1. More efficient capture, manipulation and swallowing of food can reduce the time spent foraging or increase food intake, both of which are beneficial for survival. Feeding performance is therefore generally regarded to be under strong natural selection.

2. In granivorous songbirds, feeding is a complex process as seeds need to be dehusked before they can be consumed, making the feeding act a biomechanically challenging endeavour. However, most previous research has focused on how beak morphology affects feeding performance, while the influences of beak kinematics remain largely unknown.

3. In this study, we hence investigated at the individual level how feeding performance (i.e. seed processing time and success rate) relates to both the athletic capacity of the beak (i.e. beak tip speed, acceleration, frequency) and skill (i.e. seed handling tactics) in the Canary (Serinus canaria). To do so, high-speed videos during feeding were recorded and subjected to automated tracking of beak tip movements.

4. Better skills, i.e. accurate positioning of the seed for being split in half, had a positive impact on total seed handling time compared to more random positioning and crushing the husk into multiple, scattering fragments. Surprisingly, individual variation in beak speed, acceleration, or frequency generally did not relate to differences in performance.

5. Thus, our data suggest that seed positioning precision, and hence the control of coordinated beak and tongue movement, is critical to minimize feeding durations in songbirds. Further studies are needed to explore whether this develops via a positive feedback between behaviour, learning and increased efficiency or if it relates to intrinsic differences.

• Feeding performance (phase durations and success rate) and skill (cracking mode and head position) data were manually extracted from video data. Raw video data can be accessed upon request by contacting the corresponding author (Tim Andries)
• Kinematic data were calculated in Microsoft Excel from 3D-coordinate data obtained through use of DeepLabCut automated pose estimation software on the video data.
• All data were further analysed using R statistical software (version 4.2.1).

According to the guidelines of the respective journal, all data files are submitted in .csv format. We still advise opening the data in Microsoft Excel or similar programs.