Sync from GitHub: 080bffac0f304b7f3781266ca4fb761c2974f8a6

grammars/concepts.gbnf

@@ -28,14 +28,12 @@
 
 root    ::= item ("," ws item){0,7}
 item    ::= word (ws word){0,3}
-# Word: first char is any letter (allows TitleCase proper nouns
-# like "Calvin" or "RSA"); body is lowercase + digits + hyphen only.
-# Mid-word capitals are forbidden. This forces Falcon3's BPE
-# tokenizer to emit space-prefixed continuation tokens (" Sunlight")
-# rather than jamming tokens together as one camelCase "word"
-# ("thatGreenhouseCarbon"). Per arXiv 2502.14969, space-prefixed
-# tokens have better-trained embeddings (5-10% gain), which is
-# particularly important for smaller/quantized models like
-# Falcon3-10B-1.58bit.
-word    ::= [a-zA-Z] [a-z0-9-]{2,19}
+# Word: any letter + alphanumerics + hyphens. Mid-word capitals
+# are permitted because legitimate concepts frequently contain them:
+# acronyms (RSA, CPU, DNA, ATP, NADPH), patronymic proper nouns
+# (McDonalds, MacPhearson), brand/product names (iPhone, eBay).
+# Length-cap of 20 chars + the defensive parser's word-count gate
+# handle the run-7 token-jam pollution ("thatGreenhouseCarbon")
+# without sacrificing legitimate mid-capital content.
+word    ::= [a-zA-Z] [a-zA-Z0-9-]{2,19}
 ws      ::= " "