Abstract
The Distributed Partial Information Puzzle (DPIP) is a collaborative problem-solving paradigm in which essential information is distributed among participants, making communication a requirement for task success. Unlike traditional shared-information tasks, DPIPs deliberately prevent any single participant from solving the puzzle independently, thereby enforcing communication as the primary means of coordination. In this dataset, the DPIP task is implemented as a collaborative construction challenge performed in groups of four, where each of three directors is given different partial information about a structure and must collaboratively instruct one builder to assemble a single contiguous structure out of large Lego blocks. The task is complete when all three directors agree that the builder's construction is consistent with their individual views, and the group is considered successful if that structure actually matches the originally generated side views. In this data collection implementation, the goal structure has dimensions of 3 (width) by 3 (depth) by 3 (height), where the unit of height is a layer of blocks and the unit of width and depth is 2 of the pegs atop a Lego block (meaning a single square Lego is considered to be 1x1x1). The structure is composed entirely of square or rectangular blocks, and there are no gaps between blocks in any of the walls. While a variant of the task featured a 4x4x3 footprint and allowed for curved sides or gaps, the annotations and evaluation provided in this dataset extend only to the 3x3x3 variant described above.