0.9.0

What's Changed

Major Changes 🛠

• Add Pipeline workflow with sensitivity checks, report generation, and documentation by @drbenvincent in #749 [EXPERIMENTAL - see below]
• Add PanelRegression experiment class for fixed effects estimation by @drbenvincent with @Copilot in #628
• Add Bayesian Synthetic Difference-in-Differences by @thomaspinder in #823

New Features 🎉

• feat: add legend_kwargs parameter to BaseExperiment.plot() by @haoyu-haoyu in #831
• Implement summary method for InstrumentalVariable by @anevolbap in #851
• Document and test the outcome-scale prediction contract by @drbenvincent in #1020

Bugfixes 🐛

• Fix simulate_data.py reproducibility + remove synthetic CSVs by @louismagowan in #794
• Fix mypy type: ignore codes in panel_regression.py by @anevolbap in #853
• ci: pin pyproject2conda==0.22.1 to work around 0.23.0 regression by @drbenvincent in #878
• Standardise HDI_PROB usage across experiments and reporting by @drbenvincent in #891
• Wire hdi_prob through every Bayesian plot method by @drbenvincent in #893
• Fix inline bullet list rendering in docstrings (closes #892) by @drbenvincent in #897
• Fix IPW plot crash with extreme propensity scores by @anevolbap in #854
• fix(ipw): preserve treatment design_info instead of overwriting with covariates by @haoyu-haoyu in #842
• Warn when staggered DiD ATTs lack untreated-period support by @drbenvincent in #949
• Fix #664: clone sklearn models to avoid mutating fit_intercept by @aahammer-eu in #942
• Fix DiD interaction-term matching by @drbenvincent in #1001
• Parse DiD interaction terms with Patsy by @drbenvincent in #1003
• Use Patsy response vectors for transformed formula LHS by @drbenvincent in #1004
• Validate exact treated predictors in RD and RK by @drbenvincent in #1005
• Classify panel fixed effects with Patsy metadata by @drbenvincent in #1006
• Parse PiecewiseITS transforms with Patsy metadata by @drbenvincent in #1008
• Treat datetime formula predictors as continuous by @drbenvincent in #1007
• Reject unsorted or duplicate ITS time indexes before fitting by @drbenvincent in #1019
• Fix residual-shape broadcasting bug in OLS DiD effect_summary (~70x SE inflation) by @Mari1988 in #1026
• Fix biased SE denominator in OLS DiD effect_summary (SSR/n -> SSR/(n-p)) by @Mari1988 in #1025
• Fix biased SE denominator in OLS RD effect_summary (SSR/n -> SSR/(n-p)) by @Mari1988 in #1027

Documentation 📖

• Add knowledgebase guide for custom PyMC models by @drbenvincent in #741
• Update README.md and docs homepage by @drbenvincent in #648
• Elevate consulting CTA up page + minor docs tweaks by @drbenvincent in #815
• Add central guide for pipeline sensitivity checks by @drbenvincent in #818
• docs: fix Sphinx build warnings by @anevolbap in #834
• Self-hosted contributors image + weekly regen workflow by @drbenvincent in #856
• Replace plot() *args/**kwargs with explicit signatures (#886) by @drbenvincent in #894
• Add PyMarkdown pre-commit hook for repository Markdown files by @drbenvincent in #900
• Add full notebook re-execution mode by @anevolbap in #855
• chore: enforce stricter ruff on docs notebooks (#574) by @anevolbap in #868
• Add StaggeredDiD group-time effect plots by @drbenvincent in #939
• Widen numpydoc-validation checks across causalpy by @drbenvincent in #902
• Enforce single top-level markdown heading per docs notebook by @drbenvincent in #866
• Update contributors image by @github-actions[bot] in #965
• Fix numpydoc docstring section ordering by @drbenvincent in #971
• Comparison notebook: Why flexibility matters? by @cetagostini in #703
• Example Gallery by @juanitorduz in #558
• Fix how-to gallery hidden toctree MyST syntax by @drbenvincent in #986
• Update contributors image by @github-actions[bot] in #1010
• Clarify causal estimand extraction approaches by @drbenvincent in #1033
• docs: restore per-document bibliographies after sphinxcontrib-bibtex 2.7.0 by @anevolbap in #909

Agent-related 🤖

• Add scope-drift escalation to pr-to-green by @drbenvincent in #812
• Clarify reuse-first Python environment guidance by @drbenvincent in #808
• Add sub-issue discovery to github-issues skill by @drbenvincent in #817
• Add maintainer PR review skill by @drbenvincent in #920
• Add ARCHITECTURE.md for agent and contributor onboarding by @drbenvincent in #924
• Separate developer and user agent skills by @drbenvincent in #927
• Improve user/agent facing skills by @drbenvincent in #929
• Slim ARCHITECTURE.md and reference it from CONTRIBUTING.md by @drbenvincent in #945
• Add deterministic structural check scripts by @drbenvincent in #948
• Refresh agent skills layout by @drbenvincent in #952
• Redesign CausalPy method selection skill by @drbenvincent in #931
• Align agent skill names with directory and file conventions by @drbenvincent in #1015
• Strengthen PR review probes and summaries by @drbenvincent in #1061

Maintenance 🔧

• Swap pre-commit workflows to prek by @drbenvincent in #780
• Fix blocking Matplotlib windows during pytest by @drbenvincent in #782
• Update github-issues skill with parent-child sub-issue workflow by @drbenvincent in #784
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #803
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #805
• Fix its_lift_test notebook compatibility with newer pymc-marketing by @drbenvincent in #811
• Handle xtensor BSTS pymc-marketing components by @drbenvincent in #813
• Cap ArviZ below 1.0 pending migration by @drbenvincent in #810
• Add maketables plugin MVP for core experiments by @drbenvincent in #773
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #828
• Clean up experiments/: typos, annotations, and constants by @louismagowan in #796
• Add SoftmaxWeightedSumFitter for synthetic control by @thomaspinder in #822
• docs: improve parameter docstrings in InversePropensityWeighting class by @haoyu-haoyu in #832
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #839
• Remove numpy<2.4 pin now that pytensor fix is released by @drbenvincent in #781
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #859
• chore: pin pymc<6 and pytensor<3 to current major versions by @anevolbap in #906
• fix: SDiD plot() API and sdid_pymc notebook ruff compliance by @drbenvincent in #919
• Fix ITS plot when post-period has a single observation by @drbenvincent in #895
• docs: clarify IV example labels (#499) by @anevolbap in #907
• Fix scalar score coercions for mypy by @drbenvincent in #940
• ci(docs-preview): cancel when not in defined scopes of git diff by @RussellSB in #943
• Add local patch coverage gate by @drbenvincent in #951
• Add outcome falsification, random placebo folds, and model clone support by @cetagostini in #826
• Skip welcome workflow for bot accounts by @drbenvincent in #956
• Bump pre-commit hooks and fix PyMarkdown MD032 violations by @drbenvincent in #957
• ci: harden welcome.yml and silence zizmor dangerous-triggers by @anevolbap in #908
• ci: install test job env via micromamba (refs #279) by @anevolbap in #910
• ci: consolidate pyproject2conda invocation through the prek hook by @drbenvincent in #882
• Consolidate experiment design-matrix attributes into xr.Dataset by @drbenvincent in #849
• refactor: centralize PyMC/sklearn dispatch in ModelAdapter by @drbenvincent in #961
• refactor: normalize sklearn fit inputs and collapse backend branches by @drbenvincent in #962
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #969
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #973
• [plotting] Allow user to specify type of posterior data visualisation by @AllenDowney in #680
• fix: make three-period ITS test fixtures order-independent by @anevolbap in #960
• Rename internal plot_xY to plot_posterior_over_x by @drbenvincent in #987
• Require a valid group*post_treatment interaction term in DiD formulas by @Mari1988 in #993
• Make OLS causal_impact lookup order-independent for the DiD interaction term by @Mari1988 in #994
• docs: rename demo notebooks to descriptive slugs (#840) by @anevolbap in #861
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #1012
• Add validation to _data_setter for non-standard data node names by @drbenvincent in #847
• Add pymc_forecast model-provider adapter for InterruptedTimeSeries by @twiecki in #1014
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #1031
• Align local patch coverage gate with Codecov (split from #1024) by @drbenvincent in #1034
• Unify predict() to a canonical prediction container and delete the parallel legacy paths by @drbenvincent in #1036
• Unify _bayesian_plot / _ols_plot into a single backend-agnostic _plot by @drbenvincent in #1053
• Add seeded batch plot-generation script for visual regression review by @drbenvincent in #1054
• Dispatch remaining Bayesian/OLS branches on posterior size, not backend identity by @drbenvincent in #1057
• Make InferenceData capability explicit at adapter seam by @drbenvincent in #1058
• Canonical score contract across model backends by @drbenvincent in #1059
• Prepare 0.9.0 release by @drbenvincent in #1095

New Contributors

• @thomaspinder made their first contribution in #822
• @haoyu-haoyu made their first contribution in #832
• @RussellSB made their first contribution in #943
• @aahammer-eu made their first contribution in #942
• @AllenDowney made their first contribution in #680
• @Mari1988 made their first contribution in #993
• @twiecki made their first contribution in #1014

Full Changelog: 0.8.0...0.9.0

---

Experimental features and known limitations

The pipeline API is usable, but its newer diagnostic workflows are still evolving. Treat PlaceboInTime results as exploratory: fold selection and end-to-end random-seed reproducibility have known limitations.

The optional pymc-forecast backend for Interrupted Time Series is experimental. Core fitting and impact estimation are supported, but .summary() is not yet supported for forecast-backed models.

BayesianBasisExpansionTimeSeries default priors may be unsuitable for large-magnitude outcomes; inspect fitted counterfactuals and use scale-appropriate priors.