[owenthcarey]

Summary

Rewrites the README as a front page for the project. It was 445 lines, about 300 of them RFC-by-RFC history, and its status and performance claims had gone stale (the last performance figure was 2.85 times CPython's time).

• Drops the history instead of moving it. Every wave it described already has its own RFC in docs/rfcs/, and docs/PERFORMANCE-PROGRESS.md covers the recent work. The README now links to both.
• Replaces the claims with dated, measured results, each linked to its source, and adds a section on known limitations.
• Fixes setup details: the regrtest path that didn't exist, the crate list (7 of 13), and the unsafe claim, which is gone. It also now follows the style rules in AGENTS.md.

Where the numbers come from

Claim	Source
459 of 468 Lib/test entries pass	Full --all-cpython sweep run for this PR on macOS at 666e32b6
460 bundled tests	Same sweep; all pass
48 of 48 packages	tests/ecosystem/expectations.toml and RFC 0077's re-measure; not re-run here
Warm, cold, and memory ratios	docs/PERFORMANCE-PROGRESS.md, October 30, 2026, PGO builds; not re-run here

Worth a look

• The sweep isn't clean. It reports 6 unexpected rows on main, while docs/CONFORMANCE.md still describes a clean baseline:

  • test_capi, test_getpath, and test_sysconfig fail.
  • test_multiprocessing_spawn, test_multiprocessing_forkserver, and test_pyrepl time out.

  test_sysconfig's failing test passes when run on its own, and a small spawn and forkserver pool works, so some of these may come from the sweep's setup. I didn't root-cause them. The README lists them as open rather than claiming a pass.

• The new CI badge can show red. bench gate (blocking, macos-latest) failed on main for perf: reach warm parity with CPython via direct JIT calls #102 and also for docs: record the merged performance work #100.

Testing

• Ran the quick-start commands from a release build, including -m venv and pip install requests.
• Ran the regression, CPython source fetch, and pybench commands with filters. The fetch went to a scratch directory rather than vendor/.
• Confirmed WEAVEPY_JIT=0 and WEAVEPY_FRAME_JIT=0 control separate tiers, from the code and a timing probe.
• Checked that the README's relative links resolve, and python3 tools/check_repository_artifacts.py passes.
• Not run: tools/pgo/build.sh, cargo xtest, cargo xclippy, and the full package lane. Only the README changed.