Measure graph-guided writing under generation-time length matching

Determine how graph-guided story generation compares with event-list story generation when output length is held fixed during generation, thereby separating the effect of the narrative graph from the effect of producing longer texts.

Background

Stories generated with the full narrative graph were longer than stories generated from a bare event list, and the apparent reader preference for graph-guided stories disappeared after statistical adjustment for word count. Regression adjustment cannot establish what would happen if the two conditions had been generated at equal length.

A length-matched generation run would directly test whether the graph improves story quality independently of verbosity.

References

We can say that the graph-versus-event-list preference does not survive adjusting for word count, and we cannot say what that comparison would look like with length held fixed at generation time.

— Can LLMs Tell Stories Out of Order? A Corpus and Benchmark for Nonlinear Narrative Generation  (2610.21468 - Raiyan et al., 15 Oct 2026) in Limitations, paragraph “The length confound is diagnosed, not removed”