Developing unobtrusive systems for digital health monitoring requires addressing fundamental design challenges, such as understanding the role of spatial context, the detectability of repeated activities, and the impact of sensor placement. The SLICE study investigates these aspects in a feasibility setting. In a multiroom living lab, twelve participants performed routine and deliberately repetitive variants of everyday tasks, including handwashing, table cleaning, and checking if a door is closed. Using wearable motion sensors and Bluetooth beacons, we evaluated traditional and deep learning models across four tasks: activity recognition, execution pattern differentiation, spatial context, and sensor placement. In the multiclass activity recognition task, the Random Forest model reached F1 scores of 0.63 for handwashing, 0.68 for table cleaning, and 0.34 for door checking, resulting in a macro F1 of 0.55. The LSTM model performed less well overall, with F1 scores of 0.40, 0.42, and 0.17 (macro F1 = 0.47). Across tasks, distinguishing routine from repetitive execution remained inconsistent, suggesting that simulated repetition alone may not yield distinct sensor signals. Location features improved accuracy, and the right wrist provided stronger input for asymmetric activities. We also describe the modular data collection platform D4L Collect, which supported synchronized multimodal sensing, task annotation, and questionnaire delivery. The findings demonstrate the potential of combining wearable and location data for behavioral monitoring and illustrate how feasibility studies can inform the design of robust digital health systems for everyday use.