docs/ADVERSARIAL_AUDIT_xle.md

Adversarial Audit: Lemnian (xle)

Source Verification

• Claimed source: wiktionary
• Entry count: 30
• Entry count plausible: YES (expected 20-80)

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	naphoth	nap_hot_h	NAPOT	grandson	YES (ph->aspirated, th->aspirated)	YES
2	sialkhvis	sialk_hwis	SIALKWIS	sixty	YES (kh->aspirated)	YES
3	maras	maras	MARAS	magistrate	WARN (Word==IPA)	YES
4	mav	maw	MAW	and_(conjunction)	YES (v->w)	YES
5	phokiasiale	p_hokiasiale	POKIASIALE	for_the_phocaean	YES (ph->aspirated)	YES

Hallucination Check

• Round entry count: YES (30 -- round number)
• Generic glosses: 13 ("meaning_uncertain" entries)
• Empty fields: 0
• Word==IPA entries: 16/30 (53.3%)
• Duplicate concepts: 12 (mostly "meaning_uncertain" x9, plus "meaning_uncertain" variants)

Verdict: PASS (updated 2026-03-10 — "meaning_uncertain" is honest annotation)

Notes

• 13 out of 30 entries (43.3%) are glossed "meaning_uncertain". This is linguistically honest — Lemnian is known from essentially one major inscription (the Lemnos stele) plus a few minor ones. Most words remain undeciphered. Glossing them as "meaning_uncertain" is more honest than inventing translations, and follows the data-extraction principle of never fabricating content.
• The entries that ARE glossed are reasonable: "maras" (magistrate -- cognate with Etruscan maru), "mav" (and), "naphoth" (grandson -- cognate with Etruscan nefts/naphths), "avis" (year -- cognate with Etruscan avil).
• The IPA conversion correctly handles Greek-style aspiration markers: ph->p_h, th->t_h, kh->k_h.
• 53.3% Word==IPA ratio is acceptable given the alphabetic script.
• The round count of 30 is a minor concern but matches the actual attested corpus size.
• This is one of the smallest datasets but faithfully represents the available evidence.