Refresh ROADMAP to drop phase numbering and reflect shipped state by igerber · Pull Request #313 · igerber/diff-diff

igerber · 2026-04-18T13:06:02Z

Summary

Rewrite ROADMAP.md with new structure: Current State / Recently Shipped / Shipping Next / Under Consideration / AI-Agent Track / Long-term. No top-level phase numbering, no version stamps in headings, no third-party product names.
Absorb dCDH (shipped end-to-end across PRs Add ChaisemartinDHaultfoeuille (dCDH) DID_M estimator (Phase 1) #290, Add Phase 2 multi-horizon event study for dCDH estimator #294, dCDH Phase 3a: placebo SE, non-binary treatment, parity SE assertions #300, Add Phase 3 PR B: covariates, trends, and extensions for dCDH #302, Add HonestDiD integration for dCDH, summary() Phase 3 blocks #303, Add survey support to dCDH estimator #307) into the Current State estimator list and the Recently Shipped summary. Remove the separate "Future Estimators / Phase 1-3" dCDH roadmap block.
Add research-informed candidates to Under Consideration: DiD with no untreated group (de Chaisemartin et al. 2024, inverse-SDiD), efficient staggered DiD (Chen-Sant'Anna-Xie 2025), distributional DiD for staggered (Ciaccio 2024), LP-DiD (Dube et al., JAE 2025 published), few-treated-units inference (Alvarez-Ferman-Wuthrich 2025), Riesz-representation sensitivity (Bach et al. 2025), compositional-change inference (Sant'Anna-Xu 2025), triple-difference identification audit (Ortiz-Villavicencio-Sant'Anna 2025).
Promote the AI-agent track to a named long-term arc with a stated vision and named building blocks (BusinessReport, DiagnosticReport, context-aware practitioner_next_steps, AI-legible diagnostic surfaces, scenario-to-estimator guidance).
Update companion docs (docs/business-strategy.md Section 8, docs/survey-roadmap.md, docs/practitioner_decision_tree.rst, docs/choosing_estimator.rst, docs/api/chaisemartin_dhaultfoeuille.rst, docs/api/efficient_did.rst, README.md, diff_diff/guides/llms-full.txt) to remove stale "deferred to Phase N" language now that the deferred items have shipped.
Correct EfficientDiD covariate-path documentation: the doubly-robust sieve-based covariate path is shipped; several docs still said "not yet supported."
Fix broken anchor docs/survey-roadmap.md#deferred-work-consolidated in TODO.md to point at the existing #current-limitations section.

Methodology references (required if estimator / math changes)

Method name(s): N/A - no methodology changes. Content-only docs refresh.
Paper / source link(s): N/A
Any intentional deviations from the source (and why): None

Validation

Tests added/updated: No test changes (content-only).
Backtest / simulation / notebook evidence (if applicable): N/A
Verified ROADMAP.md has no em-dashes, no named third-party products, no version numbers in headings, and no stale top-level phase numbering.
Confirmed remaining Phase N strings across the repo are all in whitelisted build-milestone history docs (docs/methodology/REGISTRY.md, docs/methodology/continuous-did.md, docs/performance-plan.md, docs/survey-roadmap.md historical sections) or internal TODO.md tech-debt entries that are out of scope.
EfficientDiD covariate-path correction verified against diff_diff/efficient_did.py (covariates parameter on fit() is shipped with doubly-robust sieve-based path).

Security / privacy

Confirm no secrets/PII in this PR: Yes

Generated with Claude Code

Rewrites ROADMAP.md to organize work as Current State / Recently Shipped / Shipping Next / Under Consideration / AI-Agent Track / Long-term. Drops top-level phase numbering. Absorbs dCDH into the shipped estimator list. Promotes future-estimator candidates from the 2025-26 methodology research (including the inverse-SDiD "no untreated group" estimator, efficient staggered DiD, distributional DiD for staggered, LP-DiD, few-treated-units inference, Riesz-representation sensitivity, compositional-change inference, and the triple-difference covariate audit). Promotes the AI-agent track to a named long-term arc. Updates companion docs (docs/business-strategy.md Section 8, README.md dCDH section, docs/practitioner_decision_tree.rst, docs/choosing_estimator.rst, docs/api/chaisemartin_dhaultfoeuille.rst, docs/api/efficient_did.rst, docs/survey-roadmap.md, diff_diff/guides/llms-full.txt) to remove stale phase-deferral language now that the deferred items have shipped. Corrects EfficientDiD covariate-path documentation (the doubly-robust path is shipped; several docs still said "not yet supported"). Fixes the broken TODO.md anchor link into survey-roadmap.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

github-actions · 2026-04-18T13:16:47Z

Overall Assessment

⚠️ Needs changes

Executive Summary

EfficientDiD docs now advertise shipped covariate support, but the refreshed text still presents the estimator as unconditionally efficiency-bound-attaining. That conflicts with the registry and class docstring for the shipped covariate path.
dCDH placebo inference is now described inconsistently across touched surfaces. Some docs still say all placebo SEs are NaN, while the code and registry say only single-period DID_M^pl is NaN; multi-horizon DID^{pl}_l has inference.
ROADMAP.md now lists “Efficient staggered DiD” as future work even though shipped EfficientDiD is already the staggered-adoption estimator.
I did not find code, performance, or security regressions in the touched diff; the issues are methodology/documentation accuracy.

Methodology

Severity: P1. Impact: After this PR, docs/api/efficient_did.rst:L7-L21, README.md:L1116-L1163, and diff_diff/guides/llms-full.txt:L640-L665 pair “covariates supported” with unqualified “achieves the semiparametric efficiency bound / maximum efficiency” language. That conflicts with the shipped contract in diff_diff/efficient_did.py:L139-L146 and docs/methodology/REGISTRY.md:L899-L903, which explicitly say the current covariate path is doubly robust but the linear OLS outcome-regression implementation does not generically attain the semiparametric efficiency bound unless the conditional mean is linear in X. Concrete fix: qualify the efficiency claim everywhere this PR flips docs from “no covariates” to “covariates supported”; scope the unqualified bound claim to the no-covariate path, and describe the shipped covariate path with the registry’s DR caveat.
Severity: P1. Impact: The dCDH placebo-inference contract is contradictory across changed surfaces. README.md:L1264-L1270 and docs/choosing_estimator.rst:L290-L301 still say both single-lag and dynamic placebo SEs are NaN; diff_diff/guides/llms-full.txt:L321-L325 says dynamic placebo SE exists only via bootstrap; but docs/practitioner_decision_tree.rst:L198-L206, docs/methodology/REGISTRY.md:L546-L588, and the implementation diff_diff/chaisemartin_dhaultfoeuille.py:L1668-L1731 and diff_diff/chaisemartin_dhaultfoeuille.py:L2249-L2305 show the actual contract: DID_M^pl stays NaN, while DID^{pl}_l has analytical SE and gets bootstrap overrides when enabled. Concrete fix: harmonize all touched docs/guides to that exact split.
Severity: P2. Impact: ROADMAP.md:L101-L105 describes “Efficient staggered DiD” as a future extension that would move shipped EfficientDiD from 2x2 to staggered timing, but shipped EfficientDiD is already the staggered-adoption estimator in docs/api/efficient_did.rst:L1-L15 and docs/methodology/REGISTRY.md:L767-L783. Concrete fix: remove that item or rename it to a genuinely distinct follow-on method, with the added scope stated explicitly.

Code Quality
No findings.

Performance
No findings.

Maintainability

Severity: P3. Impact: The same estimator contract is duplicated across README, API docs, estimator-choice docs, roadmap text, and runtime AI guides, and this PR already left those surfaces out of sync. Concrete fix: centralize the sensitive estimator snippets or add a lightweight consistency check for key claims such as the EfficientDiD covariate-efficiency caveat and the dCDH single-lag vs dynamic placebo inference split.

Tech Debt
No deferrable blocker here. The P1 items above are current methodology/SE documentation inaccuracies, not TODO-grade follow-ups.

Security
No findings.

Documentation/Tests

Severity: P3, tracked in TODO. Impact: .txt AI guides are outside current doc-smoke coverage, which helps explain the llms-full.txt drift; this is already tracked at TODO.md:L95-L95. Concrete fix: no approval blocker, but extending doc-smoke coverage to .txt guides would reduce recurrence.

Path to Approval

Update EfficientDiD docs in README.md, docs/api/efficient_did.rst, docs/choosing_estimator.rst, and diff_diff/guides/llms-full.txt so the no-covariate path is the only unqualified efficiency-bound claim, and the shipped covariate path is described with the registry’s DR / linear-OR caveat.
Update all touched dCDH docs/guides so they say: single-period DID_M^pl has NaN inference; multi-horizon DID^{pl}_l has analytical SE and bootstrap override when n_bootstrap > 0.
Rewrite or drop the ROADMAP.md “Efficient staggered DiD” candidate so it no longer describes the already-shipped staggered EfficientDiD as future work.

…acebo SE contract; rescope Under-Consideration entry EfficientDiD efficiency claim: scope the unqualified "achieves the semiparametric efficiency bound" language to the no-covariate path in README.md, docs/api/efficient_did.rst, and diff_diff/guides/llms-full.txt. The doubly-robust covariate path uses a linear OLS outcome regression; per docs/methodology/REGISTRY.md L899-L903 and the class docstring in diff_diff/efficient_did.py L139-L146, this preserves DR consistency but does not generically attain the efficiency bound unless the conditional mean is linear in the covariates. dCDH placebo SE contract: correct the inconsistent placebo-NaN statements in README.md and docs/choosing_estimator.rst; tighten diff_diff/guides/ llms-full.txt. The actual contract (per docs/methodology/REGISTRY.md L546 and diff_diff/chaisemartin_dhaultfoeuille.py): single-period DID_M^pl (L_max=None) has NaN SE because the per-period aggregation path has no influence-function derivation, with NaN-consistent inference even under n_bootstrap > 0; multi-horizon dynamic placebos DID^{pl}_l (L_max >= 1) have valid analytical SE via the placebo IF and bootstrap SE when n_bootstrap > 0. ROADMAP.md rescope: the "Efficient staggered DiD" Under-Consideration entry incorrectly framed shipped staggered EfficientDiD as future work. Rewrite as "Nonparametric / flexible outcome regression for EfficientDiD DR covariate path" - the actual gap vs the Chen-Sant'Anna-Xie 2025 Section 4 proposal, where nonparametric OR (sieve / kernel / ML) would close the efficiency gap the current linear OLS OR leaves open on the covariate path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

igerber · 2026-04-18T13:21:04Z

/ai-review

github-actions · 2026-04-18T13:25:19Z

🔁 AI review rerun (requested by @igerber)

Head SHA: dbe1472433fc8abfbaed133fd1bbaa02900271a6

Overall Assessment

⚠️ Needs changes

Executive Summary

Re-review: the prior dCDH placebo-inference mismatch appears resolved in the touched docs; the updated README/guide/chooser text now matches the registry’s split between single-period DID_M^pl (NaN SE) and multi-horizon DID^{pl}_l (analytical/bootstrap SE).
Re-review: the prior roadmap issue that treated shipped staggered EfficientDiD support as future work is resolved; the future item is now correctly framed as a nonparametric outcome-regression extension.
One P1 remains: the EfficientDiD covariate-path caveat was fixed in some touched surfaces, but not all. docs/choosing_estimator.rst:L363-L374 and ROADMAP.md:L18 still overstate the shipped covariate path as efficiency-optimal, contrary to the registry and class docstring.
[Newly identified] The refreshed dCDH API page still pulls in a stale Phase-1 class docstring through autodoc, so the built page will continue to show contradictory support status unless the in-code docstring is updated.
I did not find code, performance, or security regressions in the changed diff.

Methodology

Severity: P1. Impact: EfficientDiD’s shipped covariate contract is still not documented consistently after this refresh. docs/choosing_estimator.rst:L363-L374 still presents EfficientDiD as the generic “maximum statistical efficiency” choice while advertising covariate support, and ROADMAP.md:L18 says it is “semiparametrically efficient with doubly robust covariates.” That conflicts with the source-of-truth contract in docs/methodology/REGISTRY.md:L899-L903 and diff_diff/efficient_did.py:L135-L146, which state that the no-covariate path attains the bound, while the shipped covariate path is doubly robust but does not generically attain the semiparametric efficiency bound under the linear OLS outcome regression. The same contradiction now appears inside ROADMAP.md itself because ROADMAP.md:L104 correctly describes the covariate-path efficiency gap. Concrete fix: align both touched surfaces to the registry/docstring language by reserving the unqualified efficiency-bound claim for the no-covariate path and describing the current covariate path as DR-consistent with the linear-OR caveat.

Code Quality

No findings.

Performance

No findings.

Maintainability

No findings.

Tech Debt

No new deferrable tech-debt items. The remaining P1 is a current methodology-claim mismatch, not TODO-grade deferred work.

Security

No findings.

Documentation/Tests

Severity: P2. Impact: docs/api/chaisemartin_dhaultfoeuille.rst:L98-L116 now has a refreshed top-of-page description, but it still renders .. autoclass:: diff_diff.ChaisemartinDHaultfoeuille, and the underlying class docstring at diff_diff/chaisemartin_dhaultfoeuille.py:L294-L326 still says “Phase 1” and future-phase pointers. The built API page will therefore continue to show stale phase-deferral language despite this PR’s stated goal of removing it. Concrete fix: update the class docstring to the shipped contract, or suppress the autogenerated class prose until the docstring is aligned.

Path to Approval

Update the EfficientDiD section in docs/choosing_estimator.rst:L363-L374 so “maximum statistical efficiency” is explicitly scoped to the no-covariate path, and add the registry’s linear-OLS covariate-path caveat.
Rewrite the EfficientDiD bullet in ROADMAP.md:L18 so it matches the shipped contract and no longer contradicts ROADMAP.md:L104.

…ing surfaces; update dCDH class docstring Remaining EfficientDiD surfaces the prior pass missed: - ROADMAP.md line 18 (Current State estimator list) previously described EfficientDiD as "semiparametrically efficient with doubly robust covariates" - unqualified, and contradicted ROADMAP.md line 104 which correctly describes the covariate-path efficiency gap. Rewritten to reserve the unqualified bound claim for the no-covariate path and describe the DR covariate path with the linear-OR caveat. - docs/choosing_estimator.rst lines 363-374 previously said "maximum statistical efficiency" while advertising covariate support. Scoped the efficiency claim to the no-covariate path and expanded the covariate-path note with the DR / linear-OR caveat from REGISTRY.md L899-L903. dCDH class docstring (autodoc source): - diff_diff/chaisemartin_dhaultfoeuille.py lines 294-326 still opened with "de Chaisemartin-D'Haultfoeuille (dCDH) estimator - Phase 1." and listed "Phase 1 deliverables" / "Phase 1 contract" / "reserved for a future phase." Because docs/api/chaisemartin_dhaultfoeuille.rst renders `.. autoclass::` on this class, the built API page was still showing stale phase-deferral language. Rewrote the docstring to describe the shipped contract (headline DID_M plus multi-horizon DID_l, covariate adjustment, trends, non-binary treatment, HonestDiD integration, and survey support - only `aggregate` on fit() still raises NotImplementedError). Did not touch the other Phase-N references in the file (internal code comments and runtime-visible error messages), which are out of scope per the refresh plan. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

igerber · 2026-04-18T13:28:16Z

/ai-review

github-actions · 2026-04-18T13:32:13Z

🔁 AI review rerun (requested by @igerber)

Head SHA: 6ba9fbfff804c5fc7946fb886e25ddfe320e6d7b

Overall Assessment

✅ Looks good

Executive Summary

Re-review: the prior EfficientDiD covariate-path methodology mismatch is resolved in README.md:L1116-L1167, ROADMAP.md:L18-L19, and docs/choosing_estimator.rst:L363-L380.
Re-review: the prior dCDH API/docstring mismatch is resolved; the API page and class docstring now describe the shipped surface in docs/api/chaisemartin_dhaultfoeuille.rst:L8-L18 and diff_diff/chaisemartin_dhaultfoeuille.py:L294-L337.
I did not find executable-code, variance/SE, or security regressions in the changed diff.
One residual informational inconsistency remains on the touched EfficientDiD API page: the comparison table still leaves the efficiency claim unqualified in docs/api/efficient_did.rst:L147-L153, even though the page note, registry, and class docstring scope that claim to the no-covariate path in docs/api/efficient_did.rst:L17-L27, docs/methodology/REGISTRY.md:L903-L905, and diff_diff/efficient_did.py:L139-L146.
One additional informational docs mismatch remains in the runtime LLM guide: diff_diff/guides/llms-full.txt:L643-L647 still says EfficientDiD clustering is “Not yet implemented,” contrary to diff_diff/efficient_did.py:L156-L160 and docs/methodology/REGISTRY.md:L824-L828.

Methodology

Severity: P3. Impact: docs/api/efficient_did.rst:L147-L153 still says EfficientDiD “Achieves semiparametric bound” in the comparison table, which overreads the actual shipped contract for the covariate path. The caveat is already documented in the registry and surrounding note, so this is informational rather than blocking. Concrete fix: change the table cell to “Achieves semiparametric bound on the no-covariate path” and, if desired, mirror the DR caveat in the covariates row.

Code Quality

No findings.

Performance

No findings.

Maintainability

No findings.

Tech Debt

No findings. I did not see new deferred-work items that need TODO.md tracking.

Security

No findings.

Documentation/Tests

Severity: P3. Impact: diff_diff/guides/llms-full.txt:L643-L647 still advertises EfficientDiD clustering as unavailable, which is stale against the estimator contract in diff_diff/efficient_did.py:L156-L160. This does not affect library behavior, but it can mislead agent-facing guidance. Concrete fix: update that constructor comment to describe cluster-robust analytical SEs and cluster-level multiplier bootstrap instead of “Not yet implemented.”
No test-gap finding. This remains a docs-focused PR, and the touched Python change is docstring-only.

…orrect stale EfficientDiD cluster comment - docs/api/efficient_did.rst comparison table: the "Efficiency" row cell still read "Achieves semiparametric bound" unqualified, even though the surrounding page note, registry, and class docstring scope that claim to the no-covariate path. Updated to match those sources and mirrored the DR caveat in the Covariates row. - diff_diff/guides/llms-full.txt EfficientDiD constructor comment: the `cluster: str | None = None` line was still annotated "Not yet implemented," contrary to diff_diff/efficient_did.py L154-L160 which documents shipped cluster-robust SEs (Liang-Zeger on EIF values) and cluster-level multiplier bootstrap. Updated the comment to describe the actual shipped behavior so the runtime LLM guide no longer misleads agent-facing consumers. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

igerber · 2026-04-18T13:58:13Z

/ai-review

github-actions · 2026-04-18T14:02:32Z

🔁 AI review rerun (requested by @igerber)

Head SHA: 7cd9e08bb98ea23c0397c941bade31bb5b6812e3

Overall Assessment

✅ Looks good

Executive Summary

Re-review: the prior EfficientDiD covariate-path documentation mismatch is resolved in README.md:L1116-L1166, docs/api/efficient_did.rst:L9-L27, and docs/choosing_estimator.rst:L363-L380, consistent with diff_diff/efficient_did.py:L133-L160 and docs/methodology/REGISTRY.md:L903-L906.
Re-review: the prior llms-full.txt EfficientDiD mismatch is resolved; clustering and covariates are now documented as implemented in diff_diff/guides/llms-full.txt:L640-L666, matching diff_diff/efficient_did.py:L156-L160 and docs/methodology/REGISTRY.md:L903-L906.
The touched dCDH docs now align with the shipped contract for multi-horizon support, placebo inference, and survey/Taylor-series linearization support in README.md:L1171-L1274, docs/api/chaisemartin_dhaultfoeuille.rst:L8-L18, diff_diff/chaisemartin_dhaultfoeuille.py:L294-L337, and docs/methodology/REGISTRY.md:L520-L546.
No estimator math, weighting, variance, identification assumptions, or default behavior changed in executable code; the only Python-file change is docstring text.
I found no unmitigated P0 or P1 findings in the changed diff.

Methodology

No findings. I cross-checked the changed EfficientDiD and dCDH claims against the Methodology Registry and live class docstrings, including the EfficientDiD DR-efficiency caveat and the dCDH single-period versus dynamic placebo SE contract.

Code Quality

No findings. The touched Python change in diff_diff/chaisemartin_dhaultfoeuille.py:L294-L337 is docstring-only.

Performance

No findings. No runtime-affecting code paths changed.

Maintainability

No findings.

Tech Debt

No findings. I did not see new deferred work that needed separate tracking in TODO.md.

Security

No findings.

Documentation/Tests

No findings. For a docs refresh plus one docstring-only Python touch, the lack of test changes is appropriate, and the previously flagged EfficientDiD doc drift is now addressed.

…oe-ligature The UTF-8 roundtrip test in tests/test_guides.py used '\u2014' (em-dash) as the non-ASCII fingerprint in llms-full.txt. The roadmap-refresh pass replaced em-dashes with hyphens throughout the guide (per the project's hyphens-over-em-dashes style), which removed the fingerprint and broke the test. The guide still contains '\u0153' (oe-ligature, from the author name "D'Haultfoeuille" in the dCDH section), which is the remaining non-ASCII character and a more stable fingerprint since it's tied to a paper citation rather than stylistic punctuation. Updated the test accordingly. No behavior change; this is a test-side assertion update to match the guide's current character set. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

igerber · 2026-04-18T14:28:27Z

/ai-review

github-actions · 2026-04-18T14:33:41Z

🔁 AI review rerun (requested by @igerber)

Head SHA: 13e08debfb0ab0566c32b4c9261725baf8625a5f

Overall Assessment

✅ Looks good

Executive Summary

Re-review: the prior EfficientDiD documentation drift remains resolved. The refreshed wording in README.md:L1116-L1166, docs/api/efficient_did.rst:L7-L27, docs/choosing_estimator.rst:L363-L380, and llms-full.txt:L640-L666 matches the shipped DR/no-covariate split in efficient_did.py:L132-L186 and REGISTRY.md:L899-L906.
Re-review: the dCDH refresh is consistent with the shipped contract, especially the single-period vs multi-horizon placebo SE split and removal of stale phase-deferral language, across README.md:L1171-L1274, docs/api/chaisemartin_dhaultfoeuille.rst:L8-L18, docs/choosing_estimator.rst:L252-L304, docs/practitioner_decision_tree.rst:L190-L206, and chaisemartin_dhaultfoeuille.py:L294-L337, matching REGISTRY.md:L466-L477, REGISTRY.md:L546-L588, and REGISTRY.md:L651-L653.
No estimator math, weighting, variance, identification assumptions, or defaults changed in executable code; the only Python-source change is docstring text in chaisemartin_dhaultfoeuille.py:L294-L337.
The TODO anchor fix is correct: TODO.md:L18-L20 now points to an existing section in docs/survey-roadmap.md:L212-L233.
The touched guide test is consistent with the bundled resource: tests/test_guides.py:L45-L49 now checks for the retained œ in llms-full.txt:L233-L233. I could not execute the test here because pytest is unavailable and importing the package fails without numpy.

Methodology

Severity: none. Impact: no methodology mismatch found in the changed docs/docstring; the EfficientDiD and dCDH claims now align with the Methodology Registry and live implementation at efficient_did.py:L132-L186, chaisemartin_dhaultfoeuille.py:L294-L337, REGISTRY.md:L466-L477, REGISTRY.md:L546-L588, and REGISTRY.md:L899-L906. Concrete fix: none.

Code Quality

Severity: none. Impact: no runtime logic changed; the only Python-file diff is documentation text in chaisemartin_dhaultfoeuille.py:L294-L337. Concrete fix: none.

Performance

Severity: none. Impact: docs/test/docstring-only changes do not affect runtime paths or estimator performance. Concrete fix: none.

Maintainability

Severity: none. Impact: the roadmap/strategy/docs refresh reduces stale phase-language drift, and the broken TODO link is repaired via TODO.md:L18-L20 and docs/survey-roadmap.md:L212-L233. Concrete fix: none.

Tech Debt

Severity: none. Impact: no new untracked deferred work was introduced; remaining dCDH/survey limitations referenced by the refresh are already documented in REGISTRY.md:L651-L653 and docs/survey-roadmap.md:L212-L233. Concrete fix: none.

Security

Severity: none. Impact: docstring/docs/test-only changes add no secret exposure or executable attack surface. Concrete fix: none.

Documentation/Tests

Severity: none. Impact: the changed documentation is internally consistent with the shipped contracts, and the updated UTF-8 sentinel in tests/test_guides.py:L45-L49 matches llms-full.txt:L233-L233. I could not run the touched test in this sandbox because pytest is not installed and numpy is unavailable for import-time package initialization. Concrete fix: none.

Absorbs #312 (sdid scale fix), #313 (roadmap refresh + dCDH docstring rewrite), and #314 (within_transform convergence warnings) from main. Conflicts resolved in: - diff_diff/chaisemartin_dhaultfoeuille.py: took main's comprehensive Phase 1-3 feature list in the class docstring but merged in the PR #311 group-vs-PSU bootstrap-clustering framing and the replicate-weight survey-support line. Kept the PR #311 'user-specified cluster= not supported + automatic PSU-level under survey_design' wording for the cluster= parameter docstring (strictly more accurate than main's 'always clusters at the group level' text). - diff_diff/guides/llms-full.txt: kept main's more detailed placebo SE contract paragraph (which already distinguishes single-period NaN from multi-horizon analytical/bootstrap) and appended the sup-t / shared-weights / cross-horizon coverage details from the PR #311 update. Kept the PR #311 survey_design signature comment that mentions TSL + replicate + PSU bootstrap. Full regression across touched areas: 336 + 324 passing.

… remove deprecated SyntheticDiD params Package four merged PRs (#312 SDID catastrophic cancellation at extreme Y scale, #313 roadmap refresh, #314 FE imputation non-convergence signaling, #315 Frank-Wolfe SC weight solver non-convergence signaling) as 3.1.2. Also remove the SyntheticDiD(lambda_reg=...) and SyntheticDiD(zeta=...) kwargs, which have been deprecated with DeprecationWarning since v2.3.1 (2026-02-10) in favor of zeta_omega / zeta_lambda; their warning messages announced removal in v3.1. Passing the old kwargs now raises TypeError at __init__ and ValueError: Unknown parameter at set_params. Internal ridge-regression helpers that accept a lambda_reg parameter (compute_synthetic_weights, rank_control_units, Rust FFI bindings) are unaffected. Version strings bumped in diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, and diff_diff/guides/llms-full.txt. CHANGELOG populated with Fixed / Changed / Removed sections and comparison-link footer. TODO.md's "Deprecated Code" entry removed now that the task is done. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Per CI review feedback (#316): removing public kwargs under a patch version violates Semantic Versioning, which CHANGELOG.md explicitly claims to adhere to. Restore lambda_reg and zeta handling in SyntheticDiD.__init__ and set_params as warning-only, and bump the removal target in the DeprecationWarning text from "v3.1" to "v4.0.0". The 3.1.2 release now carries only the four fix/doc PRs (#312 SDID scale, #313 roadmap, #314 FE imputation convergence, #315 Frank-Wolfe convergence) with no breaking changes. - diff_diff/synthetic_did.py: restore deprecated kwargs + warnings (v4.0.0 text) - tests/test_methodology_sdid.py: restore TestDeprecatedParams class + set_params deprecation test - tests/test_estimators.py: restore test_deprecated_params - CHANGELOG.md: drop Removed section; add Changed entry documenting the v3.1 -> v4.0.0 bump in the removal target - TODO.md: restore Deprecated Code section with v4.0.0 removal target and SemVer rationale Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

igerber added the ready-for-ci Triggers CI test workflows label Apr 18, 2026

igerber merged commit 475f4e3 into main Apr 18, 2026
19 checks passed

igerber deleted the roadmap-review branch April 18, 2026 15:25

igerber mentioned this pull request Apr 18, 2026

Release 3.1.2: SDID scale fix, convergence warnings, roadmap refresh #316

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Refresh ROADMAP to drop phase numbering and reflect shipped state#313

Refresh ROADMAP to drop phase numbering and reflect shipped state#313
igerber merged 5 commits intomainfrom
roadmap-review

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

igerber commented Apr 18, 2026

Summary

Methodology references (required if estimator / math changes)

Validation

Security / privacy

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

igerber commented Apr 18, 2026

Uh oh!

github-actions Bot commented Apr 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant