A finite SAT experiment, not a P vs NP proof. Four variables form a cycle: three equality edges and one inequality edge. Each edge becomes two CNF clauses.
Deleting any one edge leaves a satisfiable path. This refutes only the shortcut that satisfiable edge-deleted pieces imply a satisfiable whole. It does not establish consistency of compatible partial assignments, nor a resolution lower bound: propagation refutes this example easily.