Competitive sprint swimming performance is strongly influenced by the ability to generate high propulsive impulses and to rapidly accelerate the body during starts and turns, which has positioned dry-land strength and power training as a central complement to in-water practice in many programs (Crowley et al., 2017). Resistance training is widely used to target neuromuscular qualities that are difficult to overload sufficiently in water, and systematic reviews indicate that appropriately prescribed resistance training can improve swimming performance and related determinants (Crowley et al., 2017; Fone and van den Tillaar, 2022). Plyometric approaches are also commonly implemented to enhance stretch–shortening cycle function and explosive lower-limb output, with evidence supporting improvements in swim start and turn-related performance following plyometric interventions (Bishop et al., 2009).

The state of the art shows that dry-land resistance training meaningfully improves strength-related qualities and can translate to faster swim performance across sprint and middle distances, although the magnitude of transfer varies across studies and training modalities (Crowley et al., 2017). Recent syntheses further suggest that the methodological characteristics of resistance training programs, including exercise selection and emphasis on strength versus power, can differentially affect outcomes in competitive swimmers (Jin et al., 2024). Similarly, controlled and quasi-controlled work indicates that plyometric training can improve jump-based proxies of power and sport-specific swimming performance, including start-related metrics and sprint times (Bishop et al., 2009; Sammoud et al., 2019).

Despite this progress, much of the literature remains focused on group-average effects and isolated endpoints, which can obscure meaningful inter-individual variability in adaptation and limit the translation to individualized programming (Ross et al., 2019; Bonafiglia et al., 2021). Contemporary methodological reviews highlight that “responder” classification is highly sensitive to statistical choices, outcome selection, and measurement error, and they recommend rigorous approaches when quantifying inter-individual differences in trainability (Bonafiglia et al., 2021). In swimmers specifically, reviews of dry-land interventions for starts and turns note substantial heterogeneity in training content and outcome selection, which complicates direct comparison across modalities and may mask distinct profiles of adaptation (Hermosilla et al., 2021).

Addressing these gaps requires analytical frameworks that integrate multiple neuromuscular indicators into interpretable constructs, quantify adaptations as multivariate change patterns, and formally evaluate the stability of responder classifications (Ross et al., 2019; Bonafiglia et al., 2021). Moreover, because variability in training response is now recognized as a reproducible biological and methodological phenomenon across exercise domains, identifying predictors of responsiveness represents a practical step toward precision-oriented training prescription (Ross et al., 2019), however, predictive signals are highly contingent on athlete level, intervention dose, and outcome definitions, and require external validation before any broad monitoring or talent-development application is implied. In competitive swimming, this integrated approach is particularly relevant because dry-land improvements may transfer differently across upper-versus lower-limb contributions, and because start and turn determinants can be influenced by distinct strength–power qualities (Bishop et al., 2009; Hermosilla et al., 2021).

Accordingly, the objectives of the present study were to quantify training-induced changes in upper- and lower-limb power constructs derived from multiple performance indicators and to compare responsiveness between training approaches using post–pre change scores. The study further aimed to identify stable responder signatures by clustering individuals in the two-dimensional change space defined by ΔUpper (Post–Pre) and ΔLower (Post–Pre), including explicit cluster-number selection and stability evaluation. Finally, the study aimed to exploratorily model predictors of responsiveness and responder-signature membership, examining the contribution of training modality and baseline characteristics to inter-individual variation in adaptation within this sample and intervention dose.

The present study examined construct-level power adaptations and stimulus-dependent response profiles (“responder signatures”) following distinct dry-land training approaches in competitive swimmers. The main findings were that maximal strength training produced the largest improvements in the upper-limb power construct and both experimental training approaches exceeded the control condition for upper- and lower-limb responsiveness; and that swimmers segregated into two stable response profiles in the ΔUpper–ΔLower change space with a strong association between training modality and cluster membership, with the observed separation largely reflecting assigned training stimulus (i.e., group allocation accounted for most of the cluster structure) rather than indicating training-independent responder phenotypes.

The first main finding was that upper-limb responsiveness was greater in maximal strength training than plyometric training, and both exceeded the control condition, while lower-limb responsiveness was increased in both training approaches compared with control but did not clearly differ between maximal strength and plyometrics. However, interpretation must be related to the intervention dose and participant level (Tier 2; ∼3.5 h·wk^-1 swimming). In such samples, large and rapid improvements are plausibly amplified by novelty, task familiarization, and early neural/motor-learning adaptations that predominate in the initial weeks of structured resistance/power training, whereas higher-tier swimmers with extensive, consistent strength-and-conditioning exposure would be expected to show smaller relative changes over the same time window and potentially require longer or higher-dose blocks to shift comparable constructs. Our results are consistent with recent controlled evidence in swimmers indicating that maximal strength training and plyometric training can both enhance sprint-relevant outcomes, with maximal strength training often yielding comparatively larger improvements in strength-dominant or force-dominant outputs (Amara et al., 2021). Broader syntheses in competitive swimming similarly report that combining swim training with structured strength training tends to improve performance more than swim-only approaches, although the magnitude of transfer depends on training content, athlete level, and the specificity of the dry-land stimulus (Fone and van den Tillaar, 2022; Wirth et al., 2022). The comparatively larger upper-limb composite benefits in the maximal strength condition are plausibly explained by neural and contractile adaptations characteristic of heavy strength training, including increases in efferent neural drive, improved voluntary activation, and altered motor unit recruitment and rate coding, which manifest early and are particularly relevant for high-force actions that underpin propulsive phases (Sale, 1988; Del Vecchio et al., 2019; Tøien et al., 2021). The robust improvements in the lower-limb construct across both training interventions align with the shared capacity of maximal strength and plyometrics to enhance force–velocity characteristics and stretch–shortening cycle performance, which are central to jump-derived outcomes and are strongly implicated in swimming start and turn performance (Potdevin et al., 2011; West et al., 2011; Keiner et al., 2021).

The second main finding was the emergence of two stimulus-dependent response profiles in ΔUpper–ΔLower space, characterized by a lower-change cluster and a higher-change cluster, with good bootstrap stability and a pronounced association between training group and responder category; indeed, cluster membership closely mirrored training allocation (MSTG concentrated in the higher-change cluster and CG concentrated in the lower-change cluster), implying that the “responder signatures” primarily summarize between-group stimulus effects rather than intrinsic, modality-independent responder biology. These signatures are defined specifically in the ΔUpper–ΔLower change space (i.e., the constructs as operationalized in this study) rather than implying global responder “types.” This aligns with a growing body of work showing substantial inter-individual response variability to standardized training and that “responder” classification is sensitive to the outcomes selected, measurement error, and the analytic method used to form subgroups (including clustering) (Potdevin et al., 2011; Basri and Turki, 2025). Within swimming-specific contexts, controlled trials and syntheses of dry-land modalities indicate that the imposed training stimulus can materially influence both the magnitude and heterogeneity of performance-related changes, which is consistent with (though not definitive proof of) stimulus-dependent responder structure (Lopes et al., 2021). Two-cluster structure in ΔUpper–ΔLower space is consistent with the idea that adaptation is constrained by the magnitude of the imposed neuromuscular stimulus and (within a given stimulus) inter-individual differences in training dose uptake, fatigue–recovery balance, and motor learning (Mann et al., 2014). However, in the present dataset the dominant determinant of cluster membership was the assigned training stimulus, and disentangling intrinsic individual factors from modality effects will require designs with larger samples, broader outcomes (including in-water transfer), and/or clustering within-modality strata. In practical terms, the high stability estimates observed for both clusters suggest that the responder signatures were not merely stochastic partitions of noise but reflected reproducible patterns in the multivariate change signal.

In practical terms, the high stability estimates observed for both clusters suggest that the responder signatures were not merely stochastic partitions of noise but reflected reproducible patterns in the multivariate change signal. The dominance of the training stimulus accords with the evidence base in swimming indicating that targeted dry-land programs, particularly those emphasizing heavy strength or well-structured power training, are meaningful determinants of start, turn, and sprint-related outcomes, which are strongly influenced by maximal and explosive strength capacities (West et al., 2011; Keiner et al., 2021). Importantly, in a brief, low-dose block delivered to trained/developmental training group dominated the ridge/logistic models indicates that the ‘predictive’ structure is driven primarily by between-group stimulus allocation, with limited incremental personalization signal from baseline covariates in this context. The inverse association between baseline sprint performance and high-responder probability is consistent with ceiling effects and diminishing returns that are especially plausible in lower-trained or highly trainable samples over short interventions, where swimmers with higher initial status may show smaller observable changes because they operate closer to performance limits and require larger or more prolonged stimuli to shift outcomes (Mann et al., 2014; Bonafiglia et al., 2016). Accordingly, these predictive findings should be interpreted as conditional on participant level and intervention duration, and should not be generalized to athlete monitoring, talent development, or higher-tier swimmers without external validation in independent cohorts and longer training periods. A complementary explanation is that better baseline sprint performance may reflect greater technical proficiency and efficiency, reducing the relative contribution of dry-land power gains to the measured composite changes and thereby attenuating the apparent responsiveness signal, a phenomenon previously discussed in the swimming literature as limited transfer when dry-land gains do not align with in-water coordination and propulsive specificity (Wirth et al., 2022).

Some limitations should be considered. First, although eligibility required at least 2 years of resistance-training experience, we did not quantify participants’ prior dry-land strength and conditioning exposure or verify consistency with training logs. Baseline training status and familiarity with barbell versus jump/medicine-ball tasks may therefore have differed, making it difficult to disentangle modality-specific adaptations from generalized responses to the introduction of a supervised, progressive program. Second, while the use of composite constructs strengthens reliability and reduces multiplicity, it necessarily compresses information and may mask indicator-specific adaptation profiles, particularly if different neuromuscular qualities contribute heterogeneously across tests. This issue is compounded by partial overlap between training content and outcome measures, which can introduce specificity/learning bias. Specifically, bench press average power contributed to ΔUpper and was directly trained only in MSTG, whereas ΔLower comprised jump-based tests (CMJ, SJ, DJ, SLJ) that closely resemble exercises emphasized in PTG. Therefore, between-group differences, especially the MSTG advantage for ΔUpper, should be interpreted primarily as superiority in trained-task responsiveness rather than unequivocal evidence of greater generalized power transfer. Future trials should minimize exercise–outcome overlap (or pre-specify both trained-task and transfer endpoints) and prioritize outcomes with clearer in-water relevance (e.g., start/turn measures, in-water kinetics, and race performance). Finally, although clustering stability was formally evaluated, responder classification remains sensitive to feature choice and scaling and should be examined in larger multi-center samples. Importantly, the relatively small sample size limits generalizability and constrained longitudinal modeling, whereas larger samples would enable linear mixed-effects models for repeated measures, which are generally preferred for longitudinal data and can better accommodate individual trajectories and missing observations. While plyometric sessions were closely supervised and standardized with SSC-focused coaching cues, we did not collect objective kinematic or ground contact-time metrics during training; thus, we cannot confirm the extent to which all participants consistently executed a fast SSC across sessions, which may have introduced unquantified variability in the realized PTG stimulus. Additionally, because cluster membership aligned strongly with training allocation in this sample, the identified “responder signatures” should be interpreted primarily as stimulus-dependent response profiles (between-group structure) rather than training-independent individual phenotypes; future work should test whether similar profiles emerge within-modality, across longer periods, and using outcomes with clearer in-water transfer. In addition, the intervention was brief and low-frequency (12 sessions across 6 weeks) and the sample was trained/developmental with modest in-water volume (∼3.5 h·wk^-1), which likely increased short-term trainability and novelty effects; therefore, the magnitude and speed of the observed adaptations should not be assumed to generalize to higher-tier (Tier 3–5) or elite swimmers with higher training loads and longer strength-training histories.

From an applied perspective, in Tier 2 (trained/developmental) university swimmers completing a short, low-dose dry-land block (12 sessions/6 weeks), the present findings suggest that introducing supervised, progressive dry-land training, whether maximal strength–oriented or plyometric, can yield meaningful short-term improvements in the measured power constructs, with maximal strength training producing a larger increase in the upper-limb construct as operationalized here. Because responsiveness in this setting is likely amplified by high trainability and novelty/familiarization effects, these data should not be used to make strong preferential prescriptions for higher-tier or elite swimmers, for whom marginal gains over short blocks are typically smaller and more dose-dependent. Practitioners may therefore view MSTG and PTG as context-dependent options within periodized plans, and prioritize monitoring of individual change (ΔUpper/ΔLower) while integrating outcomes with in-water performance transfer (starts/turns and race metrics) when making programming decisions. However, any model-based ‘prediction’ shown here should be treated as exploratory.

Maximal strength training elicited the greatest improvement in the upper-limb composite construct (ΔUpper; bench press average power and medicine-ball throw), while both maximal strength and plyometric training improved upper- and lower-limb composite constructs relative to the control condition. Because the outcome constructs partially overlapped with the training content (bench press power in MSTG; jump-based tests in PTG), between-modality differences, particularly for ΔUpper, should be interpreted as construct-/trained-task responsiveness that may be influenced by specificity/learning effects, rather than as definitive evidence of generalized neuromuscular superiority or a basis for strong preferential recommendations. Swimmers also displayed two stable responder signatures in ΔUpper–ΔLower change space that were strongly associated with training modality. However, these responder signatures should be interpreted as modality-aligned patterns within this short intervention and sample, rather than as definitive evidence of baseline-determined “responder” profiles. Therefore these predictive findings should be viewed as context-conditional and not generalized without external validation. Larger and longer-duration trials that better quantify/equate training dose across modalities and incorporate in-water transfer outcomes are needed to confirm modality-specific transfer to swimming performance.