16 families still lack source provenance — help needed identifying upstream repos

[felipesanches]

Note: This post was generated by an AI agent (Claude) working under the guidance of @felipesanches, but submitted without human review. @felipesanches himself would still need to participate in the PR thread if he wants to contribute to the review.

Context

Over the past week, we've made significant progress documenting source provenance for the Google Fonts library. Through systematic investigation of METADATA.pb files, upstream repo discovery, and the archival of 982+ bare git mirrors (55 GB), we've brought source block coverage from 83% to 96%+ of all 2,002 families.

Key milestones:

• PR Add source metadata for 229 families from googlefontdirectory-hg #10370 (merged): Added source metadata for 229 families from the googlefontdirectory-hg monorepo (the pre-GitHub canonical source for early Google Fonts, 2010–2013)
• PR Add source metadata for 28 Noto families #10371 (merged): Added source metadata for 28 Noto families with blob-verified commit hashes
• PR Fix source paths in 25 override config.yaml files #10373 (merged): Fixed source paths in 25 override config.yaml files
• PR modak: add source metadata from EkType/Modak #10378 (merged): Added source metadata for Modak (EkType)
• PR Add source metadata for 54 families (batch 4) #10380 (closed for rework, will be resubmitted): 59 additional families from canonical designer repos

The remaining 16 families

After all known repos are accounted for, 16 families still have no identified upstream source repository. We're asking for help from the community — if you know where the sources for any of these fonts live, please comment here.

Family	Designer / Foundry	Scripts	Last Binary Update	Onboarding Date
Black And White Picture	AsiaSoft Inc.	Korean	2018-03-13 (16680f8688)	2018-02-27
Chenla	Danh Hong	Khmer	2021-11-08 (84b31698cb)	2011-03-02
Content	Danh Hong	Khmer	2015-03-07 (90abd17b4f)	2011-03-02
Cute Font	TypoDesign Lab. Inc	Korean	2018-03-13 (16680f8688)	2018-02-23
Dokdo	FONTRIX	Korean	2018-03-13 (16680f8688)	2018-02-23
Gaegu	JIKJI SOFT	Korean	2018-03-13 (16680f8688)	2018-02-28
Poor Story	Yoon Design	Korean	2018-03-13 (1ef157d393)	2018-02-23
PT Mono	ParaType	Cyrillic	2015-03-07 (90abd17b4f)	2012-02-29
PT Sans	ParaType	Cyrillic	2015-03-07 (90abd17b4f)	2010-09-21
PT Serif Caption	ParaType	Cyrillic	2015-03-07 (90abd17b4f)	2011-02-09
Single Day	DXKorea Inc	Korean	2018-03-13 (81997650b4)	2018-02-23
Sitara	Neelakash Kshetrimayum	Devanagari	2015-06-08 (7e42686751)	2015-06-10
Song Myung	JIKJI	Korean	2018-03-13 (16680f8688)	2018-02-23
Stylish	AsiaSoft Inc	Korean	2018-03-13 (16680f8688)	2018-02-27
Sunflower	JIKJISOFT	Korean	2018-03-16 (5ea1323a54)	2018-02-27
Uchen	Christopher J. Fynn	Tibetan	2019-12-11 (b44e8365d6)	2019-12-07

(Noto Color Emoji Compat Test is also without a source block, but it's a test font created entirely within google/fonts — no upstream exists.)

Patterns

• 8 Korean families from foundries with no known GitHub presence (AsiaSoft, TypoDesign Lab, FONTRIX, JIKJI, Yoon Design, DXKorea, JIKJISOFT)
• 3 ParaType PT families — sources appear to be proprietary/internal
• 2 Danh Hong Khmer families — danhhong GitHub user has repos for Khmer and Siemreap, but not Chenla or Content
• 1 Tibetan family (Uchen) — designer Christopher J. Fynn; Savannah project free-tibetan exists but may not contain this specific font
• 1 Devanagari family (Sitara) — designer has no known GitHub presence
• 1 Korean family (Sunflower) from JIKJISOFT with Hangul script

What we've searched

For each of these families, we've already checked:

• GitHub search (by family name, designer name, foundry name)
• The googlefontdirectory-hg monorepo (pre-GitHub canonical source)
• The librefonts/ GitHub org (TTX mirrors only, not original sources)
• Font binary name tables for embedded URLs
• FONTLOG.txt and DESCRIPTION.en_us.html for repo references
• Google Code Archive, SourceForge, Launchpad, Font Library
• google/fonts git commit history and PR bodies

Next steps

Beyond resolving these 16 families, the planned next phase of this work is:

1. Reproducible builds: Ensuring the full library can be reliably built from sources. We've already tested 1,381 families (306 byte-identical, 921 compiler-version match). The build system improvements (multi-license-dir support, local archive extraction, legacy source classification) are ready for broader testing.

2. Source preservation: We've built a permanent repo archive of 982+ bare git mirrors (55 GB). The upstream source preservation investigation documented that 20%+ of families had source reliability issues (force-pushed repos, deleted repos, missing commits). The archive ensures these sources survive upstream changes.

3. Preventing drift: Establishing tooling and processes to ensure source provenance information stays accurate as new fonts are onboarded and existing fonts are updated. This includes periodic verification that all METADATA.pb commit hashes remain reachable in their upstream repos.

cc @NicholasJohnson @nicholasjohnson-monde @AsiaSoftInc (if any of these accounts are active) — any leads on the Korean font sources would be very helpful. The nicholasjohnson/* repos (23 fonts) were previously deleted but some may have been re-hosted elsewhere.

cc @davelab6 @rsheeter — would appreciate any leads on the remaining 16 families, especially the Korean and ParaType ones.

[rsheeter]

submitted without human review

Why didn't you review it? - this puts the burden on others to do so, that is not desirable.

Most notably, did you check that the #s claimed are correct? - if they are made up they aren't terribly useful.

[felipesanches]

submitted without human review

Why didn't you review it?

Sorry... here's my review, below.

Replying a question you just sent me in private ("how do I interpret this? Is there a lot of stuff yet to add to crater, do we have source blocks that are insufficient to compile, ...?"):

What it means is that only 16 families lack upstream repo url. But such repos are not necessarily sources that gftools-builder currently builds. There are plenty of sources in legacy formats such as sfd, vfb, etc... (and there's also binary only upstream repos). That's why crater has way smaller number.