feat(v4): semantic rename wave - post/control_group/vcov-only/level shims + missed-rename amendments (2(c)-ii PR-B, 15 rows)

[igerber]

Summary

• Second and final 2(c)-ii rename-sweep PR (v4-design §9 item 4, the semantic wave): flips 12 ledger rows planned → shimmed — [M-030]/[M-031] (time=→post= on the 2x2/2x2x2 post dummy), [M-043]/[M-095] (clean_control→control_group, estimator + results field), [M-044]/[M-086]/[M-087] (the Wooldridge level/event_study/summary triple), [M-045]..[M-047]/[M-115] (the robust drop, all four sites), [M-084] (ContinuousDiD covariates moves to fit()) — and adds three missed-rename amendment rows [M-136]..[M-138] (LPDiD level="event" value; the two post-dummy diagnostics params), born shimmed with ledger + v4-design §3.4/§8/§9 edited in the same diff (EXPECTED_INITIAL_IDS/ROW_COUNT_FLOOR → 117).
• All shims ride the PR-A _deprecation.py machinery: single FutureWarning with the uniform migration message, both-spellings ValueError, positional binding preserved, deprecated paths route to bit-identical numbers. M-031's warning notes time= becomes calendar-only at 4.0 ([M-085]).
• BaseEstimator interplay per the approved plan: robust keeps the RESOLVED legacy bool on the public attr while get_params() returns the raw sentinel-era arg (silent clone round-trips; configured instances re-warn on probe re-init, documented); StackedDiD's lossless shim carries _DERIVED_CONFIG_ATTRS + a both-directions _normalize_set_params hook + a deprecated estimator-attr property (guard-allowlisted).
• Consumer migrations (rule 11): _reporting_helpers/business_report read the renamed field with dual serialized keys through 3.9; power.py's survey gate is now estimator-scoped so a default StackedDiD stays accepted (bit-exact with the pre-rename gate); the four newly-shimmed diagnostics forwards migrated (the except Exception wrapper paths pinned zero-warning + non-error-dict); DiD/DDD power builders migrated while TWFE builders stay on time= ([M-082] is phase 3).
• The run_placebo_test/run_all_placebo_tests wrappers keep their genuinely OVERLOADED time (calendar for timing/group tests, post dummy for the renamed callees) — a rename cannot express dual semantics; the signature redesign is a TODO.md row with honest guard-allowlist reasons.
• WooldridgeDiDResults.summary adopts the library-wide §5 contract: alpha (keyword-only during 3.9) RAISES on a value different from the stored one — no label-only relabeling of stored intervals.

Methodology references (required if estimator / math changes)

• Method name(s): N/A - API renames only; every shim resolves into the pre-existing body-local variable before estimation, and per-surface routing-parity tests pin old-path == new-path numbers. No estimand, weighting, comparison-group, variance, or inference change.
• Paper / source link(s): naming rules per docs/v4-design.md §8 (rule 1 post vs calendar time, rule 6 control_group, §7's robust retirement, rule 10 public-function completeness, the missed-rename clause for [M-136]..[M-138]).
• Any intentional deviations from the source (and why): None. Calendar time= surfaces (TWFE, CS, MPD, dCDH, SDID's own fit, CiC/QDiD) are deliberately untouched; CS-family cohort vocabulary unchanged.

Validation

• Tests added/updated: new tests/test_v4_rename_semantic.py (47 pins incl. warning contracts, routing parity, field trio + pickle + dual keys, probe re-warn counts, wrapper zero-warning assertions, the power survey-gate regression, and the summary-alpha reject contract); ~600 test call sites migrated to canonical names (scope-aware AST rewriter + hand triage; deliberate alias-semantics tests keep exercising deprecated spellings under pytest.warns); tests/test_v4_matrix.py + tests/test_naming_guard.py green with all 15 rows and the amendment bookkeeping.
• Backtest / simulation / notebook evidence (if applicable): the ten touched tutorials (01, 03, 04, 08, 09, 13, 16-survey, 16-wooldridge, 17, 27) execute end-to-end via nbmake; 16_wooldridge_etwfe's stored outputs re-executed in place after its display calls changed. Numbers unchanged (rename-only routing).

Security / privacy

• Confirm no secrets/PII in this PR: Yes

[github-actions]

Overall assessment

✅ Looks good — no unmitigated P0 or P1 findings.

Executive summary

• Estimator math, weighting, identification, and variance formulas remain unchanged.
• Rename shims preserve positional calls, defaults, serialization, cloning, and old/new routing parity.
• The overloaded placebo-wrapper parameter is explicitly tracked in TODO.md.
• One nonblocking P2 documentation issue remains: several docstrings still present deprecated spellings as current.
• Tests could not be executed because the review environment lacks pytest and numerical dependencies.

Methodology

No defects found.

Cross-checking against docs/methodology/REGISTRY.md confirmed preservation of:

• 2×2 DiD and TripleDifference estimands and variance behavior.
• Wing–Freedman–Hollingsworth StackedDiD clean-control definitions.
• ContinuousDiD conditional-parallel-trends covariate path, including bootstrap propagation.
• Wooldridge ETWFE cell/cohort-share weighting and inference.
• The documented randomization-inference exception for permutation_test’s inline t-statistic.

Code Quality

No findings. The shared rename resolver and safe_inference() are used consistently; invalid dual spellings fail explicitly.

Performance

No findings. Changes add only constant-time argument resolution and compatibility bookkeeping.

Maintainability

No findings. Raw versus resolved constructor state and bidirectional set_params() handling are explicit and tested.

Tech Debt

• Severity: P3 — informational, tracked
• Impact: run_placebo_test and run_all_placebo_tests retain an overloaded time argument whose meaning varies by delegated diagnostic.
• Concrete fix: Implement the signature split already recorded at TODO.md:L24. Per the deferred-work policy, this is not a blocker.

Security

No findings. No secrets, credentials, unsafe evaluation, or new external-input execution paths were identified.

Documentation/Tests

• Severity: P2
• Impact: Public docstrings still describe deprecated parameters and values as current, encouraging calls that now emit FutureWarning. Examples include ContinuousDiD(covariates=) at diff_diff/continuous_did.py:L91-L97, robust= at diff_diff/estimators.py:L67-L71, diff_diff/triple_diff.py:L410-L414, diff_diff/had.py:L2725-L2731, diff_diff/linalg.py:L4303-L4308, and diff_diff/twfe.py:L43-L44. The Wooldridge weighting documentation also still says type="event" at diff_diff/wooldridge_results.py:L264-L272.
• Concrete fix: Mark robust= as deprecated with the vcov_type= replacement, describe ContinuousDiD covariates under fit(), and replace canonical Wooldridge examples with event_study.

The new compatibility tests cover warnings, dual-name rejection, positional stability, routing parity, pickle migration, and parameter round-trips by inspection.

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 447e57e189a5f8344822331e407368b07d2b31f9

---

Overall assessment

✅ Looks good — no unmitigated P0 or P1 findings.

Executive summary

• Estimands, weighting, control groups, variance, and inference formulas are unchanged.
• Rename shims preserve legacy routing, positional calls, cloning, and serialization.
• The previous P2 docstring finding is resolved.
• The overloaded placebo-wrapper parameter is tracked in TODO.md.
• Static syntax checks passed; pytest is unavailable in the review environment.

Methodology

No findings.

Affected methods were checked against their Registry entries: DiD, TripleDifference, ContinuousDiD, StackedDiD, Wooldridge ETWFE, LPDiD result formatting, and placebo diagnostics (docs/methodology/REGISTRY.md:L47, L1199, L1946, L2091, L2857, L4442). The changes do not alter their mathematical or inferential contracts.

The inline permutation t-stat remains the documented randomization-inference exception at docs/methodology/REGISTRY.md:L4453-L4459.

Code Quality

No findings. Deprecated aliases consistently reject dual spellings and resolve to canonical body-local values.

Performance

No findings. The shims add only constant-time argument and compatibility handling.

Maintainability

No findings.

Keeping triple_difference(time=) unchanged is intentional: the wrapper is removed under M-075 (docs/v4-deprecations.yaml:L856-L867), and removal-bound wrappers are explicitly exempt from rename propagation (docs/v4-design.md:L573-L581).

Tech Debt

• Severity: P3 — informational, tracked
• Impact: run_placebo_test and run_all_placebo_tests retain an overloaded time parameter.
• Concrete fix: Implement the signature split recorded at TODO.md:L24. No action is required for this PR.

Security

No findings. No secrets, unsafe evaluation, or new external-input execution paths were identified.

Documentation/Tests

No findings.

The previous documentation issue is resolved: deprecated robust=, constructor-level ContinuousDiD covariates=, and Wooldridge "event" now identify their canonical replacements.

AST syntax validation passed for all 60 changed Python/test files inspected. Runtime tests could not be executed because pytest is not installed.

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: e4f0164f325ca6ed2d35bf9cd69209d79c83c129

---

Overall assessment

✅ Looks good — no unmitigated P0 or P1 findings.

Executive summary

• Rename shims do not alter estimands, weighting, controls, variance, or inference.
• The latest commit correctly migrates StackedDiD test stubs to control_group.
• The previous documentation finding remains resolved.
• The overloaded placebo-wrapper parameter is tracked in TODO.md.
• Static parsing passed; runtime tests were unavailable because NumPy and pytest are absent.

Methodology

No actionable findings.

Affected methods were cross-checked against the Registry: DifferenceInDifferences, ContinuousDiD, StackedDiD, Wooldridge ETWFE, LPDiD, TripleDifference, and placebo diagnostics (docs/methodology/REGISTRY.md:L47, L1199, L1946, L2091, L2277, L2857, L4429).

• Severity: P3 — informational
• Impact: The inline permutation t-stat at diff_diff/diagnostics.py:L716 is an intentional randomization-inference exception, documented at docs/methodology/REGISTRY.md:L4453-L4470.
• Concrete fix: None required.

Code Quality

No findings. Canonical StackedDiD reporting now consistently reads control_group at diff_diff/_reporting_helpers.py:L199-L215.

Performance

No findings. Compatibility shims add only constant-time argument handling.

Maintainability

No findings. The re-review delta consistently replaces obsolete test-stub fields across all control-group branches.

Tech Debt

• Severity: P3 — informational, tracked
• Impact: Placebo wrappers retain an overloaded time parameter.
• Concrete fix: Implement the signature split recorded at TODO.md:L24; no action is required for this PR.

Security

No findings. No new unsafe execution paths, secrets, or sensitive data were identified.

Documentation/Tests

No findings.

The follow-up commit correctly exercises control_group at tests/test_target_parameter.py:L94-L100 and all three supported branches at tests/test_target_parameter.py:L382-L395. All Python files under diff_diff/ and tests/ passed AST parsing.