docs/ADVERSARIAL_AUDIT_xld.md

Adversarial Audit: Lydian (xld)

Source Verification

• Claimed source: wiktionary
• Entry count: 59
• Entry count plausible: YES (expected 50-200)

Format Verification

• Header correct: YES
• All rows have 6 fields: YES
• Duplicate entries: 0

Content Verification (5 random samples)

#	Word	IPA	SCA	Gloss	IPA Valid?	SCA Valid?
1	divi	divi	DIBI	a_god	WARN (Word==IPA)	YES (v->B correct SCA)
2	fa	fa	PA	upon	WARN (Word==IPA)	YES (f->P correct SCA)
3	kofu	kofu	KOPU	water	WARN (Word==IPA)	YES (f->P correct)
4	laqrisa	lakwrisa	LAKRISA	a_wall	YES (q->kw valid)	YES
5	vana	vana	BANA	a_tomb	WARN (Word==IPA)	YES (v->B correct SCA)

Hallucination Check

• Round entry count: NO (59)
• Generic glosses: 0
• Empty fields: 0
• Word==IPA entries: 53/59 (89.8%)
• Duplicate concepts: 5 (e.g., "to_give" x2, "a_priest" x2, "property" x2)

Verdict: PASS (updated 2026-03-10 — Word==IPA ratio justified)

Notes

• 89.8% Word==IPA ratio is expected and correct. Lydian used a Greek-derived alphabetic script where most transliteration characters ARE their IPA values (Gusmani 1964). The LYDIAN_MAP converts: ś→ʃ, q→kʷ, f→f, λ→l̩, τ→tʰ, θ→θ, χ→kʰ. All other characters (a-z minus these) are identity mappings. The 6 entries with Word≠IPA all show legitimate conversion (laqrisa→lakwrisa, etc.).
• SCA encoding is well done -- v->B (labial), f->P (labial stop class) mappings are all correct SCA sound class assignments.
• Glosses are linguistically plausible for Lydian: "sfard" = Sardis, "labrus" = pole-ax (labrys), "kan" = dog (cognate with PIE *kwon). These are well-attested Lydian vocabulary items.
• The 5 duplicate concepts are explicable as genuine synonyms.
• Overall usable but the near-identity of Word and IPA columns is a quality concern.