[ronaldtse]

Summary

Pins the two Khmer lineage sources precisely in WO68:

• The khm-latn lineage (27.4 CER) trains on secryst/data-khmer-translit — 17,911 pairs, currently fetched from raw.githubusercontent (a data-law violation to retire), and the repo carries no license file. It is the predecessor project's own repository, so the fix is the owner adding a license or blessing the migration into our Releases channel — no third-party negotiation.
• khmer-dict-spice is the DICT/IPA line (ipa_dic.txt 3,286 gold pairs, SBBIC 72,048 words, the dictionary DB), LGPL-2.1, already recorded — it does not contain the latn pairs.

Consequence: the CER < 15 recovery path needs no third-party license; SEAlang / full Chuon Nath licensing only gates expansion beyond the current data. The recovery plan (Releases re-release with sha256, Modal-to-HF-Jobs port, retrain, eval vs 27.4, strict publish) is recorded.