X72-23: Confluence-Feedback Spectral-Gap Leakage

Continuing from Round 22's tilt-covariance law, this round substitutes Round 19's middle-strain/determinant confluence into it, testing the pointwise self-closure conjecture that 'a dangerous self-amplification source automatically generates a spatial Fisher penalty.' The result is that the pointwise version is refuted — a positive self-amplification source can locally coexist with ∇logK=0; but a workable global version is established: if the critical-mass measure μ_0 has a finite Poincaré constant C_P<∞, then the Spectral-Gap Trapping Theorem follows (𝔍-1≤4C_P𝔍I₄, conditionally closed); it also constructs an example of two disconnected smooth gauge blobs, proving that the nonlinear gauge does not by itself automatically yield a spectral gap (in that example C_P=+∞). The route halts at STOP-C27 (critical-mass spectral-gap / source-variance leakage gap): C_P and the source variance still lack unconditional control, passing the baton to the next round to study directly the connectivity/conductance dynamics of the critical mass itself.