The approach works without a labelled answer. The model looks for groups, patterns or a compressed representation in the data it is given. Because there is no target, the framework's evaluation logic cannot use the same measures as supervised learning; the result is judged by how useful or stable the structure is, and that judgement needs to be recorded. Clustering and dimensionality reduction are named in the task clause, so the two clauses are read together. The clause text is not in the public sample, and this page does not pretend to quote it.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.