Validation status by solver and reference

Caveat

First-order Rusanov smears the contact, so the bound is on mean accuracy, not on peak resolution. The nonlinear flux and EOS are evaluated pointwise (dequantize → compute → requantize); the rank-preserving TT-cross form is the large-L upgrade.

Caveat

The uncalibrated finite-rate network lands within +0.35 decade of the flight anchor, inside the ±0.70 chemistry-spread band. The closed-form Park-2T controller lands 1.27 decades below the anchor after the N₂–N₂ reduced-mass correction (μ = 14.007); its former near-anchor landing was an artifact of an invalid μ = 7.0 (the N–N atomic pair, which has no vibrational mode), and the offset is reported rather than re-admitted. Still a two-temperature Saha surrogate; the T_e = T_ve lumping is worth roughly 2×, and the landing is sensitive to the Millikan–White τ_vt model within the documented 2–5× chemistry-model spread. γ = 1.1 is an effective-γ closure, not perfect gas.

Caveat

Periodic, smooth, low-Reynolds and single-mode. It does not test immersed-body boundary conditions, turbulent rank growth, or multi-mode cascade. Gate 2 exists because single-mode Taylor–Green's convective term is a pure gradient the projection removes; a solver with a broken or zero u·∇u would still pass gate 1.

Caveat

The absolute C_d ≈ 23.8 is NOT an isolated-cylinder drag coefficient: roughly 30% blockage, a penalization-integral force over a smoothed 2-cell skirt, and a fixed-horizon read rather than a steady state. The verification claim is the convergence trend plus no-slip and positivity, never the absolute number. Reproducing an isolated C_d needs an inflow/outflow domain, out of scope for the periodic QTT solver.

Caveat

This gates rank, not physical accuracy; the quantitative accuracy gate for the compressible solver is qtt_sod, against the exact Riemann solution. The marched peak bond is reported and explicitly not asserted: a plain flux-through-front marcher injects angular structure and grows the bond to 64 over 6 steps even in the fitted coordinate. Bounding that is design D9 and the qtt_repin_marcher study.

Caveat

Forebody only, and structural: it bounds rank, not physical accuracy. The wake is explicitly out of scope: a separated unsteady wake needs turbulence and is a multi-feature structure no single fitted coordinate aligns; its bond is reported, never gated. The dynamic marched forebody rank is likewise reported, not gated: there is no 3-D body-fit metric yet, so the marcher runs Cartesian and grows the bond to 16 over 6 steps. A 3-D body-fit metric plus re-pinning is the open remainder.

Caveat

Superseded. The Tier-A closure over-predicts by three decades: it rides an incompressible rollout with a recovery-temperature reconstruction rather than a true post-shock thermodynamic path, and Saha equilibrium at the frozen RH temperature drives near-full ionization. No absolute coupled-CFD match is claimed. Retired by the Tier-B compressible marcher; use qtt_ramc_stagline instead.

Caveat

The 6%-of-span vortex offset is at a coarse 33² grid. Reporting resolution is 129² with t_end ≥ 150, Ghia's own grid, which takes hours. The committed baseline.txt for this target is a partial run log and records no RMSE; the figures above come from the target README.

Caveat

Acceptable but not DNS-grade at this grid. The integrated drag is close for the wrong reason: the pressure/friction split is off, with friction at 13% against the ~25% reference. Most of the +4.3% Strouhal excess is LY_D = 16 blockage (≈6.25%), leaving ~1–2% method error. A defensible accuracy claim needs a grid-convergence study (16→24→32/D, Richardson-extrapolated) plus C_L,rms, θ_sep and C_pb. This target has no baseline.txt; the figures are the committed run output re100_16_resolved.txt. Its staircase companion at the same 16 cells/D does not shed at all: the wake decays to a steady residual. That run's printed St 0.2444 is therefore the crossing detector firing on 7th-decimal noise, and its C_d 1.356 is a steady-flow value rather than a cycle mean. The aperture-resolved cut cells are what produce a sustained street here.

Caveat

At strong grading the coarse-pair order dips to ~1.7 and recovers to ~2.0 as the mesh refines. An earlier revision of this study mis-measured a convective order collapse; the cause was a measurement bug: pointwise 1-form values instead of edge integrals. The cochain convention is load-bearing.

Caveat

Only the energy-monotonicity invariant is gated; the DNS comparison is informational. 16³ is grossly under-resolved and cannot represent the small-scale dissipation peak, hence the −80%. Strictly the curve never peaks at this grid: the reported maximum falls at the final sample, t* = 10.05, so it is a monotone-rising tail rather than a resolved peak, where the DNS peak sits near t* ≈ 9. Reporting resolutions of 64³–128³ close this gap. Do not read the −80% as a solver error.

Caveat

The DEC solver has no inflow/outflow surface here; the sensor drives a prescribed moving wall in a confined periodic-x channel. At 25% blockage the run reports no clear shedding in the developed signal, so the printed Strouhal is a qualitative check for that confined case, never gated, and no isolated-cylinder Reynolds ladder is claimed. The committed baseline.txt is the 200-row probe stream; the divergence figure is the maximum over its residual column; the pass/fail summary lines are in cli_output.txt.

Caveat

Off-grid step counts introduce a phantom floor: dt = 0.005 is not a binary fraction, so steps·dt misses 1 by ~2e-17, which becomes a fixed ~1.9e-18 amplitude error. Guarded by evaluating the reference at t_final = dt·steps. Past a few thousand steps the two low-precision columns drift upward.