Release v0.4.0: Major module architecture overhaul

- DID: Add DDD, did_analysis() workflow, 8 plot functions
- Synth: Modular rewrite with demeaned/robust/gsynth/staggered/conformal variants
- Panel: Hausman, BP-LM, Pesaran CD, Wooldridge tests; diagnostic plots
- RD: Comprehensive rd_diagnostics() suite
- IV: Rewritten with proper first-stage diagnostics and weak IV detection
- Matching: Add CEM, optimal matching, genetic matching
- DAG: 15+ example DAGs, dag_simulate(), backdoor/frontdoor identification
- AI Registry: Expanded function schemas for LLM tool-use
- CausalResult: Added .to_json(), .to_dict()

884 tests passing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: brycewang-stanford

.gitignore

@@ -164,4 +164,5 @@ Thumbs.db
 # Files
 Start Superpower for R.md
 
-test-notebooks/
+test-notebooks/
+StatsPAI/statspai-推文-agent时代的统计包生态.md

README.md

@@ -6,7 +6,7 @@
 [![Tests](https://github.com/brycewang-stanford/statspai/workflows/CI%2FCD%20Pipeline/badge.svg)](https://github.com/brycewang-stanford/statspai/actions)
 [![PyPI Downloads](https://static.pepy.tech/personalized-badge/statspai?period=total&units=INTERNATIONAL_SYSTEM&left_color=BLACK&right_color=GREEN&left_text=downloads)](https://pepy.tech/projects/statspai)
 
-StatsPAI is a unified Python package for causal inference and applied econometrics. One `import`, 150+ functions, covering the complete empirical research workflow — from classical econometrics to cutting-edge ML/AI causal methods to publication-ready tables in Word, Excel, and LaTeX.
+StatsPAI is a unified Python package for causal inference and applied econometrics. One `import`, 170+ functions, covering the complete empirical research workflow — from classical econometrics to cutting-edge ML/AI causal methods to publication-ready tables in Word, Excel, and LaTeX.
 
 It brings R's [Causal Inference Task View](https://cran.r-project.org/web/views/CausalInference.html) (fixest, did, rdrobust, gsynth, DoubleML, MatchIt, CausalImpact, ...) and Stata's core econometrics commands into a single, consistent Python API.
 
@@ -344,7 +344,7 @@ sp.subgroup_analysis(df, formula="wage ~ education + experience",
 ## API at a Glance
 
 ```text
-150+ public functions/classes
+170+ public functions/classes
 
 Regression:     regress, ivreg, panel, heckman, qreg, sqreg, tobit, xtabond
 DID:            did, did_2x2, callaway_santanna, sun_abraham, bacon_decomposition, honest_did
@@ -373,6 +373,67 @@ Output:         modelsummary, outreg2, sumstats, balance_table, tab, coefplot, b
 
 ---
 
+## Release Notes
+
+### v0.4.0 (2026-04-05) — Module Architecture Overhaul
+
+**Major refactoring and expansion of core modules (+5,800 lines of new code):**
+
+- **DID**: Added Triple Differences (`ddd()`), one-call `did_analysis()` workflow (auto design detection → Bacon decomposition → estimation → event study → sensitivity), and 8 publication-ready plot functions (`parallel_trends_plot`, `bacon_plot`, `group_time_plot`, `enhanced_event_study_plot`, `treatment_rollout_plot`, `sensitivity_plot`, `cohort_event_study_plot`)
+- **Synthetic Control**: Modular rewrite — `demeaned_synth()`, `robust_synth()` (penalized SCM), `gsynth()` (Generalized SCM with interactive fixed effects), `staggered_synth()` (multi-unit staggered adoption), `conformal_synth()` (distribution-free inference), and comprehensive `synth_plot()` / `synth_weight_plot()` / `synth_gap_plot()`
+- **Panel**: Major expansion of `panel()` — Hausman test, Breusch-Pagan LM, Pesaran CD, Wooldridge autocorrelation, panel unit root tests; added `panel_summary_plot()`, `fe_plot()`, `re_comparison_plot()`
+- **RD**: New `rd_diagnostics()` suite — bandwidth sensitivity, placebo cutoffs, donut-hole robustness, covariate balance at cutoff, density test
+- **IV / 2SLS**: Rewritten `ivreg()` with proper first-stage diagnostics (Cragg-Donald, Kleibergen-Paap), weak IV detection, Sargan-Hansen overidentification test, Anderson canonical correlation test, Stock-Yogo critical values
+- **Matching**: Enhanced `match()` — added CEM (Coarsened Exact Matching), optimal matching, genetic matching; improved balance diagnostics with Love plot and standardized mean difference
+- **DAG**: Expanded `dag()` with 15+ built-in example DAGs (`dag_example()`), `dag_simulate()` for data generation from causal graphs, backdoor/frontdoor criterion identification
+- **Causal Impact**: Enhanced Bayesian structural time-series with automatic model selection and improved inference
+- **AI Agent Registry**: Expanded `list_functions()`, `describe_function()`, `function_schema()`, `search_functions()` for LLM/agent tool-use integration
+- **CausalResult**: Added `.to_json()`, `.to_dict()`, enhanced `.summary()` formatting
+
+### v0.3.1 (2025-12-20)
+
+- Fix PyPI badge displaying stale version
+
+### v0.3.0 (2025-12-20) — ML & Advanced Causal Methods
+
+- **Meta-Learners**: S/T/X/R/DR-Learner for CATE estimation with `compare_metalearners()` and CATE diagnostics (`gate_test`, `blp_test`)
+- **Neural Causal Models**: TARNet, CFRNet, DragonNet for deep CATE estimation
+- **Causal Discovery**: `notears()` (continuous DAG optimization), `pc_algorithm()` (constraint-based)
+- **TMLE**: Targeted Maximum Likelihood Estimation with Super Learner
+- **Policy Learning**: `policy_tree()` optimal treatment rules, `policy_value()` evaluation
+- **Conformal Causal**: Distribution-free prediction intervals for ITE
+- **Bayesian Causal Forest**: `bcf()` with separate prognostic/treatment functions
+- **Dose-Response**: Continuous treatment GPS curves
+- **Bounds**: Lee bounds (sample selection), Manski bounds (partial identification)
+- **Interference**: `spillover()` direct + indirect effect decomposition
+- **DTR**: `g_estimation()` multi-stage optimal treatment regimes
+- **Multi-Treatment**: AIPW for multi-valued treatments
+- **Bunching**: Kink/notch bunching estimator with elasticity
+- **Matrix Completion**: `mc_panel()` nuclear-norm panel estimator
+- **Robustness**: `spec_curve()`, `robustness_report()`, `subgroup_analysis()`
+- **New Regression**: DeepIV, Heckman selection, quantile regression, Tobit, Arellano-Bond GMM
+- **New Diagnostics**: E-value, Anderson-Rubin weak IV test, Sensemakr, RD density test
+- **Other**: Entropy balancing, Sun-Abraham event study, Bacon decomposition, HonestDiD
+
+### v0.2.0 (2025-11-15) — Post-Estimation & Output
+
+- **Post-Estimation**: `margins()`, `marginsplot()`, `test()`, `lincom()`
+- **Output Tables**: `modelsummary()`, `outreg2()`, `sumstats()`, `balance_table()`, `tab()`, `coefplot()`, `binscatter()`
+- **Inference**: `wild_cluster_bootstrap()`, `aipw()`, `ri_test()`
+- **New Modules**: DML, Causal Forest, Matching (PSM/Mahalanobis), Synthetic Control (ADH + SDID), Panel (FE/RE/FD), Causal Impact, Mediation, Bartik IV
+- **Diagnostics**: `oster_bounds()`, `mccrary_test()`, `hausman_test()`, `het_test()`, `reset_test()`, `vif()`
+- **Utilities**: Variable labeling, `describe()`, `pwcorr()`, `winsor()`, `read_data()`
+
+### v0.1.0 (2025-10-01) — Initial Release
+
+- Core regression: `regress()` OLS with robust/clustered/HAC standard errors
+- Instrumental variables: `ivreg()` 2SLS
+- Difference-in-Differences: `did()`, `did_2x2()`, `callaway_santanna()`
+- Regression discontinuity: `rdrobust()`
+- Unified `CausalResult` object with `.summary()`, `.plot()`, `.to_latex()`, `.to_docx()`, `.cite()`
+
+---
+
 ## About
 
 **StatsPAI Inc.** is the research infrastructure company behind [CoPaper.AI](https://copaper.ai) — the AI co-authoring platform for empirical research, born out of Stanford's [REAP](https://reap.fsi.stanford.edu/) program.
@@ -405,7 +466,7 @@ pytest
   author={Wang, Bryce},
   year={2025},
   url={https://github.com/brycewang-stanford/statspai},
-  version={0.3.1}
+  version={0.4.0}
 }
 ```
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "StatsPAI"
-version = "0.3.1"
+version = "0.4.0"
 description = "The Causal Inference & Econometrics Toolkit for Python"
 readme = "README.md"
 license = {text = "MIT"}

src/statspai/__init__.py

@@ -22,27 +22,35 @@
 >>> sp.outreg2(result, filename="results.xlsx")
 """
 
-__version__ = "0.3.1"
+__version__ = "0.4.0"
 __author__ = "Bryce Wang"
 __email__ = "bryce@copaper.ai"
 
 from .core.results import EconometricResults, CausalResult
 from .regression.ols import regress
-from .regression.iv import ivreg, IVRegression
+from .regression.iv import iv, ivreg, IVRegression
 from .causal.causal_forest import CausalForest, causal_forest
-from .did import did, did_2x2, callaway_santanna, sun_abraham, bacon_decomposition, honest_did, breakdown_m, event_study
-from .rd import rdrobust, rdplot
+from .did import (
+    did, did_2x2, ddd, callaway_santanna, sun_abraham,
+    bacon_decomposition, honest_did, breakdown_m, event_study,
+    did_analysis, DIDAnalysis,
+    parallel_trends_plot, bacon_plot, group_time_plot, did_plot,
+    enhanced_event_study_plot, treatment_rollout_plot,
+    sensitivity_plot, cohort_event_study_plot,
+)
+from .rd import rdrobust, rdplot, rdplotdensity, rdbwsensitivity, rdbalance, rdplacebo, rdsummary
 from .synth import (
-    synth, SyntheticControl, sdid, augsynth,
+    synth, SyntheticControl, synthplot, sdid, augsynth,
+    demeaned_synth, robust_synth, gsynth, staggered_synth, conformal_synth,
     synthdid_estimate, sc_estimate, did_estimate,
     synthdid_placebo, synthdid_plot, synthdid_units_plot, synthdid_rmse_plot,
     california_prop99,
 )
-from .matching import match, MatchEstimator, ebalance
+from .matching import match, MatchEstimator, ebalance, balanceplot, psplot
 from .dml import dml, DoubleML
 from .deepiv import deepiv, DeepIV
-from .panel import panel, PanelRegression
-from .causal_impact import causal_impact, CausalImpactEstimator
+from .panel import panel, panel_compare, PanelResults, PanelCompareResults, PanelRegression
+from .causal_impact import causal_impact, CausalImpactEstimator, impactplot
 from .mediation import mediate, MediationAnalysis
 from .bartik import bartik, BartikIV
 from .output.outreg2 import OutReg2, outreg2
@@ -76,7 +84,7 @@
 from .multi_treatment import multi_treatment, MultiTreatment
 from .robustness import spec_curve, SpecCurveResult, robustness_report, RobustnessResult, subgroup_analysis, SubgroupResult
 from .survey import svydesign, SurveyDesign, svymean, svytotal, svyglm
-from .dag import dag, DAG
+from .dag import dag, DAG, dag_example, dag_examples, dag_example_positions, dag_simulate
 from .registry import list_functions, describe_function, function_schema, search_functions, all_schemas
 
 __all__ = [
@@ -85,23 +93,45 @@
     "CausalResult",
     # Regression
     "regress",
+    "iv",
     "ivreg",
     "IVRegression",
     # DID
     "did",
     "did_2x2",
+    "ddd",
     "callaway_santanna",
     "sun_abraham",
     "bacon_decomposition",
     "honest_did",
     "breakdown_m",
     "event_study",
+    "did_analysis",
+    "DIDAnalysis",
+    "parallel_trends_plot",
+    "bacon_plot",
+    "group_time_plot",
+    "did_plot",
+    "enhanced_event_study_plot",
+    "treatment_rollout_plot",
+    "sensitivity_plot",
+    "cohort_event_study_plot",
     # RD
     "rdrobust",
     "rdplot",
+    "rdplotdensity",
+    "rdbwsensitivity",
+    "rdbalance",
+    "rdplacebo",
+    "rdsummary",
     # Synthetic Control
     "synth",
     "SyntheticControl",
+    "demeaned_synth",
+    "robust_synth",
+    "gsynth",
+    "staggered_synth",
+    "conformal_synth",
     "sdid",
     "synthdid_estimate",
     "sc_estimate",
@@ -116,6 +146,8 @@
     "match",
     "MatchEstimator",
     "ebalance",
+    "balanceplot",
+    "psplot",
     # Double ML
     "dml",
     "DoubleML",
@@ -124,6 +156,9 @@
     "DeepIV",
     # Panel
     "panel",
+    "panel_compare",
+    "PanelResults",
+    "PanelCompareResults",
     "PanelRegression",
     # Causal Impact
     "causal_impact",
@@ -278,6 +313,10 @@
     # DAG
     "dag",
     "DAG",
+    "dag_example",
+    "dag_examples",
+    "dag_example_positions",
+    "dag_simulate",
     # AI / Agent Registry
     "list_functions",
     "describe_function",

src/statspai/causal_impact/__init__.py

@@ -14,6 +14,6 @@
 *Annals of Applied Statistics*, 9(1), 247-274.
 """
 
-from .impact import causal_impact, CausalImpactEstimator
+from .impact import causal_impact, CausalImpactEstimator, impactplot
 
-__all__ = ['causal_impact', 'CausalImpactEstimator']
+__all__ = ['causal_impact', 'CausalImpactEstimator', 'impactplot']

src/statspai/causal_impact/impact.py

@@ -326,6 +326,135 @@ def _predict(self, model, X_full):
         return Y_pred, Y_pred_se
 
 
+# ------------------------------------------------------------------
+# Plotting
+# ------------------------------------------------------------------
+
+def impactplot(
+    result: CausalResult,
+    type: str = 'all',
+    ax=None,
+    figsize: tuple = (12, 9),
+    title=None,
+):
+    """
+    Causal Impact visualization (Google-style 3-panel plot).
+
+    Parameters
+    ----------
+    result : CausalResult
+        Result from ``causal_impact()``.
+    type : str, default 'all'
+        'all': 3-panel (original + pointwise + cumulative).
+        'original': actual vs counterfactual.
+        'pointwise': pointwise causal effect.
+        'cumulative': cumulative effect.
+    ax : matplotlib Axes, optional
+        Only for single-panel types.
+    figsize : tuple
+    title : str, optional
+
+    Returns
+    -------
+    (fig, ax) or (fig, axes)
+    """
+    try:
+        import matplotlib.pyplot as plt
+    except ImportError:
+        raise ImportError("matplotlib required.")
+
+    mi = result.model_info
+    detail = result.detail
+    if detail is None or 'actual' not in detail.columns:
+        raise ValueError("No detail table. Use causal_impact() result.")
+
+    times = detail['time'].values
+    actual = detail['actual'].values
+    predicted = detail['predicted'].values
+    effect = detail['effect'].values
+    eff_lo = detail['effect_lower'].values
+    eff_hi = detail['effect_upper'].values
+    post = detail['post_intervention'].values
+    intervention_time = mi.get('intervention_time')
+    cum_raw = mi.get('cumulative_effect')
+    if cum_raw is not None and len(cum_raw) < len(times):
+        # cumulative_effect only covers post-period; pad pre with zeros
+        cumulative = np.zeros(len(times))
+        cumulative[post] = cum_raw
+    elif cum_raw is not None:
+        cumulative = cum_raw
+    else:
+        cumulative = np.cumsum(effect * post)
+
+    if type == 'all':
+        fig, axes = plt.subplots(3, 1, figsize=figsize, sharex=True)
+        _original_panel(axes[0], times, actual, predicted,
+                        intervention_time, result.alpha)
+        _pointwise_panel(axes[1], times, effect, eff_lo, eff_hi,
+                         post, intervention_time)
+        _cumulative_panel(axes[2], times, cumulative, post,
+                          intervention_time)
+        fig.suptitle(title or 'Causal Impact Analysis', fontsize=14, y=1.01)
+        fig.tight_layout()
+        return fig, axes
+
+    if ax is None:
+        fig, ax = plt.subplots(figsize=(figsize[0], figsize[1] // 3 + 1))
+    else:
+        fig = ax.get_figure()
+
+    if type == 'pointwise':
+        _pointwise_panel(ax, times, effect, eff_lo, eff_hi,
+                         post, intervention_time)
+    elif type == 'cumulative':
+        _cumulative_panel(ax, times, cumulative, post, intervention_time)
+    else:
+        _original_panel(ax, times, actual, predicted,
+                        intervention_time, result.alpha)
+
+    ax.set_title(title or f'Causal Impact: {type}', fontsize=13)
+    fig.tight_layout()
+    return fig, ax
+
+
+def _original_panel(ax, times, actual, predicted, t0, alpha):
+    ax.plot(times, actual, color='#2C3E50', linewidth=1.5, label='Observed')
+    ax.plot(times, predicted, color='#3498DB', linewidth=1.5,
+            linestyle='--', label='Counterfactual')
+    if t0 is not None:
+        ax.axvline(x=t0, color='#E74C3C', linestyle=':', linewidth=1,
+                   alpha=0.7, label='Intervention')
+    ax.set_ylabel('Value', fontsize=11)
+    ax.legend(fontsize=9, frameon=False)
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+
+
+def _pointwise_panel(ax, times, effect, lo, hi, post, t0):
+    ax.plot(times, effect, color='#2C3E50', linewidth=1.5)
+    ax.fill_between(times, lo, hi, alpha=0.15, color='#3498DB')
+    ax.axhline(y=0, color='gray', linestyle='--', linewidth=0.8)
+    if t0 is not None:
+        ax.axvline(x=t0, color='#E74C3C', linestyle=':', linewidth=1,
+                   alpha=0.7)
+    ax.set_ylabel('Pointwise Effect', fontsize=11)
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+
+
+def _cumulative_panel(ax, times, cumulative, post, t0):
+    ax.plot(times, cumulative, color='#2C3E50', linewidth=1.5)
+    ax.fill_between(times, 0, cumulative, alpha=0.15, color='#3498DB')
+    ax.axhline(y=0, color='gray', linestyle='--', linewidth=0.8)
+    if t0 is not None:
+        ax.axvline(x=t0, color='#E74C3C', linestyle=':', linewidth=1,
+                   alpha=0.7)
+    ax.set_xlabel('Time', fontsize=11)
+    ax.set_ylabel('Cumulative Effect', fontsize=11)
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+
+
 # Citation
 CausalResult._CITATIONS['causal_impact'] = (
     "@article{brodersen2015inferring,\n"