[ricardofrantz]

Summary

This PR merges the release branch release/v0.1. It has three parts.
After the merge, pushing three tags makes the three GitHub releases.

1. machine-made Weft 0.1.0 (from weft/phase-2): a grid-built monospace with the axes wght 100 to 900 and FRAG 0 to 100, 9 static weights (TTF and OTF) and WOFF2. 948 tests pass. The build is byte-reproducible. Fontspector reports 0 FAIL. The OCR gate passes (FRAG 0: 0.23 %, FRAG 15: 0.11 %, DejaVu Sans Mono: 0.46 %).

2. machine-made Sans 0.1.0 (merged from sans/v0.1): a fork of Plus Jakarta Sans 2.071 with 2 variable fonts, 14 TTF, 14 OTF and 2 WOFF2. The release branch makes two changes to it:

   • Founder decision on the default forms. Only the straight (Lurus) forms are defaults: a j p q r y and their accented forms. All sharp (Lancip) forms are in ss02 "Lancip: sharp forms". ss01 gives the original round forms and ss03 is swirl. The f ligatures work in default text again, as in upstream. Cyrillic а р у ў ј follow the Latin defaults, and п ш ц do not change. The change is made only in sans/scripts/fork.py.
   • Line metrics. The ascender is now 906, the same as Weft 0.1 (Sans was made against the Weft phase 1 value, 875).
   • Build date. The build now uses a fixed SOURCE_DATE_EPOCH, so a rebase or squash merge cannot break byte reproducibility.

   117 tests pass and 1 skips (the italic has no Cyrillic). Fontspector reports 0 FAIL and 0 ERROR. varLib.interpolatable finds no problems.

3. Specimen site 0.1.0 (merged from site/specimen):

   • site/build.py now uses only the committed WOFF2 files. The stand-in fonts and the FONT_SOURCE switch are removed. A missing font or a missing readability report stops the build, so site.yml cannot deploy without the real fonts.
   • The site reads proofs/weft/readability.json in the shape that weft/tools/readability.py writes.
   • The Sans page describes the new defaults and has ss02 as a switch.

Repository root:

• fonts/ and proofs/ are now tracked normally; no git add -f is needed. Build scratch stays ignored.
• New README.md, STATUS.md (stage "released 0.1"), CHANGELOG.md (an index to the family changelogs plus a site entry) and LICENSING.md.
• The root OFL.txt is removed. It was a copy of weft/OFL.txt, and each family carries its own.
• New .github/workflows/release.yml and .github/scripts/package_release.sh.

Licence decision

• Fonts: SIL OFL 1.1 with the Reserved Font Name "machine-made".
  • Weft: weft/OFL.txt.
  • Sans: sans/OFL.txt. It keeps the Plus Jakarta Sans copyright line and adds the machine-made line with the RFN (OFL FAQ 5.10).
  • The Weft stroke tables (weft/loom/glyphs.py, extended.py) are font software under the OFL, as their headers say.
• Code: Apache-2.0 (LICENSE). This covers the generators, scripts, tests, CI and the site.
• LICENSING.md maps each path to its licence.
• Credit: "machine-made Sans is based on Plus Jakarta Sans by Gumpita Rahayu, Tokotype". It is in the README, the Sans docs, the site and the font name table.

What to review

• Sans defaults. Look at proofs/sans/sans-stylistic-sets.png and proofs/sans/sans-alternates-before-after.png: default, ss01, ss02, ss03, and the fi/fl ligatures.
• sans/scripts/fork.py. Look at plan_forms, rewrite_features and the tests in sans/tests/test_fonts.py.
• .github/workflows/release.yml. Three jobs, one per tag prefix:
  • The job builds again and fails if the build is not byte-identical to the committed fonts.
  • It runs the tests and checks.
  • It packages the files, attests them and runs gh release create.
  • Permissions are contents: write, id-token: write and attestations: write, on the jobs only.
  • Every action is pinned to a full SHA, verified with git ls-remote.
• site.yml. The release-download step is removed. A new step checks that the built site has the fonts and the report.
• The site build. Check the Weft FRAG slider, the readability table and the Sans ss02 switch.

Local run (fresh clone of release/v0.1 at 060e381, 2026-10-26)

Part	Result
Weft	pytest 948 passed. Build is byte-identical to fonts/weft. varLib.interpolatable: no problems. Fontspector: 0 FAIL / 0 ERROR (VF, TTF, OTF). OCR gate: pass.
Sans	Build is byte-identical to fonts/sans (32 files). pytest: 117 passed, 1 skipped. Fontspector: 0 FAIL / 0 ERROR (VF, TTF, OTF, WOFF2). varLib.interpolatable: no problems. Proofs are byte-identical to proofs/sans.
Site	Build OK. The built-site check passes. The build stops when a font is missing. In Chromium: the FRAG slider changes the glyphs (0 to 100), the readability table has 4 rows, and the Sans ss02 switch works. No console errors. No horizontal scroll at 360 px.
Release packaging	package_release.sh makes all three zips and the single files. The OCR report from the release step equals the committed report.

Not run locally: the workflows on GitHub runners, the attestation step and gh release create (gh is not authenticated here), and the Fontspector network checks.

Release steps after the merge

1. Merge this PR into main. site.yml deploys the site to Pages. Before this, set Pages source to "GitHub Actions".
2. Make the repo public (founder approved).
3. Tag the merge commit on main and push the tags:

   git fetch origin main
   git tag weft-v0.1.0 origin/main
   git tag sans-v0.1.0 origin/main
   git tag site-v0.1.0 origin/main
   git push origin weft-v0.1.0 sans-v0.1.0 site-v0.1.0

4. release.yml creates "machine-made Weft 0.1.0", "machine-made Sans 0.1.0" and "machine-made type site 0.1.0". Check the assets, then check one file with gh attestation verify MachineMadeWeft-Regular.woff2 --repo machine-made/type.

Open risks

• The workflows have not run on GitHub yet.
• Byte reproducibility on Ubuntu runners is not proven. If ttfautohint or another tool gives other bytes, the release job stops and prints the structural difference. In that case, rebuild on the runner, or relax the check to compare_builds.py.
• The Weft look-alikes (l/1, I/1, O/0) are harder for OCR than in DejaVu Sans Mono.
• Accented capitals go above the 906 ascender in both families.
• Confirm OFL FAQ 5.10 on openfontlicense.org (the build machine could not reach it).
• "Wefts" is a near-name match (web search only).

Cost: one integration agent session (about 95 tool calls; token count not measured in the session). Local CPU only: 9 Sans builds (about 1.5 min each), 2 Weft builds, 3 OCR gate runs, Fontspector runs, and one fresh-clone end-to-end run. No GitHub runner minutes and no paid APIs.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UHtj1XPp34FFs8LVpY1PRF

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Central YAML; Path: .coderabbit.yaml
• Review profile: CHILL
• Plan: Advanced
• Run ID: 538c6511-241b-49db-9f2f-4975234810a7

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[cubic-dev-ai]

This PR has 10,309 reviewable changed lines after ignored/generated files are excluded, above this repository's 5,000-changed-line automatic review limit. The raw diff is 219,742 lines before ignored/generated files are excluded.

Most of the diff comes from:

• weft/loom/pieces.py (~834 changed lines)
• sans/scripts/fork.py (~650 changed lines)
• weft/reference/weft-sketch-0.1.html (~639 changed lines)
• site/build.py (~624 changed lines)
• weft/loom/build.py (~444 changed lines)

Comment @cubic-dev-ai review this to review it anyway. If the largest files are generated or fixture data, add them to your ignored files in review settings or reviews.ignore.files in cubic.yaml. Later automatic reviews cover only subsequent changes. Request a full review to revisit earlier changes. You can also raise this limit in review settings.