Nadarasa · Track 3 · Synesthete's PQP Frontier
Resource frontier map
Each kernel is a point in (spider-count, T-count UB). The Pareto frontier (lower-left) collects kernels that no other kernel dominates on both axes. Frontier kernels are hardware-demo candidates; dominated kernels are structurally wasteful for the same scientific question. Anchored in PQP Ch. 13.
n = 4
- G1 · modal projector(58, 3)
- G2-real · two-coset combiner(81, 3)
- G12 · modal-projection scaling(58, 3)
- G3 · per-window cost(58, 3)
n = 6
- G1 · modal projector(129, 10)
- G2-real · two-coset combiner(160, 10)
- G12 · modal-projection scaling(129, 10)
- G3 · per-window cost(129, 10)
n = 8
- G1 · modal projector(228, 21)
- G2-real · two-coset combiner(267, 21)
- G12 · modal-projection scaling(228, 21)
- G3 · per-window cost(228, 21)
What it reveals
- G1 / G12 / G3-cost occupy the same frontier point — they are diagrammatically identical, so the chart can't distinguish them. That is itself a certification: the v0.3 closed forms applied to one apply to all.
- G2-real sits strictly above and to the right on every
n ≥ 3. It is dominated on the resource axes — its extra cost is paying for the two-coset combiner, not for efficiency. - The resource-monotone picture is a slice of a larger validation hierarchy. G16 gives the ideal frontier, G17 adds noise and ZNE, G18 adds an independent estimator, and G19 frames the cross-platform corroboration layer. Step C-2q (v0.4.4) confirms the symbolic rule (M) identity itself survives realistic H2-class noise on Selene shots, and G20 (v0.4.5) closes the corroboration layer by reproducing the G19 Floquet subharmonic collapse natively on Selene, with Richardson ZNE recovering the ideal peak to within about one percent.
- The v0.4.3 changelog ties the G16–G19 layers together; the v0.4.4 changelog records the completed Step C-2q noise sweep.