perf: avoid full-width post-clean narrative profiling

[WilliamK112]

Summary

• restrict explain_clean()'s post-clean context build to columns that can actually contribute a narrative
• skip the post-clean context build entirely when there are no rationale-bearing actions or remaining missing values
• retain full profiling for the pre-clean role table and the pipeline's separate post-representation engine cache, avoiding stale context reuse
• add a 120-column regression test and an Unreleased changelog entry

Why

explain_clean() unconditionally profiled every cleaned column even though _narratives() only reads contexts for columns with rationale-bearing actions or for columns that remain missing. On a wide no-op frame, that doubled per-column context construction from 120 to 240 calls. Reusing the pipeline cache directly would be unsafe because it is built after representation repair and only covers engine-visible columns, while the role table describes the original frame.

The new selection preserves the existing narrative semantics while avoiding the redundant full-width pass.

Closes #32

Validation

• red-first regression: the new wide-frame test failed before the implementation with 240 context builds instead of 120
• pytest -m "not online and not large" — 4,052 passed, 6 skipped, 12 deselected; 93.16% coverage
• pytest --no-cov -q tests/test_explain.py — 14 passed
• ruff check src tests
• mypy src/freshdata — no issues in 201 source files
• git diff --check

AI assistance

I used OpenAI Codex to inspect the current profiling flow, implement the optimization, and run validation. I reviewed the final diff and test evidence before submission.

[coderabbitai]

Important

Review available on request

• 🔍 Trigger review

Reviews should be triggered manually for repositories with fewer than 10 stars. Select Trigger review above or comment @coderabbitai review to review the latest changes. For a full review, comment @coderabbitai full review.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: ba5d7352-9498-4518-8269-76d602a63669

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[strix-security]

Strix is installed on this repository, but we couldn't run this PR security review because this workspace's trial has ended. Add a card to resume code reviews here.

So far, Strix has reviewed 36 pull requests, surfaced 3 security issues (1 critical/high) and blocked 1 risky merge across this workspace.

[JohnnyWilson16]

Hi @WilliamK112,

Apologies for the delay in getting around to reviewing this!

First and foremost, thank you so much for your continued dedication to the freshdata project. From your work on integration recipes (Great Expectations, PyJanitor, ydata-profiling), sentinel handling, and test coverage improvements, to this performance fix—your contributions have made a substantial, high-quality impact across the entire codebase.

Regarding PR #181:

• The _narrative_contexts filtering approach is clean, targeted, and safe—it avoids redundant profiling on wide frames without introducing stale cache risks.
• The 120-column regression test and changelog entry are spot on.

LGTM! 🚀