AERO / CANONICAL PUBLIC EVIDENCE INDEX
Bounded verification and rejection behavior. Analytical agreement is not experimental validation; a FAIL can be the correct engineering outcome.
DATED EVIDENCE ADDITION · OCTOBER 10, 2026
A frozen adapted reader plus deterministic format, unit-provenance and solver checks. Fresh acceptance cases, legacy regressions, measured overhead, actual CalculiX renders, and retained failures from two stock reviewer models. This is a bounded local pipeline study, not a true MoE or public runtime deployment.
PUBLIC INFERENCE BENCHMARK · SEPTEMBER 24, 2026
Five-repeat measurements with fresh-prefix controls. A mocked 12B agent loop took 15.25 s on the workstation and 27.93 s on GB10; GB10 completed the 128k context ladder. This measures inference—not CFD execution or accepted engineering decisions.
ALEX BLYTHE · SEPTEMBER 21, 2026 · LATEST AUDIT
40 behavioral tests and 36 topology-verified field comparisons. Native flux, energy/storage and interface heat rates recovered. Cold-side field/flux settling and unresolved acceptance definitions still block qualification. No new CFD or historical rescoring.
DATED STATUS REPORT · 2026-09-20
Fresh coarse, medium, and fine runs reached 800 s under the frozen V2 Core + two-fluid CHT profile contract. Direct outer telemetry was retained, but frozen trend, field-stationarity, energy/interface, and local-manifold gates were not all established; the qualification remains FAILED_NUMERICAL_GATES.
DATED STATUS REPORT · 2026-09-19
The recent pressure-residual, outer-corrector telemetry, matched-fixture, and relative-criterion work was designed to define a defensible convergence measurement without retroactively tuning HX-CFD. The result is intentionally bounded: U outer updates are useful diagnostic evidence, but no universal or cross-validated relative threshold is established.
7 evidence items
CFD · FAILURE INVESTIGATION
NUMERICAL FAIL · INVESTIGATION COMPLETEv8 moved h/e down and U up; all target gates still fail.
Includes a valid pressure-cap test, an invalid configuration test kept visible, and a corrected one-versus-two outer-corrector comparison. The stop rule ended the solver-control branch without a v9.
No mesh independence, matched experimental validation, wall-treatment suitability, or design readiness is claimed.
CFD · VERIFICATION + NEGATIVE CONTROLS
MIXED0.053% fine-grid error. Two real failures.
Three pipe grids agree with Poiseuille within 2%. Their errors decrease from 0.996% to 0.243% to 0.053%. The negative controls miss by 13.963% and 99.605%.
Both nozzle treatments fail near-wall readiness. y+ and all-component residual histories distinguish an active-field convergence problem from an ill-conditioned negligible component.
7 new OpenFOAM runs · 295 hashed archive files, including CAD and 5 retained CalculiX decks · 7-page paper. Criteria correction disclosed; not a blinded preregistration.
Analytical pipe verification and numerical-readiness rejection, not an experiment or a general nozzle validation. The original Rev 1.0 page, results, and bundle remain frozen.
CFD · ANALYTICAL BENCHMARK
BOUNDED PASS2.5 mm is the smallest tested passing gap.
9 OpenFOAM solves: 3 gaps × 3 grids. At 25 mL/s and a 40 Pa budget, the fine-grid losses are 177.964, 75.078, 38.440 Pa for 1.5, 2.0 and 2.5 mm gaps.
The smaller gaps fail the budget even though numerical checks pass. The 2.5 mm recommendation includes the declared numerical allowance.
200 hashed paper-package artifacts · analytical parallel-plate comparison, profile, residual and conservation checks. Not a sidewall, fitting-loss, or hardware validation.
This capability study uses 25 mL/s / 40 Pa. The separate interactive default uses 20 mL/s / 70 Pa; their recommendations are not interchangeable.
THERMAL · ANALYTICAL SCREENING
BOUNDED PASSLayout B: two 2 mm passages.
Three layouts assessed under a 100 W load, 72 °C inner-wall limit and 2 kPa straight-channel pressure budget. Nominal and adverse scenarios are retained.
Energy closure error: 0. The final radial heat-transfer check differs from its analytical reference by 0.001942%.
41 hashed paper-package artifacts. Analytical thermal screening—not thermal CFD or experimental validation. Manifold, contact and boiling effects remain outside the model.
THERMAL + STRUCTURAL · SYNTHETIC PREBENCHMARK
PREBENCHMARK10 runnable gates passed; paper-specific gates remain separate.
A source-derived printed-circuit heat-exchanger packet exercises prediction freezing, criteria hashing, curved-interface pressure and heat-rate transfer, generic OpenFOAM CHT, and generic three-level CalculiX acceptance.
These six operating points are synthetic pipeline tests—not publication results. The later Samarmad-specific run has its own failed numerical outcome and unscored experimental comparison.
STRUCTURAL · NUMERICAL ADEQUACY + REJECTION
MIXED21 retained solves. Seven configurations.
Bracket, pressure tubes, a port and thermal restraint were tested at three mesh levels. The port fails mesh sensitivity. The thermal-gradient case passes numerical checks but fails its stress requirement.
Pressure-tube numerical acceptability remains NOT ESTABLISHED with reaction-balance evidence missing. Every configuration retains design readiness NOT ESTABLISHED.
CalculiX 2.20 · Gmsh / CadQuery. Decks and analytical comparisons are bounded evidence, not validated structural design capability.
| Configuration | Numerical | Requirements |
|---|---|---|
| bracket Manifest | PASS | PASS |
| tube Manifest | NOT_ESTABLISHED | PASS |
| tube3d Manifest | NOT_ESTABLISHED | PASS |
| port Manifest | FAIL | NOT_ESTABLISHED |
| thermal-free Manifest | PASS | PASS |
| thermal-fixed Manifest | PASS | PASS |
| thermal-gradient Manifest | PASS | FAIL |
All 7 public configuration manifests were checked against their retained files: 0 mismatches. The Verification Bench bundle contains 5 of these retained solve decks, not all 21.
Download retained deck subset ↓CAD / GEOMETRY · NEGATIVE CONTROL
FAIL — correctly rejected2 solids. Solver not submitted.
The STEP fixture has intersecting ports and fillets. The intact reimport passes topology only; the defective variant violates the single-solid intake contract. No silent deletion or automatic repair.
Two synthetic STEP reimports · OpenCASCADE / CadQuery. This does not establish general imported-CAD robustness.
EXTERNAL BENCHMARKS
NASA TM-4074 Table XIII: 18 operating points, three grids per point, and 36 experimental targets. Criteria and execution code were publicly committed before production. 54 / 54 attempts finished in the published snapshot.
Reference withheld from solver execution; supervisory setup was not fully blinded. Numerical failures remain visible.
The links above lead to the original versioned records. This index does not replace or revise them.
62 committed source artifacts audited. Source snapshot 2ca8e5cbfbdb03515e961be8fc4143f2891e637b. Index hashes establish identity; they do not independently establish scientific validity.
Failures remain evidence: HARCC and Samarmad numerical failures, CalculiX NOT_ESTABLISHED states, port mesh-sensitivity and thermal-gradient requirement failures, imported-CAD rejection, the Blind Validation Challenge failure, and CEP-01 model qualification still open. No item establishes universal Aero validation, certification, autonomous design authority or design readiness.
CFD · TWO-FLUID CHT QUALIFICATION
FAILED NUMERICAL GATESThe original shell/tube/solid chtMultiRegionFoam qualification failed its 1% mesh gate. A separate high-resolution v7 remediation passes mesh independence, mass and energy checks, but residual convergence and monitor stability remain NOT_ESTABLISHED. A bounded nOuterCorrectors=2 control test did not establish an improvement and is diagnostic-only.
Real retained solver evidence, now with downloadable ParaView field snapshots and an outlet-monitor animation. No physical validation, generic CHT qualification, or HX-03 Stage B authorization is claimed.