v3.0.0: Deprecation removals, version bump, docs refresh

[igerber]

Summary

• Remove 3 deprecated parameters (breaking changes justifying major version):
  • CallawaySantAnna(bootstrap_weight_type=) → use bootstrap_weights
  • TROP(method="twostep") → use method="local"
  • TROP(method="joint") → use method="global"
• Rename internal attribute bootstrap_weight_type → bootstrap_weights in bootstrap mixin and StaggeredTripleDifference
• Add TROP set_params() validation — invalid method values now raise ValueError
• Schedule SyntheticDiD deprecations — lambda_reg/zeta formally scheduled for removal in v3.1
• Bump version 2.9.1 → 3.0.0 across all 5 locations
• ROADMAP refresh — Phase 10 items marked shipped, survey_aggregate() future item added
• Documentation consistency — remove all stale WooldridgeDiD survey gap notes, update survey-roadmap.md, REGISTRY.md, troubleshooting.rst, llms.txt, llms-full.txt, TROP tutorial

Methodology references (required if estimator / math changes)

• Method name(s): N/A — no methodology changes, only parameter/alias removal and documentation
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: test_trop.py (replaced 4 deprecated alias tests with 1 rejection test), test_staggered.py (updated param names), test_methodology_callaway.py (removed deprecated param tests), test_staggered_triple_diff.py (updated assertion)
• Backtest / simulation / notebook evidence: N/A — no methodology changes

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good. No unmitigated P0/P1 findings. I did not find a methodology deviation, incorrect variance/SE change, or silent correctness regression in the changed estimator paths. Review limitation: I could not execute pytest in this environment because it is not installed.

Executive Summary

• The estimator-facing code changes are API cleanups, not methodology changes: TROP still exposes the same local and global methods documented in the registry, and CallawaySantAnna still uses the same multiplier bootstrap weight families under the renamed bootstrap_weights attribute (docs/methodology/REGISTRY.md:L1754-L1864, diff_diff/trop.py:L64-L78, diff_diff/trop.py:L138-L142, diff_diff/staggered.py:L288-L349, diff_diff/staggered_bootstrap.py:L329-L350).
• The bootstrap_weight_type → bootstrap_weights rename is propagated consistently through the Callaway bootstrap mixin and the staggered triple-diff wrapper (diff_diff/staggered_bootstrap.py:L329-L350, diff_diff/staggered_triple_diff.py:L161-L188).
• TROP now rejects the removed aliases in both constructor-time validation and set_params(), which matches the v3.0 breaking-change intent (diff_diff/trop.py:L138-L142, diff_diff/trop.py:L892-L903).
• The refreshed survey-support claims are consistent with the current WooldridgeDiD code path, which does accept survey_design and documents its current constraints (diff_diff/wooldridge.py:L367-L425).
• Minor gap only: the breaking removals are not directly regression-tested at constructor level for all removed APIs; that is P3, not approval-blocking (tests/test_trop.py:L2898-L2924, tests/test_staggered.py:L1587-L1600).

Methodology

• No findings. The diff does not change estimator math, weighting formulas, variance/SE computation, or identification assumptions. TROP remains documented/implemented as local vs global, and the Callaway bootstrap still uses the same valid weight choices; only deprecated names were removed or renamed (docs/methodology/REGISTRY.md:L1754-L1864, diff_diff/trop.py:L64-L78, diff_diff/trop.py:L138-L142, diff_diff/staggered.py:L288-L349, diff_diff/staggered_bootstrap.py:L329-L350).

Code Quality

• No findings. The internal rename is consistent across estimator state and bootstrap consumption, and the touched parameter surfaces remain coherent after the cleanup (diff_diff/staggered.py:L343-L350, diff_diff/staggered_bootstrap.py:L329-L350, diff_diff/staggered_triple_diff.py:L161-L188, diff_diff/trop.py:L892-L903).

Performance

• No findings. The reviewed changes are parameter validation, attribute renaming, and docs/test updates; they do not alter hot numerical loops, bootstrap algorithms, or estimator complexity.

Maintainability

• No findings. Removing the deprecated aliases simplifies the public surface and eliminates dual-name branches in the touched paths without introducing an undocumented new behavior (diff_diff/trop.py:L138-L142, diff_diff/staggered.py:L288-L349).

Tech Debt

• No findings. The remaining SyntheticDiD deprecated names are explicitly tracked for v3.1, so the carried debt is documented rather than silently extended (TODO.md:L115-L121, diff_diff/synthetic_did.py:L143-L157, diff_diff/synthetic_did.py:L1122-L1135).

Security

• No findings. I did not see secrets, new user-controlled execution paths, unsafe subprocess/network behavior, or data-handling changes in the diff.

Documentation/Tests

• Severity: P3. Impact: The breaking API removals are only partially regression-tested. tests/test_trop.py checks generic invalid constructor input and set_params() rejection, but not explicit constructor rejection of the removed twostep/joint aliases; tests/test_staggered.py removes the old bootstrap_weight_type deprecation tests without adding a direct assertion that the removed constructor parameter now fails. That leaves a minor gap where accidental alias reintroduction or a different failure mode could slip through unnoticed (diff_diff/trop.py:L138-L142, diff_diff/staggered.py:L288-L349, tests/test_trop.py:L2898-L2924, tests/test_staggered.py:L1587-L1600).
• Concrete fix: Add explicit negative tests for TROP(method="twostep"), TROP(method="joint"), and CallawaySantAnna(bootstrap_weight_type="mammen"), asserting the intended exception type/message for the v3.0 removal.
• Residual review gap: I could not run the targeted test slice because pytest is unavailable in this environment.