igerber · wenddymacro · Aug 6, 2026 · Aug 6, 2026 · Aug 6, 2026 · Aug 6, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -8,6 +8,123 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- **R/Python parity fixtures for the new compatibility layers.** The
+  `twfeweights`, `ptetools`, and linear `badcontrols` paths now execute the
+  installed R packages on identical CSV fixtures and compare point estimates,
+  weights, and standard errors at `1e-8` tolerance.
+- **Initial R `badcontrols` port.** Added the explicit two-period
+  ``didbc`` / ``imputation_bad_control`` API, the parametric doubly robust
+  score, random-forest cross-fitting path, influence-function output, and ATT
+  extraction helper. Unknown nuisance methods fail closed.
+- Added ``simulate_bad_controls`` with dgp1--dgp5, continuous/binary bad
+  controls, and known group-time treatment effects.
+- Binary bad-control imputation now uses the logistic first stage and
+  Bernoulli-information influence-function correction with R parity coverage.
+- Added staggered linear imputation by looping over estimable ``(g,t)`` cells
+  and aggregating them with cohort/post-period weights.
+- Added staggered parametric and random-forest DR cell loops with the same
+  cohort aggregation surface.
+- Added DR overlap and minimum-treated-group safeguards with explicit
+  imputation fallback.
+- Added seeded cohort-stratified empirical bootstrap, bootstrap SE, and
+  percentile confidence intervals to ``didbc``.
+- Added imputation support for general and bad-control covariate changes.
+- Extended the same covariate-change design matrices to parametric and
+  random-forest DR nuisance models.
+- **R `ptetools` compatibility primitives.** Added panel setup, two-period
+  group-time subsetting, ATT(g,t) influence-function containers, unadjusted
+  DID estimation, the generic ``pte`` group-time loop, and group/dynamic
+  aggregation building blocks. Pre-period covariates now use a conditional
+  AIPW path in ``did_attgt`` and ``pte``. ``pte`` also supports a seeded
+  unit-level empirical bootstrap for overall ATT inference.
+- Added repeated-cross-section ``two_by_two_rcs_subset`` and
+  ``did_rcs_attgt`` compatibility primitives.
+- Added full-history ``keep_all_untreated_subset`` and
+  ``keep_all_pretreatment_subset`` helpers for multi-period estimators.
+- Added ``setup_pte_basic``, ``pte_default``, and ``pte_attgt`` convenience
+  wrappers matching common R ``ptetools`` entry points.
+- Corrected dynamic/group aggregation to use treated-unit cohort weights and
+  normalize dynamic weights separately at each event time.
+- Expanded ``PTEResults`` and ``PTEAggregateResult`` with summary,
+  serialization, weight tables, and bootstrap-distribution surfaces.
+- Added percentile ``overall_conf_int`` to bootstrapped ``PTEResults``.
+- Added ``crit_val_checks`` simultaneous-band fallback utility.
+- Added ``group_time_att`` and ``aggte_obj`` result-container factories.
+- Added the R-compatible ``attgt_noif`` result-container factory for custom
+  group-time estimators without influence functions.
+- Added ``covid_attgt``, reusing the DRDID-validated doubly-robust core for
+  Callaway--Li levels and first-difference outcomes.
+- Added ``ggpte_cont`` as a matplotlib/Plotly-compatible wrapper for
+  ``DoseResult`` curves.
+- Added ``ggpte`` as an event-study plotting wrapper for ``PTEResults``.
+- Extended ``PTEResults.aggregate()`` with influence-function standard errors,
+  normal-based confidence intervals, and ``to_dataframe(level=...)`` views for
+  ATT(g,t), group, and dynamic results.
+- Added ``plot_qtt`` for overall and dynamic QTT visualization.
+- Added ``mp_weights_obj`` and broader ``ggtwfeweights`` result support, plus
+  the R-style ``dr_ml_attgt`` bad-controls cell wrapper.
+- Added callback hooks to ``pte`` for custom setup, subset, ATT(g,t), and
+  aggregation functions.
+- Added explicit Python counterparts for the R ``autoplot``/``plot`` methods
+  on PTE, QTT, empirical-bootstrap, and dose-response result objects.
+- Added DRDID covariate adjustment to ``did_rcs_attgt`` and optional
+  multiplier-bootstrap bands to dynamic ``PTEResults.aggregate()`` results.
+- Added ``process_att_gt`` and ``attgt_pte_aggregations`` aggregation aliases.
+- **Cross-fit parametric DR bad-control score** matching R's `dr_ml_attgt`:
+  per-arm fold splits, OLS/logit cross-fitted nuisances, a `fold_ids` ingress
+  for exact shared-fold R/Python parity, and the `max(propensity)>0.99`
+  imputation fallback.
+- Documented dropped-parity surfaces as REGISTRY deviations: `implicit_twfe_weights`
+  (R `fixest` segfaults on the parity fixture) and `did_post_lasso` (incomplete R
+  source with a `browser()` debug path) are verified for internal consistency, not
+  byte-level R parity.
+- Added tutorial 28 (`docs/tutorials/28_did_badcontrols.ipynb`) demonstrating the
+  R-package ports end-to-end: `pte` group-time ATT on the Callaway--Sant'Anna panel,
+  `twfe_weights` negative-weight decomposition, and `didbc` bad-control estimates on
+  `simulate_bad_controls` output.
+- Added ``DoseResult``, ``dose_obj``, and ``pte_dose_results`` containers for
+  dose-response outputs.
+- Added ``process_dose_gt``, which combines per-cell dose results into
+  ATT(d) / ACRT(d) curves and overall ATT/ACRT with multiplier-bootstrap
+  standard errors and (optionally) simultaneous critical values, plus a
+  ``bspline_basis`` helper that reproduces ``splines2::bSpline``/``dbs``
+  design matrices and ``mboot_se_and_crit`` for R-style sup-t inference.
+- **R `ptetools` QTT/QoTT compatibility.** Ported the quantile-treatment-effects
+  machinery: ``compute_pte`` runs the ``(g,t)`` loop over F0/F1 cumulative
+  distribution functions, ``qtt_pte_aggregations`` / ``qott_pte_aggregations``
+  mix per-cell CDFs into overall / dynamic / group quantile curves (with a
+  documented fix for R's latent merge-reorder misalignment), and
+  ``qtt_empirical_bootstrap`` derives pointwise and uniform (sup-t) confidence
+  bands from seeded unit-level block bootstrap. ``pte_qtt`` / ``PTEQTTResult``
+  hold the resulting curves and ``block_boot_sample`` resamples a panel by unit.
+  Numeric parity with R ``ptetools`` verified on single- and two-cohort panels.
+- Added ``panel=False`` repeated-cross-section support to the generic ``pte``
+  loop.
+- Added ``panel=False`` repeated-cross-section support to the generic ``pte`` loop.
+- Added public ``panel_empirical_bootstrap`` and ``mboot2`` bootstrap helpers.
+- Added R-style ``pte_params``, ``pte_results``, and ``pte_emp_boot`` aliases.
+- **R `twfeweights` compatibility layer.** Added `twfe_weights`, `attO_weights`,
+  `att_simple_weights`, `MPWeightsResult`, and `ggtwfeweights` for decomposing
+  ATT(g,t) tables with R-compatible output columns and normalization rules,
+  plus effective-sample-size and covariate-spread diagnostics.
+- Added the two-period ``two_period_reg_weights`` Frisch-Waugh-Lovell
+  implicit-weight calculation with time-varying changes and time-invariant
+  covariates.
+- Added the two-period ``two_period_aipw_weights`` ATT and implicit control
+  weights calculation with optional pre-period covariates.
+- Added the no-covariate multi-period ``implicit_twfe_weights`` group-time
+  decomposition and pre-trends contribution surface.
+- Added local ``GTWeightsResult`` objects and
+  ``implicit_twfe_weights_gt`` / ``combine_twfe_weights_gt`` accessors.
+- Added ``twfe_cov_bal`` and ``twfe_cov_bal_gt`` covariate-balance diagnostics.
+- Added ``aipw_cov_bal`` and ``aipw_cov_bal_gt`` balance diagnostics.
+- Added ``mp_covariate_bal_summary_helper`` for weighted multi-period balance
+  summaries.
+- Added ``gt_weights`` and ``two_period_covs_obj`` result-object factories.
+- Added ``did_post_lasso`` with post-Lasso outcome/propensity selection and
+  normalized AIPW influence-function inference.
+- Added ``did_post_lasso_ra`` outcome-only regression-adjustment variant.
+- Added multi-period ``implicit_aipw_weights`` group-time decomposition.
 - **ContinuousDiD post-fit `aggregate()` - a MIXED view/recompute adopter**
   (v4 program 2(b) PR-3c; ledger row [M-025]). `ContinuousDiD.fit(aggregate=)`
   is deprecated (`FutureWarning`; removed in 4.0, and the no-underscore

diff --git a/README.md b/README.md
@@ -130,6 +130,7 @@ Full guide: `diff_diff.get_llm_guide("practitioner")`.
 - [Honest DiD](https://diff-diff.readthedocs.io/en/stable/api/honest_did.html) - Rambachan & Roth (2023) sensitivity analysis: robust CI under PT violations, breakdown values
 - [Pre-Trends Power Analysis](https://diff-diff.readthedocs.io/en/stable/api/pretrends.html) - Roth (2022) minimum detectable violation and power curves
 - [Power Analysis](https://diff-diff.readthedocs.io/en/stable/api/power.html) - analytical and simulation-based MDE, sample size, power curves for study design
+- [TWFE ATT(g,t) weights](https://diff-diff.readthedocs.io/en/stable/api/twfeweights.html) - R `twfeweights`-compatible decompositions for TWFE, overall ATT, and simple ATT weights
 - [MMM Calibration Export](https://diff-diff.readthedocs.io/en/stable/api/mmm.html) - convert experiment results into MMM calibration inputs: PyMC-Marketing lift-test frames and Google Meridian lognormal ROI priors
 - Conley spatial HAC SE (`vcov_type="conley"`) on cross-sectional `LinearRegression` / `compute_robust_vcov` plus panel `DifferenceInDifferences` / `MultiPeriodDiD` / `TwoWayFixedEffects` (with `conley_lag_cutoff` for within-unit Bartlett temporal HAC) - Conley (1999) spatial-correlation-aware SEs with parity vs R `conleyreg` on cross-sectional + panel fixtures, optional combined spatial + cluster product kernel via explicit `cluster=`, auto-activating sparse k-d-tree fast path for `n > 5_000`
 

diff --git a/diff_diff/__init__.py b/diff_diff/__init__.py
@@ -39,6 +39,17 @@
     Comparison2x2,
     bacon_decompose,
 )
+from diff_diff.badcontrols import (
+    BadControlsResult,
+    didbc,
+    dr_ml_attgt,
+    dr_ml_bad_control,
+    extract_att,
+    imputation_bad_control,
+    simulate_bad_controls,
+    staggered_dr_bad_control,
+    staggered_imputation_bad_control,
+)
 from diff_diff.business_report import (
     BUSINESS_REPORT_SCHEMA_VERSION,
     BusinessContext,
@@ -216,6 +227,65 @@
     TreatmentDoseShape,
     profile_panel,
 )
+from diff_diff.ptetools import (
+    ATTGTResult,
+    DoseResult,
+    GTDataFrame,
+    PTEAggregateResult,
+    PTEParams,
+    PTEQTTResult,
+    PTEResults,
+    TwoByTwoSubset,
+    aggte_obj,
+    attgt_if,
+    attgt_noif,
+    attgt_pte_aggregations,
+    autoplot_dose_obj,
+    autoplot_pte_emp_boot,
+    autoplot_pte_qtt,
+    autoplot_pte_results,
+    block_boot_sample,
+    bspline_basis,
+    compute_pte,
+    covid_attgt,
+    crit_val_checks,
+    did_attgt,
+    did_rcs_attgt,
+    dose_obj,
+    ggpte,
+    ggpte_cont,
+    group_time_att,
+    gt_data_frame,
+    keep_all_pretreatment_subset,
+    keep_all_untreated_subset,
+    mboot2,
+    mboot_se_and_crit,
+    overall_weights,
+    panel_empirical_bootstrap,
+    plot_dose_obj,
+    plot_pte_emp_boot,
+    plot_pte_qtt,
+    plot_pte_results,
+    plot_qtt,
+    process_att_gt,
+    process_dose_gt,
+    pte,
+    pte_aggte,
+    pte_attgt,
+    pte_default,
+    pte_dose_results,
+    pte_emp_boot,
+    pte_params,
+    pte_qtt,
+    pte_results,
+    qott_pte_aggregations,
+    qtt_empirical_bootstrap,
+    qtt_pte_aggregations,
+    setup_pte,
+    setup_pte_basic,
+    two_by_two_rcs_subset,
+    two_by_two_subset,
+)
 from diff_diff.rdd import (
     RegressionDiscontinuity,
     RegressionDiscontinuityResults,
@@ -283,6 +353,38 @@
     TROPResults,
     trop,
 )
+from diff_diff.twfeweights import (
+    GTWeightsResult,
+    ImplicitAIPWResult,
+    ImplicitTWFEResult,
+    MPWeightsResult,
+    PostLassoResult,
+    TwoPeriodCovariatesResult,
+    aipw_cov_bal,
+    aipw_cov_bal_gt,
+    att_simple_weights,
+    attO_weights,
+    combine_twfe_weights_gt,
+    did_post_lasso,
+    did_post_lasso_ra,
+    effective_sample_size,
+    frac_treated_extreme,
+    ggtwfeweights,
+    gt_weights,
+    implicit_aipw_weights,
+    implicit_twfe_weights,
+    implicit_twfe_weights_gt,
+    log_ratio_sd,
+    mp_covariate_bal_summary_helper,
+    mp_weights_obj,
+    pooled_sd,
+    twfe_cov_bal,
+    twfe_cov_bal_gt,
+    twfe_weights,
+    two_period_aipw_weights,
+    two_period_covs_obj,
+    two_period_reg_weights,
+)
 from diff_diff.two_stage import (
     TwoStageBootstrapResults,
     TwoStageDiD,
@@ -416,6 +518,37 @@
     "TWFEWeightsResult",
     "chaisemartin_dhaultfoeuille",
     "twowayfeweights",
+    # R twfeweights compatibility
+    "MPWeightsResult",
+    "PostLassoResult",
+    "ImplicitTWFEResult",
+    "ImplicitAIPWResult",
+    "GTWeightsResult",
+    "TwoPeriodCovariatesResult",
+    "twfe_weights",
+    "twfe_cov_bal",
+    "twfe_cov_bal_gt",
+    "aipw_cov_bal",
+    "aipw_cov_bal_gt",
+    "mp_covariate_bal_summary_helper",
+    "mp_weights_obj",
+    "attO_weights",
+    "att_simple_weights",
+    "ggtwfeweights",
+    "effective_sample_size",
+    "pooled_sd",
+    "log_ratio_sd",
+    "frac_treated_extreme",
+    "gt_weights",
+    "did_post_lasso",
+    "did_post_lasso_ra",
+    "two_period_reg_weights",
+    "implicit_twfe_weights",
+    "implicit_aipw_weights",
+    "implicit_twfe_weights_gt",
+    "combine_twfe_weights_gt",
+    "two_period_aipw_weights",
+    "two_period_covs_obj",
     # WooldridgeDiD (ETWFE)
     "WooldridgeDiD",
     "WooldridgeDiDResults",
@@ -502,6 +635,57 @@
     "PreTrendsPowerCurve",
     "compute_pretrends_power",
     "compute_mdv",
+    # ptetools compatibility primitives
+    "PTEParams",
+    "GTDataFrame",
+    "TwoByTwoSubset",
+    "ATTGTResult",
+    "PTEAggregateResult",
+    "PTEResults",
+    "DoseResult",
+    "setup_pte",
+    "setup_pte_basic",
+    "gt_data_frame",
+    "keep_all_pretreatment_subset",
+    "keep_all_untreated_subset",
+    "two_by_two_subset",
+    "two_by_two_rcs_subset",
+    "attgt_if",
+    "attgt_noif",
+    "covid_attgt",
+    "aggte_obj",
+    "autoplot_dose_obj",
+    "autoplot_pte_emp_boot",
+    "autoplot_pte_qtt",
+    "autoplot_pte_results",
+    "attgt_pte_aggregations",
+    "did_attgt",
+    "crit_val_checks",
+    "group_time_att",
+    "ggpte",
+    "ggpte_cont",
+    "process_att_gt",
+    "process_dose_gt",
+    "bspline_basis",
+    "mboot_se_and_crit",
+    "pte_attgt",
+    "did_rcs_attgt",
+    "dose_obj",
+    "overall_weights",
+    "mboot2",
+    "panel_empirical_bootstrap",
+    "pte_aggte",
+    "pte",
+    "pte_default",
+    "pte_dose_results",
+    "pte_params",
+    "plot_qtt",
+    "plot_dose_obj",
+    "plot_pte_emp_boot",
+    "plot_pte_qtt",
+    "plot_pte_results",
+    "pte_results",
+    "pte_emp_boot",
     # Survey support
     "SurveyDesign",
     "SurveyMetadata",
@@ -590,6 +774,15 @@
     "Diagnostic",
     "EventStudyResults",
     "AggregationResult",
+    "BadControlsResult",
+    "didbc",
+    "dr_ml_bad_control",
+    "dr_ml_attgt",
+    "extract_att",
+    "imputation_bad_control",
+    "simulate_bad_controls",
+    "staggered_imputation_bad_control",
+    "staggered_dr_bad_control",
 ]
 
 # Agent-facing entrypoints surface first in dir(diff_diff). LLM agents

diff --git a/diff_diff/_deprecation.py b/diff_diff/_deprecation.py
@@ -52,6 +52,13 @@ class _NotSupplied:
     def __repr__(self) -> str:  # pragma: no cover - debugging aid
         return "<not supplied>"
 
+    def __copy__(self) -> "_NotSupplied":
+        return self
+
+    def __deepcopy__(self, memo: dict[int, object]) -> "_NotSupplied":
+        memo[id(self)] = self
+        return self
+
 
 NOT_SUPPLIED = _NotSupplied()