{"claim_id":"cnisa-ii-01","number":1,"kind":"theorem","section":2,"title":"Graph-distance and response bounds","statement":"At zero signal, the endpoint transfer norm is at most min(1, a^d/d!) and the absolute signed analyzer mean is at most 2a^d/d!. When d ≥ 2, the endpoint-block transfer error is at most a²/2, its signal derivative error is at most La²/2, and the signed mean slope deviates from 2L cos(2εL) by at most 2La². The leakage amplitude is at most min(1, a).","assumptions":"A passive unitary coupled-mode network of length L, a real direct endpoint signal ε, a graph-restricted Hermitian defect of integrated operator norm at most a, and a common unitary sideband evolution. The input starts in endpoint u and d is the u-to-v graph distance.","scope":"Bounds hold uniformly over the common sideband motion and the number of sidebands. The second-order response bounds use the absence of a direct nuisance edge between the endpoints."} {"claim_id":"cnisa-ii-02","number":2,"kind":"theorem","section":3,"title":"Sharp order of exact ambiguity","statement":"For fixed d and small a, the maximum exactly hidden clean rotation angle is Θ_d(a^d). It is bounded above by arcsin(a^d/d!) and below by θ_d(a/C_d), where C_d = 3·2^(d−1)−2, θ_1(h)=h and θ_j(h)=2 arcsin(sin(θ_{j−1}(h)) sin(h/2)).","assumptions":"There is a finite shortest path of distance d ≥ 1. The nuisance can apply time-varying signed rotations with controllable phases on each path edge, within integrated action a.","scope":"The constructed nuisance matches the full clean terminal unitary and leaves spectator modes unchanged. Arbitrary terminal probes, ancillary systems and terminal tomography retain the ambiguity. In-flight measurements are outside this experiment."} {"claim_id":"cnisa-ii-03","number":3,"kind":"theorem","section":4,"title":"A resource-matched lower bound","statement":"At the null, the marginal classical Fisher information obeys I(0)/(4nτL²) ≥ (1−a²)²/(1+nτb_d²), where b_d=2a^d/d!. The denominator 4nτL² in the information ratio is the value of the matched clean analyzer at the same exposure.","assumptions":"The graph distance is d ≥ 2 and the defect action budget satisfies a < 1. A balanced analyzer uses equal known scalar efficiency τ in (0,1]. Use a coherent probe of mean incident number n, or n identically prepared photons with conditionally independent outcomes. The waveform is shared by the photons within each shot and has a fixed signal-independent law across shots.","scope":"Marginal information describes a calibrated nuisance law. Information adds across independent shots. The manuscript treats an adversarial uncalibrated sign-decision problem separately."} {"claim_id":"cnisa-ii-04","number":4,"kind":"theorem","section":5,"title":"Local minimax separation","statement":"The smallest signal-angle separation that permits both worst-case testing errors to be at most a fixed α is Θ_(d,α,θ*)(a^d+n^(−1/2)) as a tends to zero and n to infinity. Divide the angle by L to obtain the coupling separation. A threshold on the absolute signed count attains the order.","assumptions":"Fix a graph of distance d ≥ 2, an error level 0<α<1/2, and a local angle window 0<|θ|≤θ*<π/4 with θ=εL. Use exactly n identically prepared photons, or the n-photon number state in the specified input mode, with full collection. The null and alternative can have different admissible waveforms with action at most a.","scope":"The full separation law concerns a single terminal exposure with the specified probe. The a^d lower-bound mechanism is exact terminal-channel equality and thus also constrains other terminal probes."} {"claim_id":"cnisa-ii-05","number":5,"kind":"corollary","section":5,"title":"Unequal edge tolerances","statement":"With a per-edge action budget c_e ρ^(w_e), the exact ambiguity exponent is D=min_path Σ_e w_e. The minimax angle separation in the same experiment has order ρ^D+n^(−1/2).","assumptions":"Take 0<ρ<1 with ρ tending to zero, positive constants c_e and w_e, a fixed graph and fixed edge budget parameters. Edge phases and signs are independently controllable.","scope":"The weighted minimum includes any admissible direct endpoint link. A weak shortcut can determine the exponent if its weight is smaller than that of the existing path."} {"claim_id":"cnisa-ii-06","number":6,"kind":"proposition","section":7,"title":"Exact carrier compression","statement":"For a carrier input, the carrier-compressed propagator equals the free carrier block D_0(L,0). Every nonempty scattering word advances the mode index, so the quotient F_0/F_1 evolves freely.","assumptions":"A bilateral mode ladder has a block-diagonal free evolution with a common exponential bound and only positive shifts S^r⊗W_r(t), r≥1, with time-integrable Σ_r ||W_r(t)||.","scope":"The equality holds pathwise for arbitrary internal matrix dimension and noncommuting shifts within the stated model. Open filtration-preserving cutoffs retain it. Cyclic wrap, population below the carrier, or forbidden return terms can invalidate it."} {"claim_id":"cnisa-ii-07","number":7,"kind":"proposition","section":7,"title":"Multiport cancellation to all orders","statement":"With a tensor-sum generator, the propagator is U_K⊗U_M. All transmitted sidebands carry the same internal projective direction. Conditioning analyzer counts on the total removes the common collected intensity factor, which equals one for lossless mixing and complete collection.","assumptions":"The generator is K(t)⊗I+I⊗M_ε(t) with well-defined evolutions. The input is a product sideband and internal state, and the same internal analyzer is applied to the collected sidebands.","scope":"Conditional information requires nonzero throughput. For a comparison per incident photon, collection and throughput remain relevant."} {"claim_id":"cnisa-ii-08","number":8,"kind":"theorem","section":8,"title":"Exact first observable defect order","statement":"Start with R_0 as the closure of the lifted input span under zero-cost letters. Form R_{j+1} by adding each lifted defect image of R_j and closing again under zero-cost letters. The first d for which the output wedge is nonzero on R_d is the first observable defect order. Stabilization inside the wedge kernel gives exact equality of output directions wherever both outputs are nonzero.","assumptions":"Finite-dimensional paired systems have a common finite control alphabet, a drift among the zero-cost letters, and independently varied piecewise-constant controls on variable-duration intervals. A common scalar δ scales the defect letters. The readouts share an output space.","scope":"There are at most nm−dim(R_0) strict extensions for system dimensions n and m. A fixed duration, tied controls, or bandwidth constraints can remove the witnessing inputs. Erasure requires a separate throughput check."} {"claim_id":"cnisa-ii-09","number":9,"kind":"proposition","section":11,"title":"Exact finite-length Gaussian quadrature","statement":"For F=∫₀ᴸ e^(ibt)w(t)dt and the specified return R, Re R=δ|F|². Hence E[Re R]=δV and Var(Re R)=δ²(V²+|P_b|²), where V=E|F|² and P_b=E[F²] have explicit finite-length formulas. The full complex variance has a covariance-operator Hilbert–Schmidt formula.","assumptions":"The forward and backward drives share one centered real stationary Gaussian path of covariance σ²e^(−|t−s|/ℓ). The two nearest sidebands use equal phase detuning b and there is no damping in the specified return kernel.","scope":"The quadrature variance uses Gaussian fourth-moment pairing. An arbitrary process sharing the same covariance need not share the result. The fixed nonzero detuning asymptotic is nonuniform near b=0."} {"claim_id":"cnisa-ii-10","number":10,"kind":"theorem","section":12,"title":"Nonuniform additive sewing","statement":"The sums of the local germ on arbitrary nonuniform tensor grids converge in norm to a separately additive rectangle increment. The error is bounded by terms of orders Δ_t^(θ−1) and Δ_z^(η−1), with constants determined by the enclosing rectangle controls and defect bounds. Appropriate domination also yields L^p convergence.","assumptions":"The Banach-valued germ vanishes on degenerate rectangles. For continuous superadditive controls U,V, the horizontal, vertical and mixed split defects are bounded respectively by C_h U^θ V^β, C_v U^α V^η and C_m U^θ V^η, with θ,η>1 and 0<α,β≤1.","scope":"The limit is for separately additive increments along tensor refinements. Normalization and both one-sided bounds are required in addition to mixed control."} {"claim_id":"cnisa-ii-11","number":11,"kind":"proposition","section":13,"title":"Finite quotient and all closed poles","statement":"At a primitive root q of order m the carrier vacuum resolvent reduces to P(z)/Q_m(z) and has m simple poles z_j=(2/b)cos((2j−1)π/(2m)), 1≤j≤m, where b²=1−q and Q_m(z)=2b^(−m)T_m(bz/2) for the Chebyshev polynomial T_m.","assumptions":"The algebraic ladder is Te_n=e_{n+1}+[n]_q e_{n−1} with e_{−1}=0, [n]_q=(1−q^n)/(1−q), and m≥2. The invariant tail starts at index m and the numerator is the trailing principal minor of zI minus the quotient matrix.","scope":"This specializes the continuous q-Hermite identity. The associated loaded-port calculation concerns a finite coupled-mode model with explicit port loading. Physical poles need a model and experimental fit."} {"claim_id":"cnisa-ii-12","number":12,"kind":"proposition","section":14,"title":"Positive measures and determinacy","statement":"The algebraic vacuum moments of the ladder admit a positive Hamburger measure on the real line exactly for real q≥−1. The measure is unique for −1≤q≤1 and indeterminate for q>1.","assumptions":"Moments are m_j=e_0* T^j e_0 on the finitely supported ladder core, taking the usual [n]_1=n convention. Representing measures must be positive Borel measures on the real line with all indicated moments.","scope":"This is the positive q-Hermite moment problem in the manuscript’s normalization. It addresses real positive measures and leaves open the separate use of complex bilinear resolvents."}