docs: utility coverage gaps (RTD audit Batch C)

[igerber]

Summary

RTD audit Batch C closes three utility-coverage gaps surfaced during the
docs refresh initiative:

• New docs/api/profile.rst - covers profile_panel plus the four
  supporting dataclasses (PanelProfile, OutcomeShape,
  TreatmentDoseShape, Alert). All five are public top-level exports
  but had no API reference. Lands under a new "Pre-Fit Profiling" toctree
  section in docs/api/index.rst.
• New docs/api/local_linear.rst - covers all 11 top-level
  local-linear infrastructure symbols (kernels, kernel-moment constants,
  local_linear_fit, mse_optimal_bandwidth,
  bias_corrected_local_linear, and result dataclasses). Lands under a
  new "Infrastructure" toctree section. Deliberately not placed under
  "Estimators" - the symbols are building blocks HeterogeneousAdoptionDiD
  composes, not estimators themselves.
• docs/conf.py - html_extra_path now includes
  llms-autonomous.txt (65KB). The file ships in the wheel but was not
  previously exposed via Sphinx, so the practitioner-profile prose link
  to it from profile.rst:11 now resolves on RTD.

docs/doc-deps.yaml: extends the existing local_linear.py entry to
point at the new api/local_linear.rst; adds a new top-level
diff_diff/profile.py entry at drift_risk: low (descriptive-only by
design, sits with prep.py / datasets.py not the linalg.py /
utils.py methodology-stake tier).

Drive-by: 1-line RST formatting fix in BaconDecomposition Parameters
docstring (blank line before bullet list) noticed during the deferred
Sphinx -W investigation.

Methodology references (required if estimator / math changes)

• N/A - no methodology changes. Only diff_diff/bacon.py is touched in
  source, and only the RST formatting of an existing docstring (no logic,
  no signature, no semantics).

Validation

• Tests added/updated: none required (docs-only PR; no behavior change).
  The new RST files contain no .. code-block:: python directives, so
  tests/test_doc_snippets.py reports zero new cases.
• tests/test_doc_snippets.py baseline (locally, pure-Python backend):
  111 passed, 4 skipped (unchanged from origin/main; new files
  introduce no snippets).
• Sphinx HTML build (no -W): zero warnings on profile.rst,
  local_linear.rst, index.rst, zero new autosummary stub
  warnings. Pre-existing backlog of 2,200+ warnings on other files stays
  out of scope (deferred to a separate -W cleanup PR).
• Rendered-HTML spot checks: all 5 profile symbols and all 11
  local-linear symbols render in their respective pages;
  docs/_build/html/llms-autonomous.txt is served from html_extra_path;
  profile.html contains exactly one href="../llms-autonomous.txt"
  matching the prose link.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment
✅ Looks good

No unmitigated P0/P1 issues in the changed files. The diff is docs-only apart from a docstring whitespace fix, and I found one P3 informational documentation mismatch in the new local_linear page.

Executive Summary

• diff_diff/bacon.py is a formatting-only docstring change; no estimator logic, defaults, weighting, or inference behavior changed.
• The new profile docs match the actual public exports and the llms-autonomous.txt link wiring added in docs/conf.py.
• The new local_linear page covers the correct public symbols and points readers to the right HAD/nprobust methodology context.
• P3 informational: the local_linear overview implies broader standalone applicability than the public Phase 1b/1c wrappers actually expose.
• Review limitation: I could not rerun Sphinx or import-based tests here because sphinx and runtime deps such as numpy are not installed in this checkout.

Methodology

• Severity P3 — docs/api/local_linear.rst:L8-L18 says the components are usable “for any one-sided boundary local-linear regression problem with a strictly nonnegative running variable,” but the public mse_optimal_bandwidth and bias_corrected_local_linear wrappers are explicitly HAD-scoped and restricted to the documented Phase 1b/1c contract in diff_diff/local_linear.py:L695-L705, diff_diff/local_linear.py:L972-L980, and docs/methodology/REGISTRY.md:L2522-L2524.
  Impact: Minor documentation overstatement only; the implementation and lower-page autodoc remain correct, so this does not change behavior or methodology.
  Concrete fix: Narrow the overview sentence to distinguish the generic helpers (local_linear_fit, kernels, kernel moments) from the HAD-scoped public wrappers (mse_optimal_bandwidth, bias_corrected_local_linear).

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• No findings.

Security

• No findings.

Documentation/Tests

• No additional findings in the changed docs.
• Verification limitation: I could not rerun the Sphinx build or tests/test_doc_snippets.py in this environment because the checkout lacks sphinx and import-time dependencies such as numpy.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: d81589e7133f0a417a74c6c8311ce18aa9f23d5a

---

Overall Assessment
✅ Looks good

Executive Summary

• No unmitigated P0/P1 findings in the changed files.
• The only source-code change is a docstring whitespace fix in diff_diff/bacon.py:L323-L333; no estimator logic, defaults, weighting, inference, or identification behavior changed.
• Re-review result: the prior P3 on docs/api/local_linear.rst has been addressed. The new scope split in docs/api/local_linear.rst:L13-L32 now matches the public-wrapper restrictions documented in diff_diff/local_linear.py:L683-L705, diff_diff/local_linear.py:L964-L980, and docs/methodology/REGISTRY.md:L2522-L2524.
• The new profiling docs accurately describe the public API and top-level re-exports in docs/api/profile.rst:L4-L28, diff_diff/profile.py:L1-L12, diff_diff/profile.py:L165-L206, and diff_diff/__init__.py:L253-L260, diff_diff/__init__.py:L505-L510.
• The docs wiring is coherent: docs/api/index.rst:L104-L115 and docs/api/index.rst:L262-L276 add discoverability, docs/conf.py:L74-L79 exposes llms-autonomous.txt, and docs/doc-deps.yaml:L405-L423 remains valid YAML.
• Validation limitation: I could not rerun import-based or pytest/Sphinx checks here because this environment lacks numpy and pytest.

Methodology

• No findings. The changed docs are consistent with the Methodology Registry and in-code local-linear docstrings; no undocumented methodological deviation was introduced.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• No findings introduced.

Security

• No findings.

Documentation/Tests

• No findings. The new profile page and autonomous-guide link align with the existing guide accessor in docs/api/profile.rst:L11-L28, diff_diff/_guides_api.py:L7-L15, and docs/conf.py:L74-L79.
• Validation note: docs/doc-deps.yaml parses successfully, but I could not run tests/test_doc_snippets.py or an end-to-end import diff_diff smoke test in this environment because pytest is absent and numpy is not installed.