Add inference-field aliases on staggered result classes

[igerber]

Summary

• Adds read-only @property aliases (att / se / conf_int / p_value / t_stat) on every staggered result class that previously only carried prefixed canonical fields (overall_*, overall_att_*, avg_*).
• Fixes a contract gap where getattr(res, "se", None) and getattr(res, "conf_int", None) returned None on every staggered result class — silently dropping se / conf_int from any external adapter that read them via flat getattr.
• 11 result classes touched, 59 read-only properties added, no behavior change. Backwards compatible (the overall_* / overall_att_* / avg_* fields remain canonical).

Methodology references (required if estimator / math changes)

• Method name(s): N/A — this is a result-class API extension, no methodology change. Aliases are pure read-throughs over canonical fields.
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None. The safe_inference() joint-NaN consistency contract (per CLAUDE.md "Inference computation") is inherited automatically because aliases never recompute — NaN canonical → NaN alias is regression-locked.

Validation

• Tests added/updated: tests/test_result_aliases.py (23 tests: 9 Pattern-B alias-mechanics + 9 Pattern-B NaN-propagation + 3 ContinuousDiD + 1 MultiPeriod + 1 balance-adapter regression)
• Backtest / simulation / notebook evidence (if applicable): N/A. Verified empirically against a real CallawaySantAnnaResults fit pre-PR (returned se=None / conf_int=None) and post-PR (returns populated values).
• Existing per-estimator suites (482 tests across test_chaisemartin_dhaultfoeuille.py / test_staggered.py / test_continuous_did.py) all green; full per-estimator sweep (1128 tests) green pre-rebase.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

Coverage map

Pattern	Mapping	Result classes
B	overall_* → flat (5 props each)	CallawaySantAnna, Stacked, EfficientDiD, dCDH, StaggeredTripleDiff, Wooldridge, SunAbraham, ImputationDiD, TwoStage
C	overall_att_* → flat + overall_* (9 props)	ContinuousDiD (ATT-side as headline; ACRT-side accessible unchanged via overall_acrt_*)
D	avg_* → flat (5 props)	MultiPeriodDiD

Per-result-class fields enumerated in docs/methodology/REGISTRY.md preamble.

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good — no unmitigated P0/P1 findings.

Executive Summary

• No estimator math, weighting, variance/SE, identification, or default behavior changes detected.
• Alias properties are pure read-throughs to canonical fields, so they inherit existing inference and NaN behavior.
• Sibling-surface audit found the intended scalar result classes covered.
• Pattern-wide inference grep found no new inline inference computation in the alias changes.
• Targeted tests could not be run in this review environment because pytest and numpy are unavailable.

Methodology

No P0/P1/P2 findings.

Severity: P3 informational
Impact: The PR is an API-surface extension over existing result fields. The affected result classes are the staggered/prefixed scalar result classes plus MultiPeriodDiDResults and ContinuousDiDResults; the aliases do not recompute ATT, SE, p-values, t-stats, or CIs. This matches the registry’s documented alias contract in docs/methodology/REGISTRY.md:L5.
Concrete fix: None required.

Code Quality

No findings. The aliases are simple read-only properties and do not introduce new inference logic.

Performance

No findings. Property access is constant-time and does not add material overhead.

Maintainability

No blocking findings. Sibling-surface mirror audit checked the parallel result naming surfaces: overall_*, overall_att_*, avg_*, and native flat fields.

Tech Debt

No findings. I did not see new untracked correctness debt, and no TODO mitigation is needed for this PR.

Security

No findings. Secret-pattern grep over the changed files found no obvious accidental credentials.

Documentation/Tests

Severity: P3 informational
Impact: The new registry preamble says “three native naming patterns” but enumerates four: flat att fields, overall_*, overall_att_*, and avg_*. This is documentation precision only, not a methodology or runtime defect. See docs/methodology/REGISTRY.md:L5.
Concrete fix: Change “three” to “four” and consider scoping “Every result class” to “Every scalar treatment-effect result class covered by this naming contract.”

Validation note: attempted pytest -q tests/test_result_aliases.py, but this environment lacks pytest; an import probe also showed numpy unavailable.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: d987f9e321e48e9259c8c2e4a3b736b78720a6c1

---

Overall Assessment

✅ Looks good — no unmitigated P0/P1 findings.

Executive Summary

• This re-review found the prior registry wording issue resolved: REGISTRY.md now documents four naming patterns at docs/methodology/REGISTRY.md:L5.
• No estimator math, weighting, variance/SE, identification assumptions, or defaults changed.
• Alias properties are pure read-throughs to canonical fields, including ContinuousDiDResults ATT-side aliases at diff_diff/continuous_did_results.py:L146-L183.
• Sibling-surface audit covered overall_*, overall_att_*, and avg_* result families; no missing targeted result class found.
• Pattern-wide inference grep found no new inline inference or partial NaN guard in the changed files.
• I could not run pytest because this environment lacks pytest and numpy; AST parsing of touched Python files passed.

Methodology

No P0/P1/P2 findings.

Severity: P3 informational
Impact: The affected methods are result API surfaces only: staggered overall_* result classes, ContinuousDiDResults, and MultiPeriodDiDResults. The aliases do not recompute effects, SEs, p-values, t-stats, or CIs, so they inherit canonical inference behavior. Examples: diff_diff/staggered_results.py:L141-L160, diff_diff/results.py:L450-L469.
Concrete fix: None required.

Code Quality

No findings. The implementation is simple property delegation and avoids new inference logic.

Performance

No findings. Property access is constant-time and adds no material runtime cost.

Maintainability

No findings. The registry contract and tests mirror the three changed prefixed surfaces: Pattern B overall_*, Pattern C overall_att_*, and Pattern D avg_*.

Tech Debt

No findings. I did not identify new untracked deferred correctness debt.

Security

No findings. Secret-pattern grep over the changed files did not find obvious credentials.

Documentation/Tests

No blocking findings.

Severity: P3 informational
Impact: Targeted validation could not be executed here because pytest and numpy are unavailable. The new tests do cover alias mechanics, NaN read-through for Pattern B, ContinuousDiD dual aliases, MultiPeriod aliases, and the balance-adapter regression at tests/test_result_aliases.py:L118-L301.
Concrete fix: None required for the PR; rely on CI to run tests/test_result_aliases.py.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 5b58d5f5722f9738eec0727ec83ff0721b09d3ae

---

Overall Assessment

✅ Looks good — no unmitigated P0/P1 findings.

Executive Summary

• No estimator math, weighting, variance/SE, identification assumptions, or defaults changed.
• Registry now documents the scalar result-field naming contract and alias semantics at docs/methodology/REGISTRY.md:L5.
• Alias implementations are pure read-throughs over canonical fields: Pattern B overall_*, ContinuousDiD overall_att_*, and MultiPeriod avg_*.
• Sibling-surface audit found the alias pattern mirrored across all targeted result families.
• Pattern-wide inference grep found no new inline inference computation or partial NaN guard in the changed result code.
• Could not run pytest: this environment has no pytest; AST parse passed for the touched Python files.

Methodology

Severity: P3 informational
Impact: Affected methods are result API surfaces only: CallawaySantAnna/staggered-family results, ContinuousDiD, and MultiPeriodDiD. The new properties do not recompute ATT, SE, t-stat, p-value, or CI, so they inherit existing safe_inference() behavior. Examples: diff_diff/staggered_results.py:L141-L160, diff_diff/continuous_did_results.py:L146-L183, diff_diff/results.py:L450-L469.
Concrete fix: None required.

Code Quality

No findings.

Performance

No findings. Property lookup is constant-time and adds no estimator runtime cost.

Maintainability

No findings. The registry contract and tests cover the three sibling naming surfaces: Pattern B at tests/test_result_aliases.py:L118-L181, ContinuousDiD Pattern C at tests/test_result_aliases.py:L189-L236, and MultiPeriod Pattern D at tests/test_result_aliases.py:L244-L258.

Tech Debt

No findings. I did not identify new untracked deferred correctness debt.

Security

No findings. Secret-pattern grep over changed areas did not reveal credentials.

Documentation/Tests

Severity: P3 informational
Impact: Tests cover alias mechanics, NaN propagation, read-only semantics, and the balance-adapter regression at tests/test_result_aliases.py:L293-L349. tests/test_practitioner.py:L125-L133 was updated for ContinuousDiD’s read-only overall_se alias.
Concrete fix: None required; rely on CI to run pytest because local pytest is unavailable.