docs: reposition around distributional regression

[jxucoder]

Summary

• Retitles the docs around distributional regression and varying-coefficient models, the established names for what NaturalBoost/WeibullAFT and FormulaBoost actually fit. The previous framing ("Boost every parameter") was not a statistical term and read as a slogan. The differentiator is now stated plainly: the full K x K natural gradient rather than a diagonal approximation.
• Slims the README from 182 to 128 lines. The four benchmark tables move to a new docs/benchmarks.md, which is now the source of truth the README summarizes.
• Stops leading with the NGBoost speedup ratio on pitch surfaces, because the suite is incomplete: three UCI datasets are unmeasured, and XGBoostLSS/LightGBMLSS are not compared at all. The measured ratio now appears only on the benchmarks and GPU-setup pages, and the README states its own limits.
• Comparison sections describe what each library targets rather than what it cannot do ("How it compares" replaces "Why not NGBoost or XGBoost").
• Removes em dashes throughout (roughly 90 occurrences), rewriting each rather than substituting, so table placeholders became n/a, list labels took colons, and prose cases became commas, semicolons, or new sentences.

New pages: docs/benchmarks.md, docs/user-guide/how-it-works.md, docs/user-guide/formulaboost.md, docs/user-guide/survival.md.

No behavior changes. The only source edits are docstrings, plus the pyproject.toml description and keywords.

Test plan

• uv run mkdocs build completes with no errors or broken links
• All three README code snippets executed against the CPU backend and produce the documented shapes and keys
• No em dashes remain in README.md, docs/, CLAUDE.md, CHANGELOG.md, or examples/README.md
• import openboost succeeds and the rewritten package docstring renders
• Reviewer check: benchmark claims in docs/benchmarks.md match the committed Modal runs

Made with Cursor

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.