference-demo/ |
Robust-SE t-tests over-reject under concentrated leverage; the Fisher randomization test has exact size and retains power (Young 2019) |
qte-demo/ |
A null ATE hiding large distributional effects; quantile regression recovers the analytic QTE curve (Koenker-Bassett 1978) |
lp-did-demo/ |
Pooled TWFE event studies contaminated under staggered adoption; LP-DiD with clean controls recovers the true dynamic path (Dube-Girardi-Jordà-Taylor) |
bunching-demo/ |
Excess mass at a tax kink identifies the earnings elasticity (Saez 2010), with oracle-vs-feasible counterfactuals and two falsification worlds |
matrix-completion-demo/ |
DiD structurally biased under interactive fixed effects; low-rank imputation recovers the truth — rank-sensitivity caveat demonstrated, not hidden (Athey et al. 2021) |
Also in this release
- StatsPAI tool registry expanded 43 → 51 validated bindings (
lp_did,ri_test,qte,bunching,notch,matrix_completion,interactive_fe,aipw), hub kept in sync by the registry gate. - Methods reference & glossary coverage for all five methods.
- Launch kit (
docs/launch-kit/) with bilingual announcement drafts and awesome-list blurbs;.zenodo.jsonfor DOI minting.
Full details in CHANGELOG.md.
🤖 Generated with Claude Code