Prospective bounded joint source proposal and exact readmission, 2026-10-18
Question: exhibit one strictly negative-successor-target-pressure ideal source for the same two-call acquired negative-outward target-minimum Mutation renewal. This is a constructive finite search, not a proof of global exclusion. No old envelope is refined and no zero-pressure candidate is silently substituted for q0<0.
Model: fresh exact represented-default normalized pressure/stability law from _native_mutation_feasibility._owned_law; paired unit doubled C5, held active capacities, dt=1/16, Gamma=0, no clipping; source Dissonance (OZ) target0 with p0>OZ_TINY and high-Si Coherence (IL) surrounding blocks, followed by acquired Mutation (ZHIR) target0 and high-Si IL surroundings. Retain supplied AL/IL/EN grammar/priorIL/ordinal3, empty initial timestamped history, remaining memory zero, counts/capacity adaptation not reached, no ZHIR_MIN_VF override. First controller consumes source phases; second masks consume first endpoint phases and refreshed pressure, second controller consumes post-Mutation phases with carried gains. This is ideal detached dynamics, not a native execution or physical identification.
Search coordinates and domain: five capacities in [1/64,VF_MAX], four source phases in [-3,3] with theta0=0 and common width<=3, four independent source pressure coordinates in [-Pbar,Pbar] with p4=-sum. Pbar=2*k+b*VF_MAX+3*(w/pi_rep) is a necessary source pressure magnitude ceiling from the full rails. The missing fifth pressure bound is an explicit constraint. All source and both endpoint forms use one common offset, eliminated by intersecting their fifteen exact/model form-offset rails. No separate endpoint offset, pressure amplitude, capacity maximum, phase or mask witness is selected. The lower capacity cutoff restricts this finite SEARCH, not the theorem's active-capacity domain.
Fixed proposal budget: six declared pairs in lexicographic owner order (first stable,transition,dissonant; second stable,transition); four matched initial points from scipy.stats.qmc.Sobol(d=13,scramble=True,seed=20261018).random_base2(m=2), used unchanged across pairs. Each run uses scipy.optimize.minimize(method='SLSQP') on unit-box coordinates, at most 400 iterations and at most 5000 objective evaluations, ftol=1e-12, eps=1e-7, no custom Jacobian, no parallel workers or polish/restart. The objective is the sum of squared negative normalized constraint margins, with fixed desired margin 1e-6 for strict predicates and zero for weak predicates. This desired margin is proposal policy, never a changed admission threshold. Normalize phase widths/gaps by 3, source-pressure bounds/sign/ratio numerators by Pbar, rate by xi, offset interval by 2, capacities by VF_MAX, and Si/controller R-squared constraints by 1. Use literal same-source maxima/absolute values and binary64 trigonometry solely for proposal. A zero peak uses zero ratio as the model does; exact positive pressure/sign conditions remain separate. Retain all original selector constraints, both controllers, complete renewal/terminal-target-minimum geometry, active capacities, fifth pressure bound and common offset. Stop on budget without launching replacements.
Retention: archive every proposal objective evaluation and its primitive unit vector, the selected lowest-penalty observed point for each of24 runs (first tie), solver termination/message/evaluation/iteration counts, source hashes and environment. No post-result objective reweighting, tolerance change, new initialization, extra source evaluation or new algorithm. Program defects, if any, require a separately recorded correction before a changed run; do not overwrite original evidence.
Exact readmission: each of24 selected points is converted once to dyadic primitives with denominator 2^40, nearest ties-to-even. Rebuild the exact shared compiler from normalized capacities/phases and four pressure coordinates; p4 is exact zero-sum. Compute the shared eta interval from all15 exact source/endpoint offsets and choose its midpoint (a failed interval remains rejected). Rebuild all relevant original strict/weak cell rows and determine actual source/successor peak owners from the same exact fields, including ties. Explicitly require q0<0 and P1>0; q0=0 cannot pass. Derive the first secant from its two associated exact form samples at t=1 and17/16, not from supplied history or a cached rate. Check all geometry. Certify both actual R-squared predicates from the retained exact phase differences using certified_cosine_bounds with terms=12,bits=128, exactly once per candidate; enclosures straddling a required boundary are unresolved, never passing. No result-dependent precision refinement or rationalization retry.
Result classes: exact rejected source (violated algebra/geometry/sign/controller predicate), unresolved controller enclosure, or conditionally admitted ideal joint source. Only a fully admitted joint source answers the constructive question. A numerical success flag or a failed24-run search proves neither existence nor nonexistence. Even an ideal source remains distinct from native measured history, bit-exact wrap/event behavior, persistence, formation and physical identification.
Integration: a private point-verification owner reuses _prepare/_cell and shared exact admission/cosine kernels; it does not import scipy, choose sources, install events or execute native dynamics. The proposer remains an archived bounded research driver. Preserve old compiler/reducers/instruments and frozen evidence byte-identical. Controls may use earlier formal coefficient frames, malformed representations, controlled row/interval boundaries and mocked wiring, without solving additional sources. Independently reconstruct ten-node fields/flows, common offset, masks, temporal secant and controller bounds. Rejected candidates retain exact failed inequalities; do not claim an all-domain certificate. Update theory, contracts, testing and sole queue to actual result; affected tests, style, links, mirror/generated documentation and strict site build.