Observed resource tradeoffs
The reciprocal construction used fewer operation nodes and edges than a strong SUM control on the mixed table. It did not dominate all coordinates on the scalar table. Deeper recurrence improved exact error without improving retained-mass accuracy past the observed precision plateau.
Limits of the result
These are measured competitors on two known tables, not an exhaustive graph optimum or a general model-quality advantage. Preserve unresolved range and evidence failures when reproducing the comparison.