Claims and assessment routes

Global source/energy: PDF Section 2 equations (carrier, energy, source) and Theorem
Global evolution and dissipation. feedback_test.py integrates the energy ledger;
independent_solver_max_error compares all eight endpoint state coordinates and
the accumulated dissipated energy against independent real-gradient RK4.
energy_balance_residual is max over 401 samples of |F(t)+loss(t)-F(0)|.
RK4_energy_balance_residual is the RK4 endpoint ledger discrepancy.
norm_decay_bound_max_excess is max over samples of squared norm minus envelope.
Memory relaxation: analytic convolution proof in PDF Section 2; numerical finite
trajectories illustrate transient behavior, not an asymptotic certification.

Time ordering: PDF Section 3 interaction-picture and commutator proofs.
pulse_test.py evaluates ordered products, reversal and integrated exponential,
and independently reconstructs 2x2 exponentials. Its leading_relative_error is
|| (U-DV)/(i gamma alpha tau^2) - [K2,K1] || / ||[K2,K1]|| in operator norm.
feedback_test.py reports ||U-DV||_2, ||U-VR||_2 and ||V*V-I||_2.
symmetry_control.py uses an equivariant symmetric preparation with moving memory.
Its asymmetric commutator integral is trapezoidal quadrature, not an upper
enclosure. structural_checks.py checks the exact collective mode at saved times.

Full feedback: PDF Section 4 derives all Hessian/tangent blocks and the causal
kernel with its initial-memory term. feedback_test.py compares the amplitude
parts of the full tangent acting on (1,0,...,0) and the clamped tangent acting on
its first six coordinates, using Euclidean norm. Central finite differences of
the nonlinear source check the complete eight-component tangent, with eps=1e-5.
exact_obstructions.py checks the rational 9/20 initial tangent norm derivative
and positive-joint-Hessian failed skew alignment. closure_checks.py supplies
the nonautonomous scalar negative defect and zero-endpoint-leakage fixtures.

Moving metric: PDF Section 4 derives the connection from differentiation of
x=M^(1/2)y. response_test.py evaluates the exact prescribed diagonal fixture;
its increase in whitened norm is a coordinate-norm effect, not active gain.

Observable test: PDF Section 5 proves Richardson bias h^2 B3/4, noise 5 eps/h
and calibration bounds in operator norm. structural_checks.py uses independent
closed diagonal propagators and imposed worst-case errors, with 1e-12 allowance
for floating arithmetic. response_test.py uses the same spline-interpolated
memory for both hypotheses; residuals are relative Frobenius, using r=2.
Diagnostic fitted r is Re<tr(H*,C)>/||H||_F^2, not independent calibration.
No hardware error or B3 certificate is inferred from the resulting residual.

Tables and figures: generate_assets.py reads the saved/recomputed outputs only.
The public runner compares scientific outputs and regenerates all assets. Exact
formulas/proofs, finite fixtures, independent formulations and same-code replay
remain separate kinds of evidence. The public package is sufficient to assess
every result claimed in this paper without prior internal handovers.
