Corpus-linguistic measures of lexical diversity and syntactic complexity are increasingly used to characterize spoken accounts, including children’s narratives, in investigative-interview contexts. Their validity for measuring narrative development, however, remains uncertain because narrative quality is defined primarily by discourse organization rather than lexical variety or clause-level complexity. We examined 173 narrative-practice sessions from 44 children aged 6 to 12 who completed up to four five-minute interviews across one school week. Child-only transcripts were analysed for lexical diversity, MLU-w, an automated subordination ratio, and output volume. Mixed-effects models evaluated session patterns under multiple defensible specifications, including analyses that addressed an atypical whole-day recount prompt at Session 2. We also tested measures that predicted independently coded story grammar outcomes from a later transfer interview. Lexical diversity measures showed a non-monotonic session pattern, with a Session 2 decrement and partial recovery, but no reliable linear improvement. MLU-w was stable. Output volume increased to Session 3 before declining, whereas the subordination ratio showed a modest positive session trend. Individual slopes formed a continuum rather than discrete trajectory groups. Most associations between corpus-derived change scores and later story grammar were small and non-significant. These findings suggest that, given this protocol design, conventional lexical diversity and clause-level indices are sensitive to the elicitations themselves and do not provide a sufficiently validated proxy for within-child narrative structure change. Corpus measures remain useful descriptive and diagnostic tools, but claims about narrative development require measures explicitly validated against discourse and structural criteria.