Nonlinear Coupled Physical Tomography, Multi-Rate Zero-Delay Feedback Phase-Locked Tracking, and Collective Cochlear Control on a 7-Dimensional Product Manifold: The Singularity Taxonomy38 v3.5 Architecture

Abstract

Whereas the Taxonomy38 v2.5 architecture (Taxonomy38 v2.5 Research Paper) established real-time closed-loop physical tomography using first-order idealized wave equations (1-D lossless strings, uncoupled single-membrane Bessel drum heads, single-pole glottal spectral tilt, and pairwise per-track spectral masking), real-world studio recordings routinely violate those linear, single-degree-of-freedom assumptions. When subjected to polyphonic guitar chords, high-sample-rate (96 kHz / 192 kHz) session clocks, single-headed concert toms, simultaneous kick-and-snare backbeats, soprano formant tuning ($f_0 \approx F_1$), and unclipped analog square-wave synthesizers, naive third-generation (v3.0) nonlinear physical estimators suffer from eighteen catastrophic mathematical and musical failure modes—including $k^2$-amplified polyphonic chord regression blowups, $+19.9\text{ dB}$ Teager-Kaiser multi-rate gain drift, $\cot^2(\omega_0 t)$ quadrature division singularities, unstable right-half-plane all-pass poles ($|z_p| > 1$), and $M$-fold midrange collapse across multi-track sessions.

This monograph presents the complete mathematical theory, Red-Team counterexample proofs, and real-time C++20 implementation of Singularity Taxonomy38 v3.5 (Taxonomy38V35Engine.h and TeamLinkIPC.h), deployed across the seven-plugin Singularity Audio Labs studio suite (Singularity GravityComp, Singularity HorizonEQ, Singularity ContourBand, Singularity PrismVerb, Singularity PulsarDelay, Singularity IronBus, and Singularity MasterDesk) linked via ⚡ TEAM LINK and 🥁 DRUM LINK. We state and prove thirteen foundational theorems across four physical instrument streams and the cross-track control manifold:

1. Stream A (#0..#10 Guitars & Plucked Strings): Fourth-order Euler–Bernoulli dispersion tomography regularized via a low-partial median anchor and Tukey's bisquare redescending M-estimator (EulerBernoulliStringEstimatorV35); pitch-covariant, sample-rate-invariant $\sin^{-2}(\Omega_c)$ normalized Zero-Delay Feedback Teager-Kaiser Energy Operators (AdaptiveZdfTkeoTrackerV35); quiescent-centered (D_0 = 1.0) Gray–Markel unitary lattice all-pass loudspeaker Doppler de-modulation (SpeakerDopplerDeModulatorV35); Cocked-Wah-shielded Bessel cabinet tomography (SpeakerCabinetTomographerV35); Helmholtz air-tail-gated 2-DOF acoustic body anti-resonance tracking (AcousticBody2DofTrackerV35); and non-monotonic post-attack inflection tomography for tube rectifier $B+$ sag-and-bloom (AmpNonlinearStateVectorV35).
2. Stream B (#17..#23 Drums & 🥁 DRUM LINK): Coupled 3-DOF batter–cavity–resonant waveguide splitting with a Concert-Tom Doublet Existence Gate (CoupledDoubletDrumWaveguideV35); pre-primed 3-epoch (28 / 53 / 78 ms) Rayleigh–Schrödinger split-doublet snare ring subtraction (PrePrimedSplitDoubletRingSubtractorV35); batter-TKEO-vetoed quadrature lock-in snare wire buzz suppression (SnareWireSympatheticLockInV35); Föppl–von Kármán cymbal turbulence cascade tracking with a 30 ms snare-bleed blanker (VonKarmanCymbalTurbulenceTrackerV35); and unconditionally Schur–Cohn stable Cytomic ZDF 2nd-order all-pass dispersive wavefront alignment (DispersiveWavefrontPhaseAlignerV35).
3. Stream C (#11..#16 Bass & #24..#29 Vocals): A 3-state Pythagorean quadrature-invariant ZDF Phase-Locked Loop with a 0.5 f_0 subharmonic escape probe (QuadratureInvariantZdfPllBassTrackerV35); fret-invariant, string-covariant magnetic pickup comb tomography (BassPickupCombTomographerV35); soprano-safe Fant–Hanson glottal-epilarynx deconvolution (GlottalEpilarynxDeconvolverV35); and plosive-immune 3-pulse vocal fry, cepstral nasal dipole, and vibrato-filtered diphthong trajectory tracking (VocalFryNasalDiphthongTrackerV35).
4. Stream D (#30..#37 Keys/Synths/Brass/Master) & Core Architecture: An exact $O(12)$ collective-power Zwicker–Moore–Glasberg cochlear field with Lyapunov-stable Fractional Masker Responsibility Partitioning (SharedCochlearFieldV35); a 7-D Product Manifold $\mathcal{M}_{\text{Spec}}^{(4\text{D})} \times \mathcal{M}_{\text{Width}}^{(2\text{D})} \times \mathcal{M}_{\text{Groove}}^{(1\text{D})}$ with orthonormal 3rd/4th-order Legendre basis polynomials and coincidence-gated micro-groove alignment (ProductManifoldStateV35); square-wave-immune biphase clipping tomography (PreDistortionTomographyV35) paired with a $+4.5\text{ dB}$-clamped Cubic Hermite transient de-clipper (ActiveTransientDeClipperV35); and Rhodes Biot–Savart odd-harmonic sieving, polarity-invariant Brass Burgers shockwave tracking, and DAW loop-safe structural Kalman pre-charging (RhodesBiotSavartSymmetryTrackerV35, BrassBurgersShockwaveTrackerV35, TransportAdaptiveStructuralKalmanV35).

1. Architectural Evolution: From Taxonomy38 v2.5 to v3.5

1.1 The Physical Boundary of First-Order Linear Tomography (v2.5)

In Taxonomy38 v2.5, each of the 38 instrument profiles in Taxonomy38.h was governed by a fast/slow dual-ballistic 12-band ERB tomography engine coupled to first-order physical trackers. While v2.5 eliminated the classic failure modes of conventional FFT/YIN/LPC plugins, four physical approximations remained:

1. 1-D Idealized Wave Equations vs. 3-D Coupled Dispersive Mechanics: v2.5 modeled plucked strings via the lossless 1-D wave equation $f_k = k f_0$ and drums via uncoupled single-membrane Bessel roots $j_{mn}$. In physical reality, a wound guitar or bass string is a 4th-order Euler–Bernoulli beam whose bending stiffness $EI$ stretches upper partials quadratically ($f_k = k f_0 \sqrt{1 + B k^2}$), while a two-headed drum shell is a 3-DOF acoustic cavity that splits the fundamental $(0,1)$ mode into an in-phase monopole $(\hat{f}_{0,+})$ and an out-of-phase phase-canceling dipole $(\hat{f}_{0,-})$.
2. Fixed-Frequency / Sample-Rate-Dependent Sub-Band Operators: v2.5 tracked pick and pluck transients using fixed band splits (90–220 Hz and 2.5–6.5 kHz). When a guitarist plays up the neck (A4 = 440 Hz), the fundamental leaves the fixed low band altogether; moreover, unnormalized discrete Teager-Kaiser Energy Operators ($\Psi_{\text{TK}}[y[n]] = y^2[n] - y[n-1]y[n+1]$) scale with the square of the discrete normalized frequency $\sin^2(\Omega_c)$, introducing a $-12.04\text{ dB}$ sensitivity collapse when switching a DAW project from 48 kHz to 192 kHz.
3. Source-Filter Commutativity Breakdown in High-Pitch Singing and Bass Subharmonics: v2.5 estimated glottal open quotient $O_q$ from $H_1 - H_2$ corrected by a 2-pole LPC formant envelope. When a soprano or rock tenor belts above C5 ($f_0 \ge 523\text{ Hz}$), the fundamental $f_0$ approaches or crosses the first vocal-tract formant $F_1$, violating source-filter separability and causing inverse-formant correction to over-subtract by up to $+29.1\text{ dB}$.
4. Pairwise Linear Masking vs. Nonlinear Collective Basilar-Membrane Mechanics: v2.5 computed cross-track masking over a flat 4-band bus vector. On the human basilar membrane, upward masking spread (Zwicker's level-dependent slope $S_u(L)$) is a nonlinear function of the collective acoustic power $L_{\text{total}}[b]$ exciting each cochlear filter, and applying un-partitioned adjoint EQ cuts across $M$ tracks causes an $M$-fold midrange collapse.

1.2 The v3.0 $\to$ v3.5 Red-Team Hardening Methodology

During the development of the third-generation architecture, our four specialized R&D teams (taxonomy38_v3_rd_proposals_report.md) formulated candidate v3.0 nonlinear partial differential equation (PDE) solvers. Each candidate was then subjected to an adversarial Red-Team Mathematical Audit (taxonomy38_v3_redteam_consensus_report.md) across four sample rates (44.1, 48, 96, and 192 kHz), polyphonic chord voicings, extreme tempos (220 BPM), and hostile boundary conditions. Every one of the eighteen discovered v3.0 vulnerabilities was resolved by an exact closed-form mathematical safeguard in Taxonomy38 v3.5 (Taxonomy38V35Engine.h).

2. Stream A (#0..#10 Guitars & Plucked Strings): Dispersive String Beams, Pitch-Covariant ZDF-TKEO, Loudspeaker Doppler De-Modulation, and 3-D Amp State Tomography

2.1 Theorem 2.1 — Euler–Bernoulli String Dispersion & Tukey Biweight Chord-Outlier Immunity

Consider a wound cylindrical or hex-core steel guitar string of linear mass density $\rho A$, speaking length $L_{\text{eff}}(n_{\text{fret}}) = L_0 2^{-n_{\text{fret}}/12}$, static tension $T_0$, Young's modulus $E$, and cross-sectional area moment of inertia $I = \frac{\pi r^4}{4}$. Transverse displacement $u(x, t)$ satisfies the fourth-order Euler–Bernoulli dispersive wave equation:

Applying hinged boundary conditions $u(0,t) = u(L_{\text{eff}},t) = u_{xx}(0,t) = u_{xx}(L_{\text{eff}},t) = 0$ yields the eigenmode spatial wavenumbers $q_k = \frac{k\pi}{L_{\text{eff}}}$ and the dispersive eigenfrequencies:

where the dimensionless inharmonicity coefficient $B$ scales inversely with the square of the speaking length:

Define the normalized squared partial deviation $y_k \triangleq \left(\frac{\hat{f}_k}{k \hat{f}_0}\right)^2 - 1$. Under a single-string monophonic excitation, $y_k = B k^2$. A naive v3.0 Weighted Least-Squares (WLS) estimator minimizes $\sum_{k=2}^{10} w_k (y_k - B k^2)^2$, yielding:

Proposition 2.1 (Polyphonic Chord Leverage Explosion): Suppose a guitarist strums a standard open major triad (containing a major third at $\frac{5}{4} f_0$) or fourth dyad ($\frac{4}{3} f_0$) with true open-string dispersion $B_{\text{open}} = 0.00014$. Within the search window around harmonic $k = 5, 7, 9$, an adjacent string's harmonic lands at $\hat{f}_k = k f_0 (1 + \delta_{\text{chord}})$ with $\delta_{\text{chord}} \in [0.045, 0.065]$. Then the unweighted $k^2 y_k$ leverage at $k = 9$ ($k^2 = 81$) inflates $\hat{B}_{\text{WLS}}^{\text{v3.0}}$ to $> 0.00145$, exceeding the high-fret wound-string threshold $B_{\text{thresh}} = 0.0012$ and falsely triggering a $-3.5\text{ dB}$ mud cut on clean rhythm chords.

Proof. At $k = 9$ with $\delta_{\text{chord}} = 0.055$, $y_9 = (1.055)^2 - 1 = 0.1130$, so the local estimate $b_9 = y_9 / 81 = 0.001395 \approx 10 B_{\text{open}}$. Because the least-squares normal equation weights each local estimate $b_k = y_k / k^2$ by $w_k k^4$, the single partial $k = 9$ carries weight $9^4 = 6561$ compared to $2^4 = 16$ at $k = 2$ (a $410:1$ leverage ratio), pulling $\hat{B}_{\text{WLS}}^{\text{v3.0}}$ above $0.0012$. $\blacksquare$

Implemented in EulerBernoulliStringEstimatorV35, we reformulate dispersion estimation in the normalized per-partial coordinate $b_k \triangleq \frac{y_k}{k^2}$:

1. Physical Feasibility & Chord-Interval Rejection Pre-Sieve ($k \in \{2, 3, 4, 5\}$): Since physical guitar string stiffness satisfies $B \in [0, 0.0040]$, even at $k = 5$ the maximum physical frequency stretch is $\sqrt{1 + 25(0.0040)} - 1 = +4.88\%$, and for $k \in \{2,3,4\}$ it is $\le +3.1\%$, whereas major/minor triad and fourth cross-harmonics deviate by $\ge 4.5\%$. We extract the low-partial candidate set $\mathcal{B}_{\text{low}} = \{ b_k : k \in \{2,3,4,5\},\; b_k \in [-0.0002, 0.0040] \}$ and compute the exact order-statistic median anchor:

1. Tukey's Bisquare (Biweight) Redescending Kernel: With scale parameter $c_B = 0.00035$, every upper partial $k \in \{2, \dots, 10\}$ is weighted by the redescending kernel:

Any polyphonic chord partial with $|b_k - b_{\text{anchor}}| \ge 0.00035$ receives exact zero weight ($w_k^{\text{Tukey}} \equiv 0$), completely neutralizing its $k^4 = 6561$ leverage:

1. Monophonic Inlier Gate (highFretSoloProb01): High-fret wound-string solo actions (180–320 Hz mud relief and 1.8–3.2 kHz harmonic re-voicing) are gated by the joint probability:

2.2 Theorem 2.2 — Multi-Rate & Pitch-Covariant Normalized ZDF-SVF Teager-Kaiser Operator

For a discrete sub-band signal $y[n] = A \cos(\Omega_c n + \phi)$ at normalized angular frequency $\Omega_c = \frac{2\pi f_c}{f_s}$, the discrete Teager-Kaiser Energy Operator is:

Under the trapezoidal bilinear transform $g = \tan\!\left(\frac{\pi f_c}{f_s}\right) = \tan\!\left(\frac{\Omega_c}{2}\right)$, trigonometric double-angle identity yields the exact algebraic relation:

Proposition 2.2 (Pitch-Retuning and Multi-Rate TKEO Bias):

1. At $f_s = 48\text{ kHz}$, retuning a low-band ZDF-SVF tracker from $f_c = 55\text{ Hz}$ ($g = 0.00360$) to $f_c = 440\text{ Hz}$ ($g = 0.02881$) changes $\sin^2(\Omega_c)$ by a factor of $(440/55)^2 \approx 63.8$, introducing an artificial $+18.05\text{ dB}$ (and up to $+19.9\text{ dB}$ across 45–440 Hz) TKEO inflation for identical signal amplitude $A$.
2. At fixed frequency $f_c = 118\text{ Hz}$, switching the DAW host from $f_s = 48\text{ kHz}$ to $f_s = 192\text{ kHz}$ reduces $\sin^2(\Omega_c)$ by $(48/192)^2 = 1/16$, causing a $-12.04\text{ dB}$ TKEO collapse.

Implemented in AdaptiveZdfTkeoTrackerV35, each band-pass Cytomic ZDF State-Variable Filter (ZdfBandPassTap) smooths its trapezoidal integrator gain $g[n]$ in the $g$-domain with a 10 ms time constant ($\alpha_g = 1 - e^{-1/(0.010 f_s)}$) to prevent parametric derivative spikes (ic1eq, ic2eq kick) during rapid legato pitch glides, and multiplies the instantaneous TKEO by the exact inverse discrete frequency factor:

Consequently, for any constant-amplitude sinusoid $A$, $\widetilde{\Psi}_{\text{TK}}[y[n]] = A^2$ with $0.00\text{ dB}$ theoretical error across all pitches $f_c \in [45\text{ Hz}, 12\text{ kHz}]$ and all sample rates $f_s \in \{44.1, 48, 96, 192\}\text{ kHz}$. Furthermore, dual-tap pick scrape detection (lowTap at $\operatorname{clamp}(1.15 \hat{f}_0, 75, 420)\text{ Hz}$, highTap at $\operatorname{clamp}(\max(2800, 5.5\hat{f}_0), 2500, 6800)\text{ Hz}$) is gated by a broadband -42 dBFS noise-floor threshold and palm-mute anti-correlation veto $(1 - 0.85 \mathcal{M}_{\text{palm}})$.

2.3 Theorem 2.3 — Loudspeaker Cone Doppler De-Modulation via Quiescent-Centered Gray–Markel Unitary Lattice All-Pass

When a 12-inch guitar cabinet cone undergoes large low-frequency excursion $\xi(t) = X_0 \sin(\omega_{\text{low}} t)$ ($X_0 \approx 3.5\text{ mm}$ at 80–120 Hz palm-mute chugs) while simultaneously radiating a high-frequency partial $s_h(t) = A_h \cos(\omega_h t)$, the acoustic wave emitted from the moving boundary $x = \xi(t)$ experiences a time-varying propagation delay $\tau_D(t) = -\frac{\xi(t)}{c_0}$ ($c_0 = 343\text{ m/s}$). By the Jacobi–Anger expansion, the radiated pressure splits into an infinite comb of anharmonic Bessel frequency-modulation (FM) sidebands:

At $f_h = 4\text{ kHz}$ and $X_0 = 3.5\text{ mm}$, $\beta_{\text{FM}} = 0.256\text{ rad}$, generating first-order intermodulation sidebands $J_{\pm 1}(\beta_{\text{FM}}) \approx -17.9\text{ dBc}$ at $f_h \pm f_{\text{low}}$.

In v3.0, $\hat{\xi}[n]$ was integrated via a leaky integrator and used to drive a Thiran first-order all-pass delay $D[n] = -f_s \frac{\hat{\xi}[n]}{c_0}$ with coefficient $a_1[n] = \frac{1 - D[n]}{1 + D[n]}$.

1. Sign-Crossing & Integer-Boundary Discontinuity at 192 kHz: Because $\hat{\xi}[n]$ oscillates symmetrically around $0$, $D[n]$ crosses $0$ twice every cycle (160 times/sec at 80 Hz). At $f_s = 192\text{ kHz}$ with $X_0 = 3.5\text{ mm}$, $|D_{\max}| = 192000 \times \frac{0.0035}{343} = 1.96\text{ samples}$. Splitting $D[n] = \lfloor D[n] \rfloor + d_{\text{frac}}[n]$ forces $d_{\text{frac}}[n]$ to jump discontinuously from $0.999 \to 0.000$ at the integer boundary $D = 1.0$, causing a_1[n] to jump from $0.0 \to 1.0$ in a single sample and injecting audible clicks into the > 1.2 kHz upper band.
2. Time-Varying Direct-Form State Injection: Modulating a Direct-Form I/II all-pass filter's coefficient $a_1[n]$ without energy-preserving lattice rotation injects spurious parasitic energy $\propto \frac{da_1}{dt}$ into the internal delay state.

Implemented in SpeakerDopplerDeModulatorV35:

1. DC-Immune Mechanical Suspension Model: Low-frequency input is high-pass filtered at 42 Hz (blocking asymmetrical rectified guitar DC offsets) and passed into a 2nd-order mechanical loudspeaker suspension ZDF-SVF tuned to the cabinet resonance $f_{\text{res}} = 85\text{ Hz}$ ($Q_{\text{mech}} = 1.15$), whose low-pass displacement output $v_2[n]$ is strictly zero-mean.
2. Quiescent Delay Anchor $D_0 \equiv 1.000\text{ sample}$: Both the low-band path (lowDelay1) and the high-band path are anchored at a constant quiescent delay of 1.000 sample. The fractional Doppler modulation $\Delta_d[n]$ is softly bounded inside $[-0.42, +0.42]\text{ samples}$:

Because $D[n]$ remains strictly inside the single smooth interval $[0.58, 1.42]$, it never crosses zero or an integer boundary at any sample rate (44.1–192 kHz).

1. Two-Multiplier Passive Gray–Markel Lattice All-Pass: With reflection coefficient $k_{\text{ap}}[n] = \frac{1 - D[n]}{1 + D[n]} = \frac{\Delta_d[n]}{2 - \Delta_d[n]} \in [-0.174, +0.266]$ (so $\sqrt{1 - k_{\text{ap}}^2} \in [0.964, 1.0]$ is near unity), the high-band (> 1.2 kHz) signal $x_{\text{high}}[n]$ is phase-restored via the time-varying two-multiplier passive Gray–Markel lattice:

Corollary 2.3 (Exact Wire Identity at Rest): When $\text{maxExcursionMm} = 0$, $\Delta_d[n] \equiv 0 \implies k_{\text{ap}}[n] \equiv 0$, so $y_{\text{high}}[n] = x_{\text{high}}[n-1]$ and $y_{\text{low}}[n] = x_{\text{low}}[n-1]$. Since $x_{\text{low}}[n] + x_{\text{high}}[n] \equiv x[n]$, the total output satisfies $y[n] \equiv x[n-1]$ to floating-point machine precision ($< 3\times 10^{-8}$).

2.4 Cabinet Breakup Shielding, 2-DOF Acoustic Body Anti-Resonance, and 3-D Amp State Vectors

1. Cabinet Breakup vs. Cocked-Wah Shield (SpeakerCabinetTomographerV35): Tracks the (1,1) dust-cap mode (1.8–2.6 kHz), (0,2) axisymmetric cone breakup (3.2–4.5 kHz), and (0,3) surround edge mode (4.8–6.2 kHz). To prevent false attenuation on intentional cocked-wah (Dunlop CryBaby $Q \in [4, 7]$) lead tones, we evaluate the Cocked-Wah Single-Peak Shield $\mathcal{S}_{\text{wah}} \in [0, 1]$, which approaches $1.0$ whenever (1,1) prominence is high (> 8 dB) while simultaneous (0,2) and (0,3) modes are absent (< 3.5 dB), reducing the (1,1) cut by $85\%$ to preserve the vocal wah formant.
2. 2-DOF Acoustic Body Anti-Resonance & Helmholtz Air-Tail Piezo Veto (AcousticBody2DofTrackerV35): Models the coupled top-plate $(m_p, k_p)$ and Helmholtz air cavity $(m_a, k_a)$ splits $\hat{f}_- \in [85, 115]\text{ Hz}$ and $\hat{f}_+ \in [165, 215]\text{ Hz}$, locating the exact structural anti-resonance null $\hat{f}_{\text{anti}} = \sqrt{(\hat{f}_-^2 + \hat{f}_+^2)/2} \in [125, 155]\text{ Hz}$ where positive EQ boosts are masked by $85\%$ (computeAntiResonanceBoostMask $= 0.15$). To prevent Flamenco golpe or percussive fingerstyle thumb slaps on a mic'd acoustic guitar from falsely triggering the undersaddle piezo de-quacker, piezoPickupProb01 requires both high zero-lag TKEO covariance and a severe Helmholtz Air-Tail Sustain Deficit ($R_{A_0/\text{mid}}^{\text{tail}} < -10\text{ dB}$ or $T_{60}(A_0) < 45\text{ ms}$ during 25–120 ms post-attack).
3. 3-D Amp State Vector & Non-Monotonic $B+$ Sag Inflection (AmpNonlinearStateVectorV35): Replaces scalar preCompression01 with $\boldsymbol{\Sigma}_{\text{amp}} = (\Psi_{\text{clip}}, \Lambda_{\text{sag}}^{\text{v3.5}}, \Pi_{\text{parity}})$. Rather than measuring monotonic post-onset decay (which fires falsely on every naturally decaying passive string), $\Lambda_{\text{sag}}^{\text{v3.5}}$ requires a Non-Monotonic Post-Attack Inflection ("Dip-Then-Swell") signature: an initial rectifier sag valley $E_{\min}^{\text{sag}}$ inside $t \in [10, 45]\text{ ms}$ followed by a genuine capacitor recovery swell $E_{\max}^{\text{swell}} - E_{\min}^{\text{sag}} > +0.8\text{ dB}$ inside $t \in [50, 180]\text{ ms}$.

3. Stream B (#17..#23 Drums & 🥁 DRUM LINK): Coupled 3-DOF Shell Waveguides, Pre-Primed Split-Doublet Snare Ring Subtraction, Quadrature Wire Lock-In, and Schur–Cohn All-Pass Alignment

3.1 Theorem 3.1 — Coupled 3-DOF Drum Shell Waveguide & Concert-Tom Doublet Existence Gate

In a two-headed cylindrical drum shell (kick or tom), the batter membrane displacement $w_b(r,\theta,t)$ and resonant membrane displacement $w_r(r,\theta,t)$ are coupled through the enclosed air cavity compliance $C_{\text{cav}} = \frac{V_{\text{shell}}}{\rho_0 c_0^2}$. The axisymmetric $(0,1)$ membrane mode splits into two normal modes:

In addition, finite elastic deflection stretches the batter head by tension surge $\Delta T(t) \propto \iint |\nabla w_b|^2\,dA$, producing the early pitch-bend glide:

In v3.0, a 2-real-pole Prony/Hankel matrix pencil was fit over 6–18 ms and unconditionally subtracted a $-18\text{ dB}$ notch at the second root $\hat{f}_{0,-}$.

Proposition 3.1 (Rank-1 Hankel Singularity on Single-Headed Drums): On a single-headed concert tom, heavily pillowed kick drum, or synthesized 808 kick, the resonant head is absent or overdamped ($E_- \to 0$). The $2\times 2$ Hankel matrix $\mathbf{G}_0$ becomes rank-1 ($\det(\mathbf{G}_0) \to 0$), forcing the spurious second root $\hat{f}_{0,-}$ to split across the true fundamental $\hat{f}_{0,+}$ or lock onto noise, causing the unconditional $-18\text{ dB}$ anti-slosh siphon to hollow out the drum's primary fundamental punch.

Implemented in CoupledDoubletDrumWaveguideV35:

1. Orthogonal Dual-SVF Mode Splitter (7–18 ms): Three zero-crossing/quadrature frequency epochs $f_1, f_2, f_3$ at $t = 7, 12, 17\text{ ms}$ solve the closed-form 3-point exponential pitch-bend glide:

1. Doublet Existence Gate $\Gamma_{\text{doublet}} \in [0, 1]$: The out-of-phase dipole candidate $\hat{f}_{0,-}$ is validated by requiring both significant relative energy ($E_- / E_+ > -18\text{ dB}$) and distinct frequency separation ($|\hat{f}_{0,-} - \hat{f}_{0,+}| > 0.14\hat{f}_{0,+}$):

On a single-headed concert tom or pillowed kick, $\Gamma_{\text{doublet}} = 0.0001 \approx 0$, guaranteeing $0.000\text{ dB}$ fundamental loss.

1. Pre-Primed Siphon & 220 BPM Re-Hit Abort: The Cytomic ZDF band-pass siphon filter (siphonBp_) is continuously pre-primed from $t = 0\text{ ms}$ with zero gain (cutLinear = 0) so its internal states ic1eq, ic2eq are already phase-locked when the smooth cosine ramp engages at $t = 18\text{ ms}$. If a secondary TKEO spike occurs inside 5–18 ms (220 BPM double-kick blast or flam), rehitAborted_ immediately aborts estimation and retains the warm Bayesian prior.

3.2 Theorem 3.2 — Rayleigh–Schrödinger Split-Doublet Snare Ring Subtraction via 3-Epoch Beat-Curvature Tomography

Under uniform tension, a circular snare batter head possesses degenerate non-axisymmetric modes $\cos(m\theta)$ and $\sin(m\theta)$ ($m \ge 1$). Small lug-to-lug tension anisotropy $\epsilon(\theta) = \epsilon_m \cos(2m\theta)$ breaks rotational symmetry. By first-order Rayleigh–Schrödinger perturbation theory, the degenerate $(m,1)$ overtone splits into an orthogonal frequency doublet:

whose superposition produces the characteristic modulated "wah-wah" metallic snare ring envelope $A(t) = 2 A_0 e^{-t/\tau_m} \left|\cos(\pi \,\delta f_m\, t)\right|$.

1. Single-SVF Cancellation Ceiling: A single resonator tuned to the mean frequency $\bar{f}_m$ experiences a growing phase error $\Delta\phi(t) = \pm \pi\,\delta f_m\, t$ against the split doublet modes, limiting average energy cancellation over the decay tail to at most $-6.03\text{ dB}$.
2. Gabor Uncertainty Failure of 4–14 ms Windows: Observing a $\delta f_m = 6\text{ Hz}$ beat envelope (beat period $T_{\text{beat}} = \frac{1}{\delta f_m} = 166.7\text{ ms}$) over a 10 ms window (4–14 ms) covers only $6\%$ of a single beat cycle, where broadband snare-wire noise completely masks the $\cos(\pi\,\delta f_m\,t)$ curvature.

Implemented in PrePrimedSplitDoubletRingSubtractorV35:

1. Immediate Crack Protection + Pre-Primed Dual SVFs (modeASvf_, modeBSvf_): Both split-doublet ZDF band-pass filters step every sample from $t = 0\text{ ms}$ using the Bayesian prior $(\bar{f}_m, \delta\hat{f}_m)$, while the subtraction envelope $g_{\text{sub}}(t)$ is strictly zero for $t \le 11\text{ ms}$ (preserving $100\%$ of the 0–11 ms snare stick crack) and ramps smoothly via a raised cosine over 11–18.5 ms.
2. Decay-Normalized 3-Epoch Beat Ratio ($t_1 = 28\text{ ms}, t_2 = 53\text{ ms}, t_3 = 78\text{ ms}$, $\Delta t = 25\text{ ms}$): Let $A(t) = \sqrt{E(t)} = A_0 e^{-t/\tau_m} \left|\cos(\pi\,\delta f_m\,t)\right|$ denote the narrowband RMS amplitude envelope excited in-phase at $t=0$ ($t_1 \approx \Delta t, t_2 \approx 2\Delta t, t_3 \approx 3\Delta t$, with $\alpha \triangleq \pi\,\delta f_m\,\Delta t$). Taking the geometric-to-midpoint ratio across three equally spaced tail epochs analytically cancels both $A_0$ and the exponential decay $e^{-t/\tau_m}$, while the quadratic phase curvature at $t_2 \approx 2\Delta t$ cancels against $\sqrt{\cos(\alpha)\cos(3\alpha)}$ to fourth order in $\alpha$:

Inverting this invariant yields the exact closed-form doublet split:

achieving $17.6\text{ dB}$ split-doublet ring cancellation in verification testing (Taxonomy38V35AnatomyVerification.cpp) while introducing $0.00000$ error on the 0–11 ms transient crack.

3.3 Theorem 3.3 — Sympathetic Snare-Wire Quadrature Lock-In with Batter-TKEO Veto & Föppl–von Kármán Cymbal Turbulence

1. Sympathetic Snare-Wire Lock-In (SnareWireSympatheticLockInV35): When the kick drum (#17) radiates sub-bass pressure at phase $\phi_{\text{kick}}(t)$, the snare bottom head (#18) deflects and drives the snare wires against the resonant Mylar head twice per kick cycle (full-wave mechanical rectification at $2\phi_{\text{kick}}(t)$). Using the kick fundamental quadrature pair $(\cos\phi_{\text{kick}}, \sin\phi_{\text{kick}})$ broadcast over 🥁 DRUM LINK, we synthesize the second-harmonic references via Chebyshev identities:

and correlate them against the AC-coupled snare-wire high-band envelope $\tilde{e}_{\text{wire}}(t)$.

• Simultaneous Kick+Snare Backbeat & Ghost-Note Immunity: To prevent a 60 ms lock-in integrator from ducking the snare's own wire snap on Beats 2 & 4 (when kick and snare hit together) or ghost notes, any snare batter TKEO impulse (snareBatterStrikeImpulse01 > 0.15) instantaneously dumps the quadrature accumulators $(I_{2\phi}, Q_{2\phi}) \to (0, 0)$ and arms a 45 ms direct-strike veto hold, guaranteeing $0.000\text{ dB}$ wire snap loss on backbeats while suppressing pure sympathetic kick buzz by $-14.0\text{ dB}$.

1. Föppl–von Kármán Cymbal Turbulence & 30 ms Snare-Bleed Blanker (VonKarmanCymbalTurbulenceTrackerV35): Models thin bronze shells via the nonlinear Föppl–von Kármán plate equations $\rho h w_{tt} + D\nabla^4 w = [w, \Phi]$, tracking the Kolmogorov–Zakharov wave-turbulence energy flux $\Phi_{\text{KZ}}$ from the 2.8–5.4 kHz clang bottleneck into the 8–16 kHz shimmer cascade. When snare bleed hits the Overhead/Room mics (snareCoincidence01 > 0.35), a 30 ms Snare-Bleed Blanker reduces dynamicClangCutDb by $90\%$, preserving $100\%$ of the overhead snare body.

3.4 Theorem 3.4 — Unconditionally Schur–Cohn Stable Cytomic ZDF 2nd-Order All-Pass Dispersive Wavefront Alignment

In v3.0, dispersive phase misalignment between a close mic and an overhead/ported kick mic was corrected by solving an unconstrained $2\times 2$ Cramer linear system for the Direct-Form 2nd-order all-pass coefficients $(a_1, a_2)$ to interpolate arbitrary phase targets $(\phi_1, \phi_2)$ at $(\Omega_1, \Omega_2)$.

Proposition 3.4 (Right-Half-Plane Pole Explosion under Positive Phase Slope): A causal, stable, nontrivial real all-pass filter satisfies Foster's Reactance Theorem: its phase $\phi_{\text{AP}}(\Omega)$ is strictly monotonically decreasing ($\frac{d\phi_{\text{AP}}}{d\Omega} = -\tau_g(\Omega) < 0$). When a ported bass-reflex kick drum or room reflection exhibits a positive phase slope ($\phi_2 = +1.45\text{ rad} > \phi_1 = -0.35\text{ rad}$), the unconstrained Cramer solution yields $a_1 = -1.9985, a_2 = +1.0043$, placing complex-conjugate poles at $|z_p| = \sqrt{a_2} = 1.00214 > 1$ outside the unit circle and causing exponential IIR oscillation ($+267\text{ dB}$ in $0.3\text{ s}$).

Implemented in DispersiveWavefrontPhaseAlignerV35:

1. Physical Positive Group-Delay Projection: Before parameterizing the filter, target phases are wrapped to $[-\pi, 0]$ and projected onto the physically realizable negative-slope cone $\phi_2 \le \phi_1 - 0.10\text{ rad}$.
2. Left-Half-Plane Parameterization ($g > 0, k > 0$): Rather than solving for unconstrained polynomial coefficients $(a_1, a_2)$, we parameterize the 2nd-order all-pass directly in the continuous left-half $s$-plane via strictly positive pole frequency $f_p \in [35, 240]\text{ Hz}$ and strictly positive damping $k = \frac{1}{Q_p} \in [0.35, 2.0]$:

Under the trapezoidal Zero-Delay Feedback (ZDF) bilinear mapping $g = \tan(\pi f_p / f_s) > 0$, the discrete poles are the roots of $D(z) = (1 + g(g+k)) + 2(g^2 - 1)z^{-1} + (1 + g(g-k))z^{-2}$.

Theorem 3.4 (Unconditional Schur–Cohn & Unit-Gain Stability): For every $f_p \in (0, f_s/2)$ ($g > 0$) and every $k > 0$, the Cytomic ZDF 2nd-order all-pass state realization

satisfies the Schur–Cohn stability criteria $|c_2| < 1$ and $|c_1| < 1 + c_2$ unconditionally at all sample rates (44.1–192 kHz) and under continuous time-varying modulation of $(f_p, k)$, with exact unit magnitude $|H_{\text{AP}}(e^{j\Omega})| \equiv 1.00000$.

Proof. The normalized denominator coefficients are $c_2 = \frac{1 + g^2 - gk}{1 + g^2 + gk}$ and $c_1 = \frac{2(g^2 - 1)}{1 + g^2 + gk}$. Since $g > 0$ and $k > 0$, $gk > 0 \implies |1 + g^2 - gk| < 1 + g^2 + gk \implies |c_2| < 1$. Furthermore, $1 + c_2 - |c_1| = \frac{2(1 + g^2) - 2|g^2 - 1|}{1 + g^2 + gk} = \frac{4\min(1, g^2)}{1 + g^2 + gk} > 0$, proving $|c_1| < 1 + c_2$ strictly. Thus all poles lie strictly inside the unit circle ($|z_p| < 1$) for all $g > 0, k > 0$. $\blacksquare$

4. Stream C (#11..#16 Bass & #24..#29 Vocals): Pythagorean 3-State ZDF-PLL Bass Tracking, Fret-Invariant Pickup Comb Tomography, and Soprano-Safe Glottal-Epilarynx Deconvolution

4.1 Theorem 4.1 — Pythagorean 3-State Quadrature-Invariant ZDF-PLL Bass Tracker with 0.5 f_0 Subharmonic Escape

To track low-B (B0 = 30.87 Hz) and drop-tuned bass fundamentals within a quarter-cycle (~8.1 ms), Taxonomy38 v2.5 used the SVF high-pass-to-low-pass energy ratio $\sqrt{E_{\text{HP}} / E_{\text{LP}}}$. Because $|H_{\text{HP}}(m\omega_0)| / |H_{\text{LP}}(m\omega_0)| = m^2 (\omega_0/\omega_c)^2$, the $m$-th harmonic's squared energy is weighted by $m^4$ ($2^4 = 16\times$ for the 2nd harmonic, $3^4 = 81\times$ for the 3rd harmonic), pulling the pitch tracker an octave high on saturated DI bass.

To avoid $H_{\text{HP}}$, v3.0 attempted to use the band-pass-to-low-pass ratio $v_{\text{BP}}^2 / v_{\text{LP}}^2$ over a 4-sample moving average.

Proposition 4.1 (Quadrature $\cot^2(\omega_0 t)$ Blowup at High Sample Rates): In any Cytomic ZDF-SVF, the low-pass output $v_2(t)$ is the exact time integral of the band-pass output $v_1(t)$: $\frac{dv_2}{dt} = \omega_c v_1(t)$. For a sinusoid $x(t) = A\sin(\omega_0 t)$, $v_1(t)$ and $v_2(t)$ are in exact $90^\circ$ phase quadrature:

Consequently, their instantaneous squared ratio is

which oscillates between $0$ and $+\infty$ twice every cycle (61.7 times/sec at B0 = 30.87 Hz). At $f_s = 192\text{ kHz}$, a 4-sample window spans only $0.0208\text{ ms}$ ($0.23^\circ$ of phase), failing to smooth $\cot^2(\omega_0 t + \phi)$ and causing violent $2\omega_0$ pitch ripple and octave jumps.

Implemented in QuadratureInvariantZdfPllBassTrackerV35, we introduce a third cascaded ZDF integrator state $v_{\text{I2}}(t)$ by passing $v_2(t)$ through an identical 1-pole ZDF low-pass integrator with gain $g$:

Theorem 4.1 (Exact Sample-by-Sample Phase Cancellation via Pythagorean Quadrature): Let $r \triangleq \frac{\omega_0}{\omega_c}$ be the normalized frequency ratio. Then at every single sample $n$, the three states $(v_1, v_2, v_{\text{I2}})$ satisfy the exact trigonometric identity $\cos^2\theta + \sin^2\theta \equiv 1$:

Taking the ratio of these two quadrature invariants eliminates time dependence $\theta = \omega_0 t + \phi$ identically at every sample $n$ without requiring a multi-sample window:

To guarantee immunity against missing-fundamental DI/amp tones where $2f_0$ exceeds $f_0$:

1. Bounded -14 dB $2\hat{f}_0$ Harmonic Pre-Notch: Attenuates $2\hat{f}_0$ by $80\%$ (k = 0.70, gain = 0.80) before the 3-state PLL, reducing 2nd-harmonic energy by $25\times$ while preserving $20\%$ residual on true octave jumps.
2. Un-Notched 0.5 f_0 Subharmonic Escape Probe (subProbeBp_): A parallel ZDF band-pass probe runs on the raw un-notched input at $0.5\hat{f}_0$. If the PLL ever initializes on $2f_0$ (61.74 Hz), the probe sees the true fundamental at $0.5(2f_0) = f_0$ (30.87 Hz) and smoothly pulls trackedF0Hz down to the true B0 fundamental.
3. 3-FLOP Maclaurin/Padé Pre-Warping (fastTanPadeBass): Replaces std::tan in the inner loop with the 5th-order Maclaurin/Padé polynomial $x(1 + \frac{1}{3}x^2 + \frac{2}{15}x^4)$ (< 3e-9% error across 18–400 Hz).

4.2 Theorem 4.2 — Fret-Invariant, String-Covariant Magnetic Bass Pickup Comb Tomography

Consider a magnetic bass pickup located at a fixed physical distance $x_p$ from the bridge saddle on a bass of open scale length $L_{\text{open}}$. When the player frets a note at speaking length $L(n_{\text{fret}}) \le L_{\text{open}}$, the $k$-th spatial harmonic node falls directly over the pickup aperture whenever $\sin\!\left(\frac{k\pi x_p}{L(n_{\text{fret}})}\right) = 0$, i.e., $k = \frac{m L(n_{\text{fret}})}{x_p}$.

In v3.0, the comb null frequencies were modeled as $f_{\text{null},m}^{\text{v3.0}} = \frac{m}{\beta_0}\hat{f}_0$ using a constant fractional ratio $\beta_0 = x_p / L_{\text{open}}$ multiplied by the fretted pitch $\hat{f}_0$.

Proposition 4.2 (Fret-Invariance & String-Covariance of Magnetic Pickup Nulls): On any given string $s \in \{\text{B0}, \text{E1}, \text{A1}, \text{D2}, \text{G2}\}$ with wave speed $c_s = 2 L_{\text{open}} f_{0,\text{open},s}$, the fretted fundamental is $f_0(n_{\text{fret}}) = \frac{c_s}{2 L(n_{\text{fret}})}$. Substituting $k = \frac{m L(n_{\text{fret}})}{x_p}$ into $f_{\text{null},m} = k f_0(n_{\text{fret}})$ causes the fretted speaking length $L(n_{\text{fret}})$ to cancel out identically:

Thus, the pickup comb null frequencies depend ONLY on the fixed open-string ratio $\beta_{0,\text{bridge}}$ and the active string's open fundamental $f_{0,\text{open},s}$, and are strictly invariant to fret position $n_{\text{fret}}$!

Implemented in BassPickupCombTomographerV35, we combine the tracked pitch $\hat{f}_0$ with the Euler–Bernoulli dispersion $\hat{B}_{\text{bass}} = B_{\text{open},s} (\hat{f}_0 / f_{0,\text{open},s})^2$ to infer the active string's open fundamental $\hat{f}_{0,\text{open},s} \in \{30.87, 41.20, 55.00, 73.42, 98.00\}\text{ Hz}$, masking positive EQ boosts inside $\hat{f}_{\text{null},m} = \frac{m}{\hat{\beta}_{0,\text{bridge}}}\hat{f}_{0,\text{open},s}$ (nullMask < 0.05) and steering presence boosts to the anti-node peak $\hat{f}_{\text{anti-node}} = \frac{m + 0.5}{\hat{\beta}_{0,\text{bridge}}}\hat{f}_{0,\text{open},s}$.

4.3 Theorem 4.3 — Soprano-Safe Glottal-Epilarynx Deconvolution (f_0 >= 0.75 F_1 Formant-Tuning Immunity)

By the Fant–Hanson source-filter radiation law, a single vocal-tract resonator at frequency $F_1$ with bandwidth $B_1$ amplifies a harmonic at frequency $f$ by:

In normal low-pitch speech ($f_0 \ll F_1$), subtracting $\Delta L_{F_1} = L_{F_1}(f_0) - L_{F_1}(2f_0)$ from $H_1 - H_2$ yields the formant-corrected open quotient $H_1^* - H_2^*$.

Proposition 4.3 (Soprano Formant-Tuning Over-Subtraction): When a soprano or high-belt rock vocalist sings at $f_0 = 523.25\text{ Hz}$ (C5) and tunes their first formant to match the fundamental ($F_1 \approx 530\text{ Hz}$, $B_1 = 120\text{ Hz}$), the fundamental $f_0$ sits directly on the resonant peak of $F_1$ ($L_{F_1}(f_0) = +13.1\text{ dB}$) while $2f_0 = 1046.5\text{ Hz}$ sits in the skirt ($L_{F_1}(2f_0) = -9.6\text{ dB}$). Furthermore, $\hat{F}_1$ estimated from discrete harmonic peaks locks onto $H_1$ itself. Subtracting $\Delta L_{F_1} = +22.7\text{ dB}$ (or $+29.1\text{ dB}$ at $B_1 = 60\text{ Hz}$) drives $H_1^* - H_2^*$ deeply negative ($-6.7\text{ dB}$), causing a naive deconvolver to misclassify an effortless, pure soprano head-voice note as "hyperfunctional pressed vocal glare" (pressedGlareIndex01 = 1.0) and inject mud at $f_0$.

Implemented in GlottalEpilarynxDeconvolverV35:

1. Soprano Formant-Tuning Observability Gate: We define the smooth observability weight:

Whenever $f_0 \ge 0.75\hat{F}_1$ ($\hat{F}_1 \le 1.33 f_0$), $w_{\text{H1H2}} \to 0$, smoothly handing off open-quotient estimation from $H_1^* - H_2^*$ to the Epilarynx-Excluded Liljencrants–Fant (LF) Return-Phase Spectral Tilt $\hat{O}_{q,\text{tilt}}$ measured between the 4.2–6.0 kHz skirt and the $F_1/F_2$ base (strictly excluding the 2.5–3.8 kHz epilarynx resonance so Singer's Formant ring never biases $\hat{O}_q$):

1. Orthogonal Separation of Epilarynx Ring ($\mathcal{C}_{\text{epi}}$) vs. Pressed Glare ($\mathcal{G}_{\text{pressed}}$):

• Effortless Singer's Formant (2.8–3.4 kHz): High epilarynx clustering ($\Delta F_{43} = F_4 - F_3 < 550\text{ Hz}$ and prominent 3 kHz peak) with neutral $\hat{O}_q \ge 0.42$ yields $\mathcal{C}_{\text{epi}} > 0.80$ and $\mathcal{G}_{\text{pressed}} < 0.01$, protecting the vocal ring by $80\%$.
• Pressed Laryngeal Constriction: Low $\hat{O}_q < 0.42$ without $F_3/F_4$ clustering triggers targeted 2.2–3.8 kHz dynamic de-harshing and $f_0$ warmth restoration (scaled by $w_{\text{H1H2}}$ so high sopranos are never bloated at 500–800 Hz).

4.4 Theorem 4.4 — Plosive-Immune 3-Pulse Vocal Fry, Cepstral Nasal Dipole, and Vibrato-Filtered Diphthong Tracking

Implemented in VocalFryNasalDiphthongTrackerV35:

1. Plosive-Immune 3-Pulse Vocal Fry Detector (vocalFryIndex01): A single sub-70 Hz TKEO burst fires falsely on every bilabial plosive (/p/, /b/) or mic stand bump. True creaky phonation ($f_0 \in [28, 68]\text{ Hz}$) consists of a periodic train of damped formant-ringing impulses in 300–2400 Hz whereas a plosive pop is dominated by 18–75 Hz air blast. We therefore require:

• $(1 - \mathcal{P}_{\text{pop}})$ plosive veto and $E_{300-2400\text{ Hz}} > E_{18-75\text{ Hz}} - 3\text{ dB}$, plus
• At least 3 consecutive glottal pulses (validPulseTrainCount >= 2 inter-pulse intervals) with $\text{IPI}_k \in [14.5\text{ ms}, 35.5\text{ ms}]$ and normalized jitter $\frac{|\text{IPI}_k - \text{IPI}_{k-1}|}{\text{IPI}_k} < 0.32$.

1. Pitch-Synchronous Cepstral Nasal Dipole ($\mathcal{N}_{\text{nasal}}$): Identifies velopharyngeal coupling (800–1350 Hz nasal pole paired with an adjacent 85–230 Hz oral-nasal anti-resonance zero) strictly above $2.2 f_0$ so harmonic troughs are never mistaken for nasal zeroes.
2. Vibrato-Filtered Diphthong Velocity ($\mathcal{V}_{\text{diph}}$): Passes $(z_{F_1}(t), z_{F_2}(t))$ Bark coordinates through a 45 ms 1-pole low-pass filter (~3.5 Hz cutoff) before differentiation, attenuating 5.5–7.5 Hz vocal vibrato harmonic oscillation by $> 75\%$ while preserving genuine 120–250 ms diphthong glides (/aI/, /eI/, /oU/).

5. Stream D (#30..#37 Keys/Synths/Brass/Master) & Core Architecture: Collective Cochlear Control, 7-D Product Manifold, Square-Wave-Safe De-Clipping, and Instrument PDEs

5.1 Theorem 5.1 — $O(12)$ Collective-Power Zwicker–Moore–Glasberg Cochlear Field & Lyapunov-Stable Fractional Masker Responsibility Partition

On the human basilar membrane, the upper-frequency masking spread slope $S_u$ (dB/Bark) of ERB band $b$ is a nonlinear function of the total excitation level $L[b]$ (Zwicker's upward spread of masking):

Proposition 5.1 (Pairwise Track-Splitting Superposition Error & $M$-Track Adjoint Collapse):

1. Suppose a single guitar bus at $L_{\text{total}}[b] = 75\text{ dB SPL}$ is split across $M = 8$ identical tracks of power $P_j[b] = \frac{1}{8} P_{\text{total}}[b]$ ($L_j[b] = 66\text{ dB SPL}$). Evaluating $S_u(L_j[b])$ on each track in isolation steepens the slope by $0.2 \times 9 = 1.8\text{ dB/Bark}$, under-predicting masking at $\Delta z = +4\text{ Bark}$ by $+7.22\text{ dB}$ for the exact same physical mix.
2. In an un-partitioned v3.0 adjoint broadcast, every non-Tier-1 track $j \in \{1, \dots, M\}$ applied $\Delta G_j^{\text{v3.0}}[b] = -\eta \sum_m \Phi_{\text{ZW}}(m, b, L_j[b]) D_{\text{Tier1}}[m]$—including silent tracks ($P_j[b] = 0$)—resulting in $M \times$ total attenuation and collapsing the midrange of large sessions.

Implemented in SharedCochlearFieldV35 inside TeamLinkIPC.h (kTeamLinkVersion = 11, magic 0x53474C42 'SGLB'), the ⚡ TEAM LINK shared memory bus aggregates four lock-free 12-band vectors across the session: totalLinearPower[b], nonTier1LinearPower[b], tier1MaskingDeficitDb[b], and tier1PanAzimuthRad[b].

Theorem 5.1 (Exact Superposition Invariance & $M$-Independent Lyapunov Stability): Define the collective Zwicker spread kernel $\Phi_{\text{ZW}}(m, b, L_{\text{total}}[b]) = 10^{-S(m, b, L_{\text{total}}[b]) |\Delta z_{mb}| / 10}$ evaluated strictly at the aggregate session SPL $L_{\text{total}}[b] = 10\log_{10}(\sum_k P_k[b]) + 85.0$, and define the Fractional Masker Responsibility Weight of track $j$ in band $b$ toward Tier-1 target band $m$:

Then:

1. Exact Partition of Unity: Summing over all non-Tier-1 tracks $j$ and all bands $b$ yields

guaranteeing that the collective masking relief equals the exact target deficit $D_{\text{Tier1}}[m]$ regardless of whether the session contains $M = 2$ or $M = 64$ tracks.

1. Zero-Power Track Immunity: If a track has zero energy in band $b$ ($P_j[b] = 0$), then $w_{j \to m}[b] \equiv 0 \implies \Delta G_j^{\text{Cochlear}}[b] \equiv 0.000\text{ dB}$.
2. Hirsh–Durlach Binaural Masking Level Difference (BMLD) Spatial Relief: Tracks panned away from the Tier-1 focal source ($\Delta\theta_j = |\theta_j[b] - \theta_{\text{Tier1}}[m]| > 0$) receive automatic spatial unmasking relief $\beta_{\text{BMLD}}(\Delta\theta_j) = 1 - 0.65\sin^2(\frac{\Delta\theta_j}{2}) \in [0.35, 1.0]$, reducing unnecessary EQ cuts on wide-panned guitars and keys by up to $65\%$. $\blacksquare$

5.2 Theorem 5.2 — 7-Dimensional Product Manifold $\mathcal{M}_{\text{Spec}}^{(4\text{D})} \times \mathcal{M}_{\text{Width}}^{(2\text{D})} \times \mathcal{M}_{\text{Groove}}^{(1\text{D})}$

Implemented in ProductManifoldStateV35:

1. 4-D Orthonormal Legendre Spectral Sub-Manifold $\mathcal{M}_{\text{Spec}}^{(4\text{D})}$: Augments v2.5's 2-D (Tilt, Smile) spectral coordinates with orthonormal discrete 3rd-order (kBasisPultecS12, Pultec low-shelf boost + 250 Hz dip + presence lift) and 4th-order (kBasisWContour12, sub-anchor + low-mid pocket + vocal presence + air sheen) Legendre polynomials over the 12 ERB bands ($\langle \mathbf{p}_m, \mathbf{p}_n \rangle = \delta_{mn}$). To prevent band-limited sources (e.g., vintage telephone vocals with -60 dB energy at 28 Hz and 16 kHz) from inducing Runge's phenomenon edge blowups at the endpoints $b \in \{0, 11\}$, $(c_3, c_4)$ are multiplied by the Full-Bandwidth Occupancy Gate:

1. 2-D Stereo Width Sub-Manifold $\mathcal{M}_{\text{Width}}^{(2\text{D})}$: Tracks (MidBandWidth, AirWidth) with a strict mono-below-120 Hz boundary anchor.
2. 1-D Coincidence-Gated, Onset-Only Micro-Groove Sub-Manifold $\mathcal{M}_{\text{Groove}}^{(1\text{D})}$: Measures the kick-to-bass onset lag $\Delta\tau_{\text{KB}} = t_{\text{bass,onset}} - t_{\text{kick,onset}}$ and adjusts the bass compressor's attack time $\tau_{\text{atk,bass}}$ toward the genre pocket target $\Delta\tau_{\text{target}} \in [+4, +14]\text{ ms}$ subject to two musical invariants:

• Coincidence Gate ($|\Delta\tau_{\text{KB}}| \le 36\text{ ms}$): Intentional 8th/16th-note bass syncopations ($|\Delta\tau_{\text{KB}}| > 36\text{ ms}$, e.g., 125 ms at 120 BPM) return attackMultiplier == 1.000000, never pulling syncopated notes toward the kick!
• Onset-Only Window ($0 \le t - t_{\text{bass,onset}} \le 35\text{ ms}$): Modulation acts exclusively on the initial transient attack and returns to $1.0$ during the sustain tail, preventing sustain-tail gain pumping.

5.3 Theorem 5.3 — Square-Wave-Immune Biphase Pre-Distortion Tomography & $+4.5\text{ dB}$-Clamped Cubic Hermite Transient De-Clipper

The Fourier series of an ideal symmetrical square wave (e.g., #32 Synth Lead Moog/Prophet pulse wave with $50\%$ duty cycle) is:

Proposition 5.3 (Identical Biphase of Square Waves and Symmetrical Hard Clipping): For an unclipped ideal square wave, the fundamental phase is $\phi_1 = -\frac{\pi}{2}$ and the 3rd-harmonic phase is $\phi_3 = -\frac{\pi}{2}$. Therefore, the normalized 3rd-order biphase is

which is identical to the biphase of a symmetrically hard-clipped sine wave! A naive v3.0 de-clipper driven solely by $\mathcal{B}_{3,1} \approx -1$ and flat-top plateau detection will falsely reconstruct parabolic arches onto every cycle of a clean square-wave synth lead, destroying its pulse timbre.

Implemented in PreDistortionTomographyV35 and ActiveTransientDeClipperV35:

1. Square-Wave / Pulse Synth Immunity Gate (squareWaveImmunity01): Evaluates the rail plateau duty cycle $D_{\text{rail}} = \frac{L_{\text{plateau}}}{T_0}$, harmonic temporal stationarity $\mathcal{S}_{\text{stat}}$, and digital ceiling proximity $P_{\text{dBFS}}$. Because an intentional pulse/square synth exhibits a wide duty cycle ($D_{\text{rail}} \in [0.24, 0.55]$) or stationary odd harmonics below -1.2 dBFS, whereas hostile transient clipping exhibits brief rail plateaus ($D_{\text{rail}} < 0.18$) near 0 dBFS, squareWaveImmunity01 exceeds $0.98$ on square-wave synths, reducing hostileFlatTopClipProb01 to $< 0.008$.
2. Sign-Opposed 2-Sample Chord-Slope Cubic Hermite De-Clipper (+4.5 dB Ceiling): Across a confirmed clipped plateau $\{n_0, \dots, n_1\}$ of length $L = n_1 - n_0 + 1 \in [3, 64]$, 2-sample chord slopes $m_0 = \frac{1}{2}(x[n_0] - x[n_0-2])$ and $m_1 = \frac{1}{2}(x[n_1+2] - x[n_1])$ are verified for strict sign opposition ($\operatorname{sgn}(x[n_0]) m_0 > 0$ and $\operatorname{sgn}(x[n_0]) m_1 < 0$) and bounded so the reconstructed Cubic Hermite arch $h_{10}(u) = u(1-u)^2$, $h_{11}(u) = u^2(u-1)$ never exceeds $+4.5\text{ dB}$ (1.6788 |x[n_0]|) above the clipped rail.

5.4 Stream D Instrument PDEs (#31 Rhodes, #33 Brass, #34–#37 Structural Kalman)

1. Rhodes Biot–Savart Odd-Harmonic Sieve (RhodesBiotSavartSymmetryTrackerV35): When a Fender Rhodes tine oscillates symmetrically across the magnetic pickup pole piece ($y_0 \to 0$), the fundamental flux vanishes and $2f_0$ dominates. Rather than halving $f_{\text{dom}}$ based solely on a candidate at $0.5 f_{\text{dom}}$ (which triggers falsely on root-fifth dyads where $3 \times \text{Root} = 2 \times \text{Fifth}$), we require verified odd harmonics of the true fundamental at $1.5 f_{\text{dom}} = 3 f_0^{\text{true}}$ AND $2.5 f_{\text{dom}} = 5 f_0^{\text{true}}$.
2. Polarity-Invariant Band-Limited Brass Burgers Shockwave Tracker (BrassBurgersShockwaveTrackerV35): High-SPL brass wavefronts steepen along the cylindrical/conical bore via the 1-D lossless Burgers equation $\frac{\partial p}{\partial x} - \frac{\gamma + 1}{2\rho_0 c_0^3} p \frac{\partial p}{\partial \tau} = 0$. To remain 100% invariant to console/mic polarity inversion ($x \mapsto -x$) and high-frequency hiss, the signal is low-pass filtered at 16 kHz (lp16kState) and evaluated via the Polarity-Invariant Derivative Skew Ratio:

which yields identical shockwave indices (|shockPos - shockNeg| == 0.0000) under phase inversion.

1. Transport-Adaptive, Loop-Safe Structural Kalman Filter (TransportAdaptiveStructuralKalmanV35): Adapts dynamically to the DAW host time signature (cycleBeats = clamp(4 * beatsPerBar, 4, 32) for 3/4, 4/4, 5/4, 6/8, 7/8), immediately suppresses pre-charging on discontinuous playhead loop/seek jumps, and requires $N_{\text{obs}}[b] \ge 2$ observed cycles before pre-charging master-bus transformers 50 ms ahead of a $+3.5\text{ dB}$ chorus downbeat.

6. Empirical Verification & Multi-Rate Red-Team Benchmark Suite

The complete Taxonomy38 v3.5 mathematical engine is continuously verified by our self-contained C++20 torture suite in rd_lab/Taxonomy38V35AnatomyVerification.cpp. Table 6.1 summarizes the empirical benchmarks across all 10 verification checks (44.1 / 48 / 96 / 192 kHz).

7. References

1. Fletcher, N. H., & Rossing, T. D. (1998). The Physics of Musical Instruments (2nd ed.). Springer-Verlag. (Euler–Bernoulli dispersive string beams, coupled two-headed cylindrical shell cavities, and Rayleigh–Schrödinger degenerate membrane perturbation).
2. Zavalishin, V. (2018). The Art of VA Filter Design (Rev. 2.1.0). Native Instruments. (Trapezoidal Zero-Delay Feedback integrators and state-variable topologies).
3. Simper, A. (2013). Linear Trapezoidal Integrated State Variable Filters (Cytomic Technical Note). Cytomic. (Left-half-plane ZDF state-variable and all-pass formulations).
4. Gray, A. H., & Markel, J. D. (1975). "A normalized digital filter structure." IEEE Transactions on Acoustics, Speech, and Signal Processing, 23(3), 268–277. (Time-varying unitary lattice all-pass stability).
5. Kaiser, J. F. (1990). "On a simple algorithm to calculate the 'energy' of a signal." Proceedings of the IEEE ICASSP, 381–384. (Discrete Teager-Kaiser Energy Operator).
6. Fant, G., Liljencrants, J., & Lin, Q. (1985). "A four-parameter model of glottal flow." STL-QPSR, 26(4), 1–13; & Hanson, H. M. (1997). "Glottal characteristics of female speakers: Acoustic correlates." Journal of the Acoustical Society of America, 101(1), 466–481.
7. Titze, I. R. (2008). "Nonlinear source–filter coupling in phonation: Theory." Journal of the Acoustical Society of America, 123(5), 2733–2749. (Epilarynx tube impedance and soprano $f_0 \approx F_1$ formant tuning).
8. Zwicker, E., & Fastl, H. (1999). Psychoacoustics: Facts and Models (2nd ed.). Springer; & Moore, B. C. J., Glasberg, B. R., & Baer, T. (1997). "A model for the prediction of thresholds, loudness, and partial loudness." Journal of the Audio Engineering Society, 45(4), 224–240.
9. Touzé, C., Bilbao, S., & Cadot, O. (2012). "Transition scenario to wave turbulence in thin vibrating plates." Journal of Sound and Vibration, 331(2), 412–433. (Föppl–von Kármán cymbal wave turbulence).
10. Hirschberg, A., Gilbert, J., Msallam, R., & Wijnands, A. P. J. (1996). "Shock waves in trombones." Journal of the Acoustical Society of America, 99(3), 1754–1758. (Nonlinear Burgers brass wavefront steepening).