Skip to content

docs: retire CLAUDE.md, keeping what is still true - #251

Open
BenjaminDEMAILLE wants to merge 1 commit into
mainfrom
docs/retire-claude-md
Open

docs: retire CLAUDE.md, keeping what is still true#251
BenjaminDEMAILLE wants to merge 1 commit into
mainfrom
docs/retire-claude-md

Conversation

@BenjaminDEMAILLE

Copy link
Copy Markdown
Contributor

Closes #196, implementing James's proposal.

What goes

CLAUDE.md in full. Its status figures ("396 tests passing"), known-issue list and PE numbers were all stale, its dependency section is superseded by DEPENDENCIES.md (#209), and its build commands duplicate CONTRIBUTING.md.

What stays, moved into CONTRIBUTING.md

Two sections were saying something the rest of the docs do not, and neither is agent-specific:

  • "When a STAR-faithful change makes the numbers worse" — that a benchmark moving the wrong way after a faithfulness fix usually means the surrounding logic has not caught up, and that keeping an approximation because it currently measures better is how a port stops converging. It sits next to the divergence rule now, with the added ask that the before/after go in the PR.
  • The code conventions — camelCase parameter spelling, num_args and allow_hyphen_values, validation in Parameters::validate(), no async, thiserror and anyhow.

What is deliberately not carried over

The source-layout tree. It was already wrong (no degnorm/, no solo/ detail), and a hand-maintained file tree drifts faster than the code it describes. cargo doc and the module-level docs do that job without going stale.

Dangling references

Removing a file that five places point at is how a docs cleanup makes things worse, so each pointer is updated rather than left:

  • Cargo.toml: dropped from the publish exclude list.
  • DIVERGENCE.md: two "see CLAUDE.md" pointers now name STAR-RS-COMPARISON.md and ROADMAP.md.
  • CONTRIBUTING.md: the faithfulness baseline points at ROADMAP.md.
  • STAR-RS-COMPARISON.md: three mentions rephrased, keeping the point that the figures there were stale.

grep -rn CLAUDE.md over the tracked tree comes back with only those rephrased mentions. cargo build and the library tests are green.

🤖 Generated with Claude Code

Per #196: the file was added so contributors using Claude would start from the
same place, and it has drifted. Its status figures, known-issue list and PE
numbers were all stale, its dependency list is superseded by DEPENDENCIES.md
(#209), and its build commands duplicate CONTRIBUTING.md.

Two sections were worth keeping and are now in CONTRIBUTING.md, next to the
rules they belong with:

- "when a STAR-faithful change makes the numbers worse" — keeping an
  approximation because it currently measures better is how a port stops
  converging, which is worth saying to every contributor and not only to an
  agent.
- the code conventions: camelCase parameter spelling, num_args and
  allow_hyphen_values, validation in Parameters::validate, no async,
  thiserror/anyhow.

The source-layout tree is deliberately not carried over: it was already wrong
(no degnorm/, no solo/ detail), and a hand-maintained file tree drifts faster
than anything it documents.

Every reference to the file is updated rather than left dangling: the publish
exclude list in Cargo.toml, two pointers in DIVERGENCE.md, the faithfulness
baseline in CONTRIBUTING.md, and three mentions in STAR-RS-COMPARISON.md.

Closes #196.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Removal of CLAUDE.md

1 participant