# AC14 gas–dust scientific replay package Save every linked scientific file in one ordinary directory. These scripts run without a private repository checkout, an interactive Lab engine or a network service. The matrices and grid code are independently authored scientific analysis, not a redistribution of the protected interactive entry. Use Python 3.10+ and the tested versions NumPy 2.0.2 / SciPy 1.13.1. The recorded host verification also ran successfully on Python 3.9.6. Create a new environment: ```sh python3 -m venv .venv source .venv/bin/activate python3 -m pip install -r requirements.txt python3 -W error replay-streaming.py --input streaming-input.json --out streaming-replay-rerun.json python3 -W error replay-radial.py --input radial-results.json --out radial-replay-rerun.json python3 -W error replay-brightness.py --data . --out brightness-replay-rerun.json ``` Keep linear-system.py and grid-patch.py beside replay-streaming.py. The full linear replay may take several seconds to a minute depending on hardware. Compare scientific values within numerical tolerance across different BLAS libraries: eigenvalue order and eigenvector phase can vary even with identical physics. The fixed input and file hashes identify exact saved inputs; they do not establish physical accuracy. The recorded local detached run reproduces the frozen outputs exactly with the pinned runtime on the same host. ## Three separate calculations 1. **Synthetic linear streaming benchmark.** The input fixes Stokes numbers, volume loadings and both radial/vertical wave numbers. A transverse projection of the gas equations produces six evolving amplitudes. A separate direct transcription of Youdin–Goodman equations 26–31 produces a generalized center-of-mass/relative-velocity pencil with six finite physical roots and algebraic constraint roots. Direct ODE and actual 2D periodic linear grid evolution independently check the growth. The grid retains all velocity components; it is not a one-dimensional radial trap. 2. **Synthetic finite-volume radial states.** radial-results.json stores actual evolved gas/dust mass and momentum fields and the measured spatial/time-step comparisons. replay-radial.py independently remeasures those saved states; it does not rerun the protected nonlinear radial solver. This pressure-trap experiment has no vertical modes and is not streaming instability. 3. **Observed DSHARP Elias24 intensity.** Elias24.profile.txt contains the exact public 300-row/eight-column source file. README_profiles is the source's full column description. Elias24_continuum.header contains only the 8640 original primary FITS-header bytes, obtained by a bounded range request. There are no image pixels or measurement sets here. replay-brightness.py uses the actual frequency and beam to reproduce brightness temperatures. Negative intensities remain; Planck temperatures are null when the inversion is undefined. ## Conventions and independent references Linear time is 1/Omega, length eta*r and velocity eta*vK. Perturbations have exp(sigma*t+i*k.x); positive real(sigma) means growth. Gas is incompressible in six-state calculations; eight-state weakly compressible references use cs=20. Stopping time is fixed. Background pressure acceleration maintains uniform drift. The complete gas-density contribution to drag is retained in the eight-state system. Benitez-Llambay/Krapp/Pessah 2019 v3, Table 4 and Appendix E give the matching LinA/B roots; Section 3.5.3, equation 86 explains the small established difference from Youdin/Johansen 2007, Table 1. The 2019 paper instead writes exp(-omega*t), so its complex omega is minus this package's sigma. Absolute versus relative dust amplitude also changes the eigenvector normalization. No benchmark coefficient is adjusted to match a printed root. The actual linear grid has one periodic wavelength in x and z, centered spatial derivatives, discrete Fourier pressure projection and RK4 time stepping. A mode has effective stencil wave number sin(k*dx)/dx. Each resolution starts from its discrete eigenvector; the integrator receives no growth rate. This isolates the integrator check, then 16/32/64 cells establish spatial convergence toward the continuum. Seeding a continuum eigenvector on a coarse grid can produce an early superposition transient. The separate DOP853ODE starts from the continuum seed. The frozen output gives every root, mode normalization, ODE history, measured finite-grid growth, half-step run and 32-cell final density array with first array axis radial x and second vertical z. No-feedback, no-vertical-mode and no-pressure-drift controls remove the positive linear growth. The terminal-velocity cubic is saved as an approximation with its wave-number/stopping-time domain measure, not as the reference growth. Equal Laplacian diffusion on every perturbation is a declared toy closure and shifts all eigenvalues by -D*k^2. It is not the full Umurhan/Estrada/Cuzzi alpha-disk model with separate viscosity, dust diffusion/pressure and dilution. The nonlinear radial patch instead uses length H=cs/Omega and velocity cs, with periodic 8H extent. Gas pressure comes from actual evolving density. The initial peak is prepared, then allowed to change, with no continuing peak force. A single-valued pressureless dust velocity cannot represent crossing streams or dense caustics. The finite early trapping calculation is not a nonlinear streaming-saturation or self-gravity calculation. Its 128-cell default relative L1 dust-density difference from a 1024-cell reference is 2.92 percent at t=8/Omega, and its peak is about 6.1 percent lower. Stored mass-conservation residuals must not be quoted as continuum accuracy. The source scheme uses first-order spatial Rusanov fluxes, SSPRK2 transport and symmetric exact drag/epicycle source steps. ## Observed-product provenance and use The official DSHARP release is project 2016.1.00484.L. The exact ASCII profile was last modified 5 August 2019; its README of 5 February 2020 explains the radius-column correction and propagation. The measured header frequency is 231943014843 Hz and the beam is 0.0367535241 by 0.0342671573 arcsec. No full FITS hash is claimed for a header-only request. The synthesized beam correlates radial samples; pointwise uncertainties are not a covariance matrix. Intensity, brightness temperature, dust amount, gas pressure and planet evidence are distinct quantities. Cite DSHARP I: Andrews et al. 2018, ApJL 869, L41, and DSHARP II: Huang et al. 2018, ApJL 869, L42, as requested by the official community data release. Original observational files retain their source terms; this package does not assert a new license for those files or for linked primary papers. Metadata and all checksums are retained in observed-provenance.json and provenance.json. ## Authored scientific-material permissions The independently authored Python scientific scripts are offered under MIT. Authored synthetic inputs, frozen outputs, prose and provenance are offered under CC BY 4.0 with attribution to Sandbox Physics, AC14, 2026-10-28. These grants cover the explicitly linked scientific package only, not the protected Lab engine. Original observed files and primary papers keep their own source terms. MIT notice: Copyright 2026 Sandbox Physics. Permission is hereby granted, free of charge, to any person obtaining a copy of these scientific scripts to use, copy, modify, merge, publish, distribute, sublicense and/or sell copies, subject to retaining this copyright and permission notice. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.