Mean seed-disperser ant detection shares were higher inside exclosures in one Harvard Forest block and lower in the other across four July surveys, 2012–2018

We tested ant trait shares inside and outside large-browser exclosures at eight Harvard Forest plots. We used 1,140 pitfall trap–species detections from four recorded July surveys in 2012, 2013, 2014, and 2018. Each year had equal weight within each plot. Each plot had equal weight. Six species lacked a binary seed-dispersal trait value. We kept their detections in the denominator and tested all 64 assignments, with one shared value per species. In the Ridge block, the mean inside-minus-outside share was +4.31 to +5.24 percentage points under all assignments. In the Valley block, it was −19.72 to −13.20 points. The overall mean was −7.54 to −4.15 points. These are sharp bounds with the observed detections fixed, not confidence intervals. Removing one Valley hemlock-control plot allowed either sign in the remaining mean. The 2018 contrast alone was −0.14 point. A paired sign-flip reference gave tail fractions from 0.383 to 0.789 and did not establish a general exclosure effect. The contribution is a reproducible trait-share analysis with exact missing-trait bounds and a small temporal extension of prior ant studies. The endpoint measures recorded ant composition, not seed transport. Only two forest blocks were sampled. Grid placement, sparse detections, and the lack of a full trap-event table limit causal and population claims. The package has unchanged source tables, provenance, code, all results, figures, and independent and clean-repeat checks.

ants · seed dispersal · browser exclusion · Harvard Forest · missing traits · partial identification · pitfall sampling