This study experimentally investigates the role of statistical learning in grammatical change. We test the hypothesis that speakers adapt their grammatical intuitions by tracking distributional frequencies in variable linguistic input. Although statistical learning is widely acknowledged as a key mechanism in language adaptation, the precise impact of changing input distributions on acceptability judgments remains unclear. To address this, we designed an experiment simulating grammatical change within intra-lingual variation. The study examines variable person-number agreement with conjoined subjects in Russian, focusing on structures containing both pronominal and nominal conjuncts. The core question we are investigating is whether manipulating the frequency ratio of competing grammatical variants in the input affects speakers' judgements of acceptability. We conducted two parallel experiments, each focusing on a distinct pair of agreement strategies. These pairs were chosen based on prior experimental work; within each pair, the two variants were matched in acceptability, with one pair occupying the high end and the other the middle range of the acceptability scale. Using a between-subjects design, we manipulated the presentation frequency of each variant across five levels. We hypothesized that exposure to different frequency ratios would not only shift the absolute acceptability of variants but also induce a significant difference between them. The results, however, were inconclusive. Although we observed a shift in ratings, the effect was anomalous and non-systematic, preventing us from interpreting it as evidence for statistical learning; it may instead reflect a methodological artifact. However, we identified a robust link between acceptability judgments and the well-formedness of the co-occurring linguistic material.