Approximate dates reconstructed from file timestamps
Experiment Note
The live-preview manuscript gathered the final repeated evidence:
- Kovasznay: five-seed reductions of roughly 33-46 percent.
- Re=100 cavity with 2 percent sparse CFD: 9.1 percent lower full-field velocity L2 and 26.2 percent lower sparse-velocity MSE.
- Cavity continuation at Re=100, 400, and 1000: 9/9 sparse-velocity-MSE wins and 8/9 full-velocity-L2 wins.
- Allen-Cahn: 29/35 metric-pair wins and 35.1 percent mean normalized reduction.
- Advection-diffusion: 35/35 wins and 58.6 percent mean normalized reduction.
- Sensitivity: moderate one-factor perturbations changed the nominal cavity score modestly; the largest reported mean increase was 1.7 percent.
- Single-seed Reynolds stress trace from Re=100 to 10,000: VARA had the smaller reference-free aggregate at 14/15 levels, averaging a 12.7 percent reduction. The Re=5000 reversal, where VARA was 13.8 percent worse, was retained rather than removed.
The Kovasznay CSVs were updated on 12 July, suggesting a final reconciliation of the analytical benchmark numbers with the manuscript tables.
Technical Essay
The paper explicitly discussed the main tradeoff. In the Re=100 cavity study, mean continuity residual moved from 0.0214 to 0.0220, about 2.7 percent worse, while full velocity error improved by 9.1 percent and sparse velocity MSE by 26.2 percent. In continuation, continuity changed by about 5.1 percent while full and sparse velocity errors improved by 7.5 and 19.3 percent. This was reported as a reconstruction-oriented allocation tradeoff in a soft UVP model, not hidden behind the aggregate score.
Runtime was also scoped carefully. Under the matched cavity schedules, VARA was approximately 2.6-4.9 percent faster in the reported settings, but the paper did not claim a general speed guarantee because diagnostics and probes add work. The transient raw runs were noted as approximately 9 percent slower.
Reading Note
The final literature synthesis supported a restrained conclusion: safe local control can improve reconstruction when error has a variable or spatial signature, but hard-incompressibility outputs, larger seed sets, joint sensitivity designs, inverse problems, three-dimensional cases, and high- fidelity high-Re references remain open work.
Progress Update
By 12 July, the repository contained anonymized manuscript, title page, cover letter, supplementary material, LaTeX source archives, an anonymized project archive, verification renders, extracted text checks, and source-archive compile tests. The package was structured for EAAI upload rather than only local compilation.