When is anti-holomorphic dependence forced? Exchange non-Hermiticity of one-way transfer kernels and a proposed time-domain test in a recirculating fiber loop

Sealed v11 of "When is anti-holomorphic dependence forced? The one-waytransfer kernel" (21 pages). This version repairs and supersedes thepublished v3 (10.5281/zenodo.21317960). WHAT THIS VERSION IS. A corrigendum-strengthened edition produced throughnine rounds of external adversarial audit (GPT-5.6), every catch verifiedon the sources and, where mathematical, machine-certified by exact SymPyjudges on the author's machine BEFORE entering the text. Intermediatedrafts v4-v10 existed only internally and were never deposited; the fullfail-closed patcher chain (build_v5.py ... build_v11.py) that regeneratesthis text from the sealed v3 is archived in the companion repository. MAIN REPAIRS SINCE v3.- The v3 conflation of exchange Hermiticity and reciprocity is repaired: R = K - S(K) (reciprocity) and Delta = K - M(S(K)) (exchange Hermiticity) are proven independent, with all four quadrants realized by certified kernels; the Bergman kernel is reclassified as a Hermitian decoy (Delta = 0, R != 0).- Projection trap: the scalar closed form is exchange-symmetric and blind to R; one-wayness lives in G12 = 0 with G21 != 0 on the matrix kernel. Every invariant now states the object it is evaluated on.- The jump-phase witness is comparative and conditional (sufficient, never necessary; separable two-sector class, port basis C = I), with the exact pointwise size law W(x) = (1-rho) D(x).- A direct measurement of R_G requires reverse injection AND phase-referenced detection; this branch is specified but not simulated, with an explicit perimeter at five places in the text.- Two laws born from audit counter-examples and machine-certified (6-8 exact clauses each) before entering the text: positive weights move the deep interference minima from N >= 3 on (the only guaranteed volume knob is a common rescaling), and the continuum dichotomy is one of AMPLITUDE - decaying is not forgetting (the paper's own uniform sinc keeps exact zeros forever).- Anteriority claims deflated throughout ("apparently new" / "appears unreported in the literature we searched"). REPRODUCIBILITY. The PDF and TeX are sealed (SEAL_v11.txt: pdf sha256f510cf18..., tex a801936d...). The companion repository(github.com/antparis/oneway-kernel) archives the certification harnesses,the findings files underlying every certified claim (#052-#061), thepatcher chain, and deposit_gate.sh, a mechanical fail-closed pre-depositverification (25 pinned checks). PERIMETER. Machine-verified mathematics and a simulated model with anexperimental proposal. No experimental data exist; nothing beyond themodel is claimed. The next step is a measurement.