Root-cause analysis (RCA) in surface-mount assembly must support rapid containment and verification while remaining auditable under imperfect evidence. We present an uncertainty-aware neurosymbolic pipeline in which a neural evidence layer converts per-board observations into probabilities for defect class, stage-wise mechanism, and parameter risk, and a deterministic semantic layer assembles and ranks defect–mechanism–parameter-violation hypotheses over an explicit causal mapping. The output is an RCA packet: a small ranked hypothesis set with evidence and provenance pointers (observation value, specification bounds, timestamp, and model version). We evaluate on a synthetic multi-shift dataset with correlated variation and graded ground truth, reporting set-level agreement (hit rate, Jaccard similarity, exact match), triage behavior (mean reciprocal rank), and edge-level fidelity with a cycle sanity check.