Conversation
…t-study survey composition + sup-t bootstrap Closes the two Phase 4.5 A NotImplementedError gates on design='mass_point' + weights/survey and aggregate='event_study' + weights/survey. Weighted 2SLS in _fit_mass_point_2sls follows the Wooldridge 2010 Ch. 12 pweight convention (w² in HC1 meat, w·u in CR1 cluster score, weighted bread Z'WX). HC1 and CR1 match estimatr::iv_robust bit-exactly at atol=1e-10 (new cross-language golden). Per-unit IF on β̂-scale scales so compute_survey_if_variance(psi, trivial) ≈ V_HC1 at atol=1e-10 (PR #359 convention applied uniformly). Event-study path threads weights + survey through the per-horizon loop, composing Binder-TSL variance per horizon and populating survey_metadata + variance_formula + effective_dose_mean (previously hardcoded None). New _sup_t_multiplier_bootstrap helper reuses generate_survey_multiplier_weights_batch / generate_bootstrap_weights_batch from diff_diff.bootstrap_utils — no custom Rademacher draws, no (1/n) prefactor. At H=1 reduces to Φ⁻¹(1-α/2) ≈ 1.96 (reduction-locked). New __init__ kwargs: n_bootstrap=999, seed=None. New fit() kwarg: cband=True. HeterogeneousAdoptionDiDEventStudyResults gains survey_metadata + variance_formula + effective_dose_mean + cband_* fields, surfaced through to_dict / to_dataframe / summary / __repr__. Unweighted event-study output (att/se) bit-exactly preserved; cband disabled on the unweighted path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
Overall Assessment ⛔ Blocker Executive Summary
Methodology
Code Quality
Performance
Maintainability
Tech Debt
Security
Documentation/Tests
Path to Approval
|
R1 P0 (methodology): cband_low/high now NaN-gated for horizons with se <= 0 or non-finite, matching the safe_inference contract for pointwise output. Previously a horizon with se=0 produced a finite band equal to the point estimate, overstating precision. R1 P1 (methodology): sup-t multiplier bootstrap now applies stratum-centered + small-sample-corrected PSU aggregation before the matmul, so Var_xi(xi @ Psi_psu_scaled) reproduces the analytical Binder-TSL variance term-for-term (V = sum_h (1-f_h)(n_h/(n_h-1)) sum_j (psi_hj - psi_h_bar)^2). Previously the bootstrap omitted centering and the small-sample correction; under stratified designs the critical value diverged from the analytical target. Verified at H=1 G=400 n_strata=4: bootstrap q=1.985 matches Phi^-1(0.975)=1.960. R1 P1 (code quality): event-study weights= filtering now translates data_filtered.index LABELS to POSITIONAL offsets via data.index.get_indexer, so non-RangeIndex inputs work correctly. Previous code treated index labels as positions and silently broke on custom int / string indices. Raises a clear ValueError when index alignment fails (duplicate or malformed labels). R1 P3 (docs): HeterogeneousAdoptionDiDEventStudyResults docstring updated to describe Phase 4.5 B weighted/survey support + new cband_* fields + variance_formula / effective_dose_mean; fit() docstring updated to reflect the Phase 4.5 B dispatch matrix and document the new cband kwarg. Regression tests: - test_zero_se_horizon_nan_gates_cband locks the P0 NaN gate - test_weights_nonrange_index_aligned_positionally locks the P1 row-order contract across RangeIndex vs custom int indices - Existing TestSupTReducesToNormalAtH1 covers the stratum-centered path via the unit-PSU reduction case Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall AssessmentExecutive Summary
Methodology
Code Quality
Performance
Maintainability
Tech Debt
Security
Documentation/Tests
Path to Approval
|
R2 P1 (methodology): reject SurveyDesign(lonely_psu='adjust') with singleton strata in _sup_t_multiplier_bootstrap. The bootstrap helper pools singletons into a pseudo-stratum with NONZERO multipliers, but compute_survey_if_variance centers singleton PSU scores around the global mean — without the matching pseudo-stratum centering transform in the bootstrap, the simultaneous band target diverges from the analytical Binder-TSL variance. Clear NotImplementedError points users to lonely_psu='remove' (matches the 'remove' analytical target) or cband=False (skips bootstrap). 'remove' / 'certainty' continue to work unchanged. Deferred transform tracked for follow-up. R2 P1 (code quality): reject cluster= + weights/survey= on design='mass_point' on both static and event-study paths. The weighted path composes Binder-TSL variance via compute_survey_if_variance which was silently overriding the CR1 sandwich while result metadata still advertised vcov_type='cr1' and cluster_name=<col>. Clean NotImplementedError with a pointer to the two supported combinations: cluster= alone (unweighted CR1) or survey/weights alone (Binder-TSL). Combined cluster-robust survey inference requires a derivation not yet in scope. Regression tests (+4): - test_mass_point_survey_plus_cluster_rejected_static - test_mass_point_survey_plus_cluster_rejected_event_study - test_lonely_psu_adjust_with_singletons_rejected_on_cband - test_stratified_h1_sup_t_matches_analytical (H=1 quantile lock under n_strata=4, psu_per_unit: q=1.985, matches Phi^-1(0.975)) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall AssessmentExecutive Summary
MethodologyAffected methods: Wooldridge-style weighted 2SLS on the HAD mass-point path, Binder-TSL survey composition, and the new sup-t multiplier bootstrap.
Code QualityNo separate findings beyond the methodology issues above. PerformanceNo findings. MaintainabilityNo separate findings beyond the methodology issues above. Tech Debt
SecurityNo findings. Documentation/Tests
Path to Approval
Review based on diff inspection; I did not run the test suite in this environment. |
R3 P1 (methodology — trivial-survey sup-t): _sup_t_multiplier_bootstrap now treats ANY non-None resolved_survey as survey-aware bootstrap input, including the trivial SurveyDesign(weights=...) case (no explicit strata / PSU / FPC). Previously the gate required explicit strata / psu / fpc, so the trivial survey path fell through to raw unit-level Rademacher — Var(xi @ Psi) = Σ psi², NOT the centered (n/(n-1))·Σ(ψ−ψ̄)² target that compute_survey_if_variance uses for the analytical SE. The existing "no strata → single implicit stratum" branch in _sup_t_multiplier_bootstrap now handles the trivial case end-to-end. H=1 trivial-survey regression locks q ≈ 1.96. R3 P1 (methodology — classical weighted mass-point): reject vcov_type='classical' on mass-point whenever the IF matrix is used (static survey= path; event-study survey= OR weights= shortcut with cband=True). The IF is HC1-scaled (sqrt((n-1)/(n-k)) factor in _fit_mass_point_2sls targets compute_survey_if_variance(psi, trivial) ≈ V_HC1[1,1]); mixing it with a classical analytical SE either through the survey Binder-TSL override or the sup-t bootstrap normalization would silently report a V_HC1-targeted SE under a classical vcov_type label. Rejection message points users to hc1. weights= shortcut + cband=False + classical continues to work unchanged (no IF consumption; analytical classical sandwich returned as-is). Regression tests (+4): - test_trivial_survey_h1_sup_t_matches_analytical: H=1 trivial SurveyDesign reduces to Normal quantile, confirms stratum-demean + sqrt(n/(n-1)) correction fires on the trivial case. - test_mass_point_classical_survey_rejected_static - test_mass_point_classical_event_study_with_cband_rejected - test_mass_point_classical_event_study_cband_false_accepts (complement: confirms classical+weights+cband=False still works, no IF consumption) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall AssessmentExecutive Summary
Methodology
Code Quality
Performance
Maintainability
Tech Debt
Security
Documentation/Tests
Path to Approval
Review based on diff inspection only; I could not run the test suite here because the environment is missing project dependencies ( |
R4 P1 (cluster guard overreach): narrow the mass-point cluster+weighted rejection. Previously rejected ANY cluster= + weighted fit, which blocked the valid weights= shortcut + cluster= weighted-CR1 pweight sandwich path (parity-tested vs estimatr::iv_robust se_type='stata'). Narrowed to: - Static path: reject only cluster= + survey= (Binder-TSL override would silently overwrite CR1 while metadata says vcov_type='cr1'). - Event-study path: reject cluster= + survey= OR cluster= + weights= + cband=True (sup-t bootstrap mixes HC1-scale perturbations with CR1 analytical SE). cluster= + weights= + cband=False is allowed (no IF consumption). R4 P1 (lonely_psu="adjust" deviation): added REGISTRY.md Note explicitly documenting the HAD-specific unsupported combo (weighted event-study + cband=True + lonely_psu="adjust" with singleton strata). The shared survey bootstrap helper supports pooled singletons but compute_survey_if_variance centers them at the global mean; matching the two requires a pseudo-stratum centering transform not yet derived. Other survey-bootstrap consumers (CS, dCDH, SDID) retain full "adjust" support. R4 P2 (n_units reporting): weighted event-study now reports the POSITIVE-WEIGHT contributing sample size in n_units and n_obs_per_horizon, matching the static-path n_obs contract. Full- design size still surfaces through survey_metadata.n_psu / effective_n / etc. R4 P2 (docstring drift): HeterogeneousAdoptionDiDEventStudyResults.se docstring now documents the three regimes separately — unweighted analytical sandwich, weights= shortcut analytical (HC1 / classical / CR1), and survey= Binder-TSL. Previously said "Binder-TSL for all weighted fits" which was wrong for the weights= shortcut. Regression tests (+5): - test_mass_point_weights_plus_cluster_shortcut_allowed - test_mass_point_weights_plus_cluster_event_study_cband_false_allowed - test_mass_point_weights_plus_cluster_event_study_cband_true_rejected - test_event_study_zero_weight_units_excluded_from_n_units - Updated test_mass_point_survey_plus_cluster_rejected_static to use survey= instead of weights= (matches narrowed guard) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall AssessmentExecutive Summary
Methodology
Code Quality
Performance
Maintainability
Tech Debt
Security
Documentation/Tests
Path to Approval
|
R5 P1 (classical-guard bypass): the classical-on-IF-consumption rejection was only checking the RAW vcov_type kwarg, which missed the default mapping (vcov_type=None, robust=False → classical). Re-ordered both mass-point branches (static + event-study) to resolve vcov_requested first, then gate the rejection on the resolved value. Default-path users who pass HeterogeneousAdoptionDiD(design='mass_point').fit(..., survey=...) without an explicit vcov_type now hit the same clear NotImplementedError as the explicit vcov_type='classical' path. R5 P2 (regression coverage): added three tests covering the default-path cases that slipped through the previous guard: - test_mass_point_default_vcov_survey_rejected_static: default vcov_type=None + survey= on mass-point → NotImplementedError - test_mass_point_default_vcov_event_study_cband_rejected: default vcov_type=None + weights= + cband=True on event-study → rejects - test_mass_point_default_vcov_robust_true_survey_allowed: complement confirms robust=True (default mapping → hc1) survey= is allowed Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment ✅ Looks good No unmitigated P0/P1 findings. One P2 performance item and one P3 test-gap note remain. Executive Summary
Methodology
Code Quality
Performance
Maintainability
Tech Debt
Security
Documentation/Tests
|
R6 P2 (event-study opt-out overhead): cband=False on the weighted
event-study path no longer allocates the stacked (G, H) IF matrix or
forces per-horizon IF return on the weights= shortcut. Split two
flags internally:
needs_per_horizon_if = survey= path OR (weights= AND cband=True)
needs_stacked_if_matrix = weights= AND cband=True (alias for
weighted_es AND cband)
- Psi allocation gated on needs_stacked_if_matrix.
- _fit_continuous force_return_influence gated on
(needs_stacked_if_matrix AND resolved_survey_unit_full is None) —
under survey= path, _fit_continuous returns the IF anyway via its
resolved_survey_unit gate, so no extra cost.
- _fit_mass_point_2sls return_influence gated on
needs_per_horizon_if — survey= path needs the per-horizon IF for
the Binder-TSL override regardless of cband.
Net effect: cband=False + weights= shortcut + weighted_es skips the
O(GH) Psi allocation and the per-horizon IF work entirely. cband=True
paths and survey= paths unchanged.
R6 P3 (event-study survey= integration coverage): added two end-to-end
integration tests for the previously-unguarded positive-path
estimator-level survey= + aggregate='event_study' dispatch:
- test_survey_event_study_continuous_end_to_end: continuous_at_zero +
SurveyDesign(strata='stratum') — asserts variance_formula=
'survey_binder_tsl', survey_metadata.df_survey=G-n_strata, cband_*
populated, PSU dispatch through _aggregate_unit_resolved_survey.
- test_survey_event_study_mass_point_end_to_end: mass_point +
SurveyDesign(strata=...) — asserts variance_formula=
'survey_binder_tsl_2sls' and that the 2SLS IF flows through
per-horizon Binder-TSL + sup-t bootstrap.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment ⛔ Blocker One unmitigated P0 remains: the new mass-point weighted event-study shortcut miscalibrates the sup-t band on Executive Summary
Methodology Affected method:
Code Quality No findings. Performance No findings. The prior Maintainability No findings. Tech Debt No separate findings. The new follow-ups in Security No findings. Documentation/Tests
Path to Approval
|
R7 P0 (sup-t under-scaling on weights= shortcut): the per-unit IF returned by _fit_continuous / _fit_mass_point_2sls is HC1-scaled per the PR #359 convention — compute_survey_if_variance(psi, trivial_resolved) ≈ V_HC1. Routing the weights= shortcut through the unit-level ``resolved_survey=None`` branch of _sup_t_multiplier_bootstrap normalized against raw sum(psi²) = ((n-1)/n) · V_HC1, producing silently too-narrow simultaneous bands. Fix: when the weighted event-study + cband=True path runs, always route the sup-t bootstrap through a ResolvedSurveyDesign. On the weights= shortcut (no user-supplied survey), construct a synthetic trivial resolved (pweight, no strata/psu/fpc, lonely_psu='remove') so the centered + sqrt(n/(n-1))-corrected survey-aware branch fires. The no-strata/no-PSU path inside that branch falls through to the "single implicit stratum — demean across all PSUs, scale by sqrt(n_psu/(n_psu-1))" block, which gives Var_xi(xi @ Psi_psu) ≈ V_HC1 as the IF scale convention requires. Net effect: weights= shortcut and survey=SurveyDesign(weights=...) now target the SAME variance family in the bootstrap (~atol=0.05 between the two quantiles at matching seeds, bounded by the bc_fit.se_robust vs Binder-TSL per-horizon SE convergence tolerance from PR #359). Previously the shortcut was under-scaled by sqrt(n/(n-1)) relative to the analytical HC1 target. Regression tests (+2): - test_weights_shortcut_mass_point_h1_cband_matches_normal: helper- level H=1 lock with mass-point HC1-scaled IF + synthetic trivial resolved. q → Phi^-1(0.975) ≈ 1.96 at atol=0.15 (MC noise at B=5000). The pre-fix under-scaling would have produced q ≈ 1.94 (systematic drift outside MC noise). - test_weights_shortcut_cband_matches_trivial_survey: weights= shortcut and survey=SurveyDesign(weights='w') event-study cband quantiles agree within atol=0.05 on the same DGP / seed. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owner
Author
|
/ai-review |
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment ✅ Looks good Executive Summary
Methodology No unmitigated P0/P1 findings. The changed estimator logic now matches the documented HC1/Binder targets materially better than in the previous review state. Code Quality No findings. Performance No findings. Maintainability No findings. Tech Debt No findings. The follow-up work on analytical cross-horizon covariance and unweighted sup-t is properly tracked in Security No findings. Documentation/Tests
|
R8 P3 (doc-consistency drift): prose in _fit_event_study docstring, REGISTRY sup-t bullet, and CHANGELOG entry still described the pre-R7-fix weighted-path behavior (Binder-TSL for all weighted fits; raw unit-level shortcut bootstrap). Updated all three to match the actual implementation: - _fit_event_study docstring: distinguishes survey= path (Binder-TSL) from weights= shortcut (analytical CCT-2014 / 2SLS pweight sandwich), documents trivial-survey bootstrap routing for cband. - REGISTRY.md: new "weights= shortcut ↔ bootstrap routing" bullet under the sup-t section explaining the synthetic trivial ResolvedSurveyDesign construction and why (centered + sqrt(n/(n-1))-corrected branch targets V_HC1, raw unit-level would give ((n-1)/n) · V_HC1 under-scaling). - CHANGELOG.md: per-horizon variance on weights= shortcut clarified as analytical (not Binder-TSL); sup-t routing through trivial resolved documented. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced Apr 25, 2026
igerber
added a commit
that referenced
this pull request
Apr 25, 2026
Closes the Phase 4.5 C0 promise (PR #367 commit 29f8b12). Linearity- family pretests now accept survey=/weights= keyword-only kwargs: - stute_test, yatchew_hr_test, stute_joint_pretest, joint_pretrends_test, joint_homogeneity_test, did_had_pretest_workflow. Stute family: PSU-level Mammen multiplier bootstrap via generate_survey_multiplier_weights_batch. Each replicate draws (B, n_psu) Mammen multipliers, broadcast to per-obs perturbation eta_obs[g] = eta_psu[psu(g)], weighted OLS refit, weighted CvM via new _cvm_statistic_weighted helper. Joint Stute SHARES the multiplier matrix across horizons within each replicate, preserving both vector-valued empirical-process unit-level dependence (Delgado 1993; Escanciano 2006) AND PSU clustering (Krieger-Pfeffermann 1997). NOT Rao-Wu rescaling -- multiplier bootstrap is a different mechanism. Yatchew: closed-form weighted OLS + pweight-sandwich variance components (no bootstrap): sigma2_lin = sum(w * eps^2) / sum(w) sigma2_diff = sum(w_avg * diff^2) / (2 * sum(w)) [Reviewer CRITICAL #2] sigma4_W = sum(w_avg * eps_g^2 * eps_{g-1}^2) / sum(w_avg) T_hr = sqrt(sum(w)) * (sigma2_lin - sigma2_diff) / sigma2_W where w_avg_g = (w_g + w_{g-1}) / 2 (Krieger-Pfeffermann 1997 Section 3). All three components reduce bit-exactly to existing unweighted formulas at w=ones(G); locked at atol=1e-14 by direct helper test. Workflow under survey/weights: skips the QUG step with UserWarning (per C0 deferral), sets qug=None on the report, dispatches the linearity family with survey-aware mechanism, appends "linearity-conditional verdict; QUG-under-survey deferred per Phase 4.5 C0" suffix to the verdict. all_pass drops the QUG-conclusiveness gate (one less precondition). HADPretestReport.qug retyped from QUGTestResults to Optional[QUGTestResults]; summary/to_dict/to_dataframe updated to None-tolerant rendering. Pweight shortcut routing: weights= passes through a synthetic trivial ResolvedSurveyDesign (new survey._make_trivial_resolved helper) so the same kernel handles both entry paths -- mirrors PR #363's R7 fix pattern on HAD sup-t. Replicate-weight survey designs (BRR/Fay/JK1/JKn/SDR) raise NotImplementedError at every entry point (defense in depth, reciprocal- guard discipline). The per-replicate weight-ratio rescaling for the OLS-on-residuals refit step is not covered by the multiplier-bootstrap composition; deferred to a parallel follow-up. Per-row weights= / survey=col aggregated to per-unit via existing HAD helpers (_aggregate_unit_weights, _aggregate_unit_resolved_survey; constant-within-unit invariant enforced) through new _resolve_pretest_unit_weights helper. Strictly-positive weights required on Yatchew (the adjacent-difference variance is undefined under contiguous-zero blocks). Stability invariants preserved: - Unweighted code paths bit-exact pre-PR (the new survey/weights branch is a separate if arm; existing 138 pretest tests pass unchanged). - Yatchew weighted variance components reduce to unweighted at w=1 at atol=1e-14 (locked by TestYatchewHRTestSurvey). - HADPretestReport schema bit-exact on the unweighted path; qug=None triggers the new None-tolerant rendering only on the survey path. 20 new tests across TestHADPretestWorkflowSurveyGuards (revised from C0 rejection-only to C functional + 2 mutex/replicate-weight retained), TestStuteTestSurvey (7), TestYatchewHRTestSurvey (7), TestJointStuteSurvey (5). Full pretest suite: 158 tests pass. Patch-level addition (additive on stable surfaces). See docs/methodology/REGISTRY.md "QUG Null Test" -- Note (Phase 4.5 C) for the full methodology. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Closes the two Phase 4.5 A
NotImplementedErrorgates:design=\"mass_point\" + weights/surveyatdiff_diff/had.py:2807-2817aggregate=\"event_study\" + weights/surveyatdiff_diff/had.py:2376-2382Ships weighted 2SLS mass-point path, event-study survey composition for all three designs, and a multiplier-bootstrap sup-t simultaneous confidence band on the weighted event-study path.
_fit_mass_point_2slsfollows Wooldridge 2010 Ch. 12 pweight convention (w²in HC1 meat,w·uin CR1 cluster score, weighted breadZ'WX). HC1 and CR1 bit-exact withestimatr::iv_robust(..., weights=, clusters=)atatol=1e-10on 5 DGPs (new cross-language golden).compute_survey_if_variance(psi, trivial_resolved) ≈ V_HC1[1,1]atatol=1e-10(PR HAD Phase 4.5: survey support on continuous-dose paths #359 convention, applied uniformly across continuous + mass-point).weights_unit_full+resolved_survey_unit_full, composes Binder-TSL variance per horizon viacompute_survey_if_variance, populatessurvey_metadata+variance_formula+effective_dose_mean(previously hardcodedNone).generate_survey_multiplier_weights_batch/generate_bootstrap_weights_batchfromdiff_diff.bootstrap_utils(stratum centering, FPC, lonely-PSU). Perturbationsxi @ Psiwith NO(1/n)prefactor (matchesstaggered_bootstrap.py:373). AtH=1reduces toΦ⁻¹(1-α/2) ≈ 1.96(locked byTestSupTReducesToNormalAtH1).Public API
HeterogeneousAdoptionDiD.__init__gainsn_bootstrap: int = 999,seed: Optional[int] = None.HeterogeneousAdoptionDiD.fit()gainscband: bool = True(weighted event-study only)._fit_mass_point_2slssignature: newweights=+return_influence=kwargs; always returns 3-tuple(beta, se, psi).HeterogeneousAdoptionDiDEventStudyResultsgainsvariance_formula,effective_dose_mean,cband_low,cband_high,cband_crit_value,cband_method,cband_n_bootstrap(allNoneon unweighted fits).to_dict,to_dataframe,summary,__repr__.Stability invariants
TestHADSurvey) unchanged on static path.aweight,fweight, replicate) rejected withNotImplementedErroron both new paths (reciprocal-guard discipline).filter_infoidentical across unweighted / uniform-weighted / informatively-weighted fits (identification-theory, not sampling-domain).qug_test,stute_test,yatchew_hr_test, joint variants,did_had_pretest_workflow) still don't accept weights/survey — deferred to Phase 4.5 C / C0.Test plan
pytest tests/test_had.py tests/test_bias_corrected_lprobust.py tests/test_nprobust_port.py tests/test_estimatr_iv_robust_parity.py -v— all 380 pass, 2 skippedRscript benchmarks/R/generate_estimatr_iv_robust_golden.R— regenerates 5-DGP goldenTestHADSurveyunchanged)TestSupTReducesToNormalAtH1confirmsq ≈ 1.96atH=1,G=500,B=5000TestEstimatrIVRobustHC1Parity/TestEstimatrIVRobustCR1Parityconfirm cross-language bit-parity atatol=1e-10TestIFScaleInvariantconfirms PR HAD Phase 4.5: survey support on continuous-dose paths #359 convention holds uniformly for 2SLS IF🤖 Generated with Claude Code