kimera 0.1.2 gate, non-ASCII churn fix, release 0.10.1

[giacope]

This PR has four commits: two tighten the kimera gate, one fixes churn for non-ASCII file names, and one cuts 0.10.1.

1. kimera 0.1.2, with the gate workaround removed

• Gemfile.lock: kimera 0.1.1 → 0.1.2, updated by bundle update kimera --conservative.
• The CI mutation job no longer passes --no-fail-on-no-coverage or sets KIMERA=1. CONTRIBUTING's command changes to match.
  • The coverage flag was there because kimera 0.1.1 couldn't overlay methods in cli/flag.rb's Data.define block, so all of them were reported as uncovered. On that file, 0.1.1 reports 11/11 as no_coverage (gate exit 2), and 0.1.2 kills 11/11.
  • KIMERA=1 is no longer needed because 0.1.2 sets it in its own process before loading the suite. spec_helper's coverage floor still skips under it.
• .kimera.yml: fail_on_no_coverage: true. A full run under 0.1.2 found 0 no_coverage mutants among 1,996.

2. fix(churn): count files whose names aren't ASCII

By default, git log --name-only quotes non-ASCII paths ("lib/caf\303\251.rb"), so churn never matched those files and scored them 0. GitLog now runs git with -c core.quotePath=false, which gives the raw bytes Churn already compares byte for byte (since #5).

The non-ASCII churn spec now runs with core.quotepath set to both true and false, under UTF-8 and US-ASCII default encodings. Before the fix it fails with churn 0 instead of 1; after the fix it passes. End to end, on a project with lib/café/menú.rb and git's default settings, churn goes from 0 to 1 under both locales.

3. Release 0.10.1

VERSION is now 0.10.1, the changelog heading ## [0.10.1] - 2026-09-26, and Gemfile.lock is regenerated by bundler. It's a patch release because the Unreleased section has only a "Fixed" heading. The format follows chore: release 0.10.0.

Verification (on this head)

• bin/ci passes both with LANG=C.UTF-8 and with LANG/LC_ALL unset: rubocop, rubocop-kata doctor, 502 examples, the gate, and Ratchet OK: 16 edges, 0 findings, unchanged.
• bundle exec kimera ci --since origin/main, run exactly as the new CI step (no KIMERA, no override flag), exits 0.
  • The only library line changed, the git log argument list, yields no mutants; the churn spec above covers it.
• gem build hashira.gemspec produces hashira-0.10.1.gem.

Out of scope: 10 survivors already on main

A full kimera run has 10 surviving mutants, the same 10 under 0.1.1 and 0.1.2, so max_survivors: 0 already fails the full run on main. CI doesn't run the full suite, only changed lines per PR, so it never showed. Most are memoization guards (return @x if @x → if false). The others are fragment.mass >= PREFILTER → > in duplication/clusters.rb, a deleted walked in complexity/cognitive_score.rb, and an unless words.empty? → unless false in coupling/echo.rb. They need triage of their own.

After merge

Tag v0.10.1 on the merge commit, create the GitHub release, and gem push hashira-0.10.1.gem.

🤖 Generated with Claude Code

https://claude.ai/code/session_01PcDfzLuRgjoSnfPLrTaA4h

---

Generated by Claude Code