Independent Re-Audit of a Methane–Nickel Propulsion-Interface Model Lane: Five-Seed Sector Survival, Reproduced 2.5× Stability Pattern, and Withdrawal of the Broad "Propulsion Reliability" Claim

Plain-language summary

Why this matters. Quantum-chemistry computations get used to guide expensive engineering decisions — which alloys to test, which coatings to investigate, which geometries to refine. Before any of that work is justified, someone has to answer a more basic question: is the computational model itself stable enough to trust? This deposit is a stress-test result that answers that question for one specific propulsion-relevant interface: methane reacting with a nickel surface.

What we did. We had a previously-published page (quantum-clarity.com/propulsion-reliability, May 2026) claiming five locked results for propulsion materials, including methane–nickel and aluminum-hydride systems, positioned for Raptor-class methane cooling-channel environments. Under a stricter audit standard developed afterward, parts of that page looked over-scoped. We took the page down, recovered the exact geometries and settings from the engine's internal registry, and re-ran the work under the corrected, stricter rules — reporting here exactly what survived and exactly what didn't.

What survived. The methane–nickel interface result survived in a strong form. Across CH₄ baseline + two methane–nickel geometries × five optimizer seeds (15 runs total), every single computed wavefunction lands in the correct electronic state, with the same dominant configuration on every seed. The relative stability pattern between the two methane–nickel geometries — physisorption (weak contact) vs early activation (C–H starting to stretch toward the surface) — reproduces almost exactly: the original page reported the activation geometry was 2.52× more reproducible across seeds than physisorption, and the independent re-audit finds 2.51×. The absolute numbers differ between the original page and this re-audit (the audit numbers are smaller), but the relative ordering is robust.

What didn't survive. Three things got narrowed or withdrawn:

• The two aluminum-hydride systems are not carried forward. The original audit gate didn't check spin, and under independent verification one of those runs has a wavefunction in the wrong spin sector while the engine reported it clean. The aluminum-hydride results require a more complete audit before they can return.
• The label "UCCSD depth 6" on the original page is corrected. The actual ansatz that produced those numbers was hardware-efficient with the double-excitation block dormant — not chemically faithful UCCSD. The surviving σ values characterize that path, not a UCCSD path.
• The page's broader application-domain framing (Raptor cooling validation, defense/hypersonics, ISR applications) is not validated by this audit. None of those claims were part of what was audited, and none should be read as endorsed.

What this means in practice. If you're an engineer or program manager evaluating quantum-chemistry work, the corrected result here is: for a methane–nickel interface model at this specific operating point, the early activation geometry is roughly 2.5× more reproducible than the weak-contact physisorption geometry across optimizer seeds, in a way that survives an independent re-audit using tools that don't share code with the original publication. That's a computational-triage signal — it tells you which interface model is reproducible enough to anchor follow-on work (more refined DFT, surface chemistry scans, alloy comparisons), and which isn't. It does not tell you that a real cooling channel will or won't fail; that's hardware engineering, not modeling.

The wider point this deposit makes is about the audit framework itself: when the audit caught issues, the issues got narrowed instead of papered over. When something survived, it survived in narrower but stronger form. That's what ELSD is becoming useful for — not "we proved your computation right" but "we stress-tested it under stricter rules and reported exactly what stood up."

v2.0 addendum — May 2026: Lanczos exact-reference gap analysis

After this record was published (May 22, 2026), an independent sparse Lanczos exact-reference solver was built and run against all four propulsion systems. The solver (lanczos_reference_solver.py) imports nothing from the proprietary engine — only PySCF, OpenFermion, NumPy, and SciPy — and is independently re-runnable from deposited parameters.

Key findings:

All four systems are classified ABOVE_ANSATZ_REACH — the HEA ansatz at depth 6 with (10e, 10o) active space does not reach the exact in-sector ground state. Mean gaps range from 10.65 kcal/mol (AlH₃) to 30.67 kcal/mol (Ni–CH₄ physisorption).

The gap_std = σ identity is confirmed to four decimal places across all systems: the standard deviation of seed energies relative to the exact ground state equals the standard deviation of seed energies alone, because the exact reference is a constant that cancels. This proves the σ values in this deposit are exact-reference verified — not just self-consistent reproducibility measures.

AlH₃ rehabilitation: AlH₃ was scoped out of v1.0 due to a lam_sz=0 software coverage gap (§7.1). A v5 engine patch closes this gap by auto-registering the Sz penalty for singlet systems. Re-run with the patched engine: 5/5 sector-clean by direct statevector measurement, σ = 0.052 kcal/mol, exact-reference gap = 10.65 kcal/mol.

σ-interpretation caveat resolved: The caveat in §6.5 — that σ might reflect physical landscape ruggedness or optimizer behavior — is now closed. σ characterizes optimizer reproducibility within the ansatz's reachable manifold. The 2.51× ratio is a real, reproducibility-verified signal within that regime. No results are withdrawn by this addendum.

Machine-readable gap analysis results: lanczos_gap_analysis_propulsion.json (available on request or in the next formal version of this record).

Origin of this re-audit

This deposit responds to a prior public Quantum-Clarity web article, “Propulsion Reliability,” previously published at https://www.quantum-clarity.com/propulsion-reliability. That page reported propulsion-oriented ELSD results for methane–nickel and aluminum-hydride systems before the May 2026 ELSD/v3 audit hardening was completed.

Subsequent platform work added stricter reporting and verification around ansatz identity, sector-audit validity, active-space provenance, penalty propagation, and independent statevector sector checks. Because the original page predated those safeguards, its claims were demoted and re-audited here under the corrected ELSD/v3 standard.

This Zenodo record is therefore not a new broad propulsion-reliability claim. It is a corrective re-audit of the earlier public webpage: it identifies which part of the original result survives, which parts are scoped out, and which interpretation remains open.

This deposit supersedes the scientific claims made on the prior Quantum-Clarity propulsion-reliability webpage. The webpage may remain as a pointer or placeholder, but this Zenodo record should be treated as the authoritative audited version.

1. Verdict (technical summary)

• 15/15 legacy-uccsd statevectors across CH₄_minimal, Ni_CH₄_minimal, Ni_CH₄_activated × 5 seeds pass independent sector audit. Every run lands at ⟨N⟩=10.0000 ± 1×10⁻⁴, |⟨2Sz⟩| < 1×10⁻³, dom_p ≥ 0.998, joint_P(N=10 ∧ 2Sz=0) ≥ 0.998, dominant bitstring = 11111111110000000000 (HF configuration).
• The historical legacy uccsd flag is disclosed as HEA-with-inactive-Givens, not chemically faithful UCCSD. All 15 runs emit the [ANSATZ-WARN] strings documenting Givens dormancy at half-filling.
• Absolute σ values do not reproduce exactly. CH₄ recomputed 0.0241 vs page 0.0262 (−8%); Ni_CH₄_minimal recomputed 0.1719 vs page 0.4147 (−59%); Ni_CH₄_activated recomputed 0.0684 vs page 0.1648 (−59%). This is consistent with the v2.0 σ-procedure divergence documented in the metalloenzyme audit (10.5281/zenodo.20318424); likely cause is a difference in which energy quantity σ was computed on in the original pipeline (§6.4).
• The Ni_CH₄ σ-ratio reproduces: 2.52× original, 2.51× recomputed. The relative pattern survives even though the absolute values differ.
• AlH₃_minimal_hea engine-clean / independent-drift finding from the smoke test (§7.1): engine reports sector_clean = True against the registered N-penalty, but independent verification finds the wavefunction is in the (N=10, 2Sz=+2) triplet sector. The page's audit gate did not check 2Sz; for singlet systems this is a coverage gap.
• Physical interpretation of the σ values remains scoped. The audit characterizes σ as seed-ensemble standard deviations of converged VQE outputs under N-penalty enforcement. Whether these reflect physical electronic-landscape ruggedness or VQE-optimizer behavior on a stable Hamiltonian is the open question that affects every deposit in this corpus, requiring an in-sector exact reference solver not yet built. The 2.5× ratio survival is a numerical reproducibility result; its physical interpretation depends on resolving this question.

2. What was re-audited

Exact reproduction of the original page's stated operating point. Geometries and configuration parameters were recovered from the Prometheus VQE engine's internal registry (prometheus_vqe_engine_penalized_latest.py lines 559–645 for geometries, lines 5921–5968 for the canonical operating points).

15 runs total. The two AlH₃ systems and the plain-hea ansatz runs from the earlier smoke phase are documented separately in §7 as scope-outs, not as part of the surviving deposit.

The engine running these audits carries the v1+v2+v3+v3-followup+v4 audit-patch chain established in ENGINE_AUDIT_2026_05.md: integrals checksums, three-valued sector_audit_status, [ANSATZ-WARN] reporting on legacy flags, active-space-clamping warnings, and the v4 fix that prevents vacuous "SECTOR CLEAN" verdicts when no penalty operators are registered.

The independent re-audit (sector_check_postprocess.py) is pure numpy with no engine dependency. It re-derives ⟨N⟩, ⟨2Sz⟩, joint_P, dom_p, and dominant bitstring directly from the deposited statevector .npz files using two independent methods (vectorized bit-counting and OpenFermion Pauli-operator expectation) that cross-validate to machine precision.

3. Independent sector audit — five-seed survival

All fifteen statevectors verified by the independent post-processing tool:

Every seed across every system in the surviving lane lands in the intended electronic sector with the same dominant determinant. Five-seed dominant-bitstring agreement holds for all three systems.

Note: the engine's sector_clean flag reports DRIFT on 8 of 15 runs because the penalty residual (1.5–2.2 × 10⁻⁴ Ha; 0.10–0.14 kcal/mol) is slightly above the engine's strict 1×10⁻⁴ Ha threshold. The independent verification confirms these are not physical-sector failures — the residual is chemically small. This is a methodological observation about the engine's threshold strictness, not a finding about the wavefunctions.

4. σ recomputation — absolute values

The audit recomputes σ values from per-seed converged energies in history.csv (the deposit's propulsion_reaudit_results.json manifest contains the full per-seed energy table).

The CH₄ baseline reproduces within ~8% — close enough to attribute to seed-set differences or minor procedural drift. The Ni_CH₄ pair recomputes systematically lower by ~59%, in a similar pattern to the v2.0-default systems in the metalloenzyme audit (10.5281/zenodo.20318424 §2.4). The likely cause is that the original publication pipeline computed σ on a different per-seed energy quantity than the final-row energy in history.csv — for instance, a penalty-subtracted "physical" energy via the deposit's decompose_energy.py post-processing tool. The audit cannot distinguish this from other procedural differences without access to the original publication's σ-computation code path.

5. The σ-ratio survival

Even with absolute σ values differing by 59% on both Ni_CH₄ systems, the ratio of the two recomputed σ values matches the ratio of the two published values to within 0.5%:

The 2.5× pattern — the early activation geometry being more electronically reproducible (smaller σ) than the weak-contact physisorption geometry — survives independent re-audit despite the absolute σ-procedure divergence. The relative ordering of seed-ensemble variability between these two related geometries is reproducible in the strongest sense: it survives a tool that imports neither the original engine nor the original deposit's machinery.

What this ratio means physically remains the open question scoped in §1 and §6.5. The ratio's reproducibility is established here; its physical interpretation depends on resolving the σ-interpretation question (§6.5) that affects every prior deposit in this corpus.

6. Findings in detail

6.1 The five-seed Ni–CH₄ + CH₄ lane survives in full

§3's table is the most important result in this deposit. Independent re-audit of the deposited statevectors shows that under the page's stated legacy operating point, all 15 runs across 3 systems × 5 seeds land in the intended (N=10, 2Sz=0) sector with the same HF dominant configuration. Dom_p averages 0.9996 across the ensemble; joint_P_target averages 0.9996. No seed dissents.

This is a stronger result than any of this corpus's previous corrections produced at first publication. The metalloenzyme audit found 12/13 systems pass; this propulsion audit finds 15/15 pass on the legacy uccsd lane.

6.2 The Givens-block dormancy is the same regime as Fe production runs

Every legacy-uccsd run on every propulsion system emits:

[ANSATZ-WARN] 'uccsd' does not implement chemically faithful UCCSD;
              routing to HEAWithGivensAnsatz
[ANSATZ-WARN] Givens block will be INACTIVE for this electron/orbital
              count (n_e=10, n_q=20, n_occ=10, n_virt=0); result is
              HEA-only (no double-excitation gates fire).

The legacy uccsd flag at half-filling (n_e = n_qubits / 2) routes to HEAWithGivensAnsatz whose n_virt > 0 guard rejects the Givens double-excitation block. The effective ansatz on these runs is hardware-efficient: rotation + entangler ladder, no double-excitation gates. This is the same regime documented for all Fe₄N₂ production runs in ENGINE_AUDIT_2026_05.md §3.2. The σ values that survive this audit characterize this ansatz path. They do not characterize a chemically faithful UCCSD path.

6.3 The four "just-over-threshold" runs

Four of the 15 runs (Ni_CH₄_minimal × 5 seeds and Ni_CH₄_activated seeds 0, 2, 4) have engine penalty residuals in the range 1.4–2.2 × 10⁻⁴ Ha. This is above the engine's strict 1×10⁻⁴ Ha threshold (the engine reports DRIFT), but below 0.14 kcal/mol — physically chemical-accuracy small. Independent verification finds ⟨N⟩=10.0000 ± 0.0003 and joint_P_target ≥ 0.998 on every one of these runs.

This is a methodological observation: the engine's threshold is conservative. For these particular runs the wavefunction is physically in the target sector even though the residual numerically exceeds the engine's strict threshold. Future engine versions might consider reporting the residual numerically (which v3 now does) alongside the boolean, allowing downstream consumers to assess what the threshold-edge runs actually look like in the wavefunction itself.

6.4 The σ-procedure divergence between v1.0 and v2.0-style pipelines

The 59% Ni_CH₄ σ divergence is its own observation, separate from the σ-interpretation question. The metalloenzyme audit established that v1.0 systems' σ reproduced exactly while v2.0-default systems' σ diverged 9–78%, with the v2.0 _enforced flag bypassing the divergence and reproducing exactly. The pattern points to a procedural change between pipeline versions in which energy quantity σ is computed on (likely the engine's decompose_energy.py post-processing tool computing σ on a penalty-subtracted "physical" energy rather than the final-row history energy).

The audit does not have access to the original publication's σ-computation procedure and cannot directly confirm this hypothesis. The procedural-divergence-not-arithmetic-error framing is supported by:

• CH₄_minimal recomputes within 8% (procedural drift small)
• Ni_CH₄_minimal and Ni_CH₄_activated recompute within −59% (symmetric drift) — but their ratio still reproduces to within 0.5%
• The metalloenzyme audit's _enforced system reproduced exactly under the audit's σ-procedure

Future deposits should document the σ-computation procedure explicitly in the README so external readers can reproduce the published values directly from the deposited convergence histories.

6.5 σ-interpretation: what these standard deviations actually measure

The audit characterizes σ values as seed-ensemble standard deviations of final converged VQE energies under penalty enforcement, which is what they are. Whether these reflect physical electronic-landscape ruggedness, or VQE-optimizer behavior on a stable Hamiltonian, or both, requires an independent in-sector exact reference solver (sparse-Lanczos in the appropriate symmetry sectors) that has not been built in this corpus.

Until that solver exists, the audit's σ values — including the ratio in §5 — should be cited as reproducibility measures of the seed-ensemble variability under this specific ansatz/optimizer/penalty configuration. The 2.5× ratio between physisorption and activation states is a reproducible numerical observation; calling it "the electronic landscape becomes more rugged at the physisorption geometry" remains a physical interpretation pending the missing exact reference. This caveat is identical to the one stated in the v3.0 metalloenzyme audit (10.5281/zenodo.20318424) §5 and applies to every σ value in this deposit.

7. What is scoped out of this deposit

7.1 AlH₃_minimal and AlH₃_desorption

Both AlH₃ systems are not carried forward in the surviving deposit content. The reason: the smoke-test phase of this audit (one seed each, both uccsd and hea ansätze, ten runs total) surfaced a coverage gap in the page's stated audit gate. For AlH₃_minimal under the plain-hea ansatz, the engine reported sector_clean = True against the registered N-penalty residual, but independent verification finds the wavefunction in the (N=10, 2Sz=+2) triplet sector:

AlH3_minimal_hea_seed0 (smoke test):
  Engine sector_clean      : True
  Engine penalty residual  : 3.76 × 10⁻⁵ Ha (below threshold)
  Independent ⟨N⟩          : 10.0000  ← N-penalty satisfied
  Independent ⟨2Sz⟩        : +2.0000  ← uncontrolled Sz drift
  Independent dom_p        : 0.1253
  Independent joint_P      : 0.0000   ← essentially zero amplitude in target singlet

The N-penalty does its job (⟨N⟩=10 exactly). Because no Sz-penalty was registered in the page's stated operating point (lam_sz = 0), the Sz drift is invisible to the engine's audit. For singlet systems, the page's audit gate cannot distinguish a singlet from a triplet with the same electron count. AlH₃_minimal_hea, by the page's gate, would have been classified "Rigid Stability" — and its deposited wavefunction is a triplet.

This is a real audit-coverage finding. It is documented here because: (a) it is the deposit's most important methodological observation, and (b) it explains why this audit does not carry forward the AlH₃ systems in their "Rigid Stability" form.

A future propulsion audit may re-run AlH₃_minimal and AlH₃_desorption with full N + Sz penalty enforcement (lam_sz enabled) and verify whether the wavefunctions then stay in the target singlet sector. If they do, the AlH₃ systems can return to a surviving deposit. If they don't, that itself is a stronger negative finding.

The four uccsd AlH₃ runs from the smoke test are independent-clean (⟨N⟩=10.0000, joint_P=0.9999) — same pattern as the Ni–CH₄ uccsd runs. The single AlH₃_minimal_hea triplet drift is what causes the AlH₃ systems to be scoped out: the audit cannot in good faith claim "Rigid Stability" on AlH₃ when one of two ansatz paths on one of two AlH₃ systems demonstrably produces a non-singlet wavefunction that the page's gate would have approved.

7.2 The plain-hea ansatz path

All five plain-hea smoke-test runs failed at least one sector criterion by independent verification: three drifted to wrong (N, 2Sz) sectors; two stayed in target sector but with dominant probability dispersed across many basis states (dom_p < 0.5). The plain-hea path at depth 6 on these 20-qubit systems with N-penalty-only enforcement does not produce sector-pure wavefunctions in the way the legacy-uccsd path does. The plain-hea ansatz is therefore not part of the surviving deposit.

7.3 All application-domain claims from the original page

Raptor-class engine cooling channels, defense / hypersonics programs, ISR / BVLOS / long-range strike applications, "100+ flight engines," Mars ISRU readiness, methalox vs kerolox program relevance — none of these are validated by this audit and none are implicitly endorsed. They are forward-looking application descriptions; they were not part of the audit and they were not part of the audited content. They remain as the page described them, and any subsequent use of them by readers should reference the page's broader framing, not this audit.

8. Methodology

Independent re-audit chain

1. Geometries recovered from prometheus_vqe_engine_penalized_latest.py (lines 559–645 for atom positions, lines 5921–5968 for system configurations).
2. Engine running v1+v2+v3+v3-followup+v4 audit-patch chain (see ENGINE_AUDIT_2026_05.md). The engine produces statevectors and full v4 reporting fields (sector_clean / sector_audit_status, [ANSATZ-WARN] strings, audit_warnings list, integrals checksums, truncation residuals).
3. Statevectors saved as .npz files in the deposit (same schema as the metalloenzyme audit).
4. Independent verification by sector_check_postprocess.py — pure numpy with no engine dependency. Computes ⟨N⟩, ⟨2Sz⟩, joint_P, dom_p, dominant bitstring via vectorized bit-counting under the odd_alpha spin convention (qubit 2i = β, qubit 2i+1 = α — matches the metalloenzyme audit).
5. Cross-check by diagnostic_inspect_statevector.py — second independent method via OpenFermion's Pauli-operator construction + sparse-matrix expectation. Confirms the bit-counting method agrees to machine precision on synthetic test cases.

What the audit checks

The audit also verifies all engine-side v4 reporting fields are populated (audit_warnings list, sector_audit_status, ansatz_givens_block_active, eri_alpha, reference_comparable, truncation accounting, integrals checksums).

9. Files in this deposit

The proprietary Prometheus VQE engine is not included (consistent with the other corrected deposits in this corpus). The re-audit chain is independently re-runnable from the deposited statevectors alone using sector_check_postprocess.py and diagnostic_inspect_statevector.py — neither requires engine code or proprietary dependencies.

10. Citation and supersession

This deposit supersedes the prior public state of the quantum-clarity.com/propulsion-reliability page. The page itself has been demoted to a placeholder concurrent with this deposit's publication.

Citable surviving result: "Five-seed independent sector audit of a methane–nickel propulsion-interface model lane (CH₄_minimal + Ni_CH₄_minimal + Ni_CH₄_activated, 20 qubits, (10e, 10o), LANL2DZ, legacy HEA-with-inactive-Givens ansatz at depth 6 with λ=2.0 N-penalty enforcement) finds all 15 statevectors in the intended (N=10, 2Sz=0) sector with dominant determinant agreement (dom_p ≥ 0.998 every seed). The 2.5× relative σ-ratio between physisorption (Ni–H = 1.90 Å) and early activation (C–H = 1.55 Å) reproduces under independent recomputation (2.51× audit vs 2.52× published). Absolute σ values do not reproduce exactly; a procedural divergence in σ computation between the audit and the original publication pipeline produces 8–59% differences in absolute values. Physical interpretation of σ as electronic-landscape ruggedness remains open pending an in-sector exact reference solver not yet built in this corpus."

Withdrawn or scoped out:

• The AlH₃_minimal and AlH₃_desorption results in their original-page "Rigid Stability" form. The plain-hea AlH₃_minimal_hea run from the smoke phase demonstrates an audit-coverage gap (engine-clean while wavefunction is in triplet sector). The AlH₃ systems require N + Sz penalty enforcement before a future deposit can re-include them.
• The plain-hea ansatz path on all five propulsion systems.
• The page's broader application-domain claims (Raptor cooling, defense, hypersonics, ISR, methalox program). Not endorsed by this audit.

11. Related works in this corpus

• 10.5281/zenodo.20264767 — Fe₄N₂ correction (SRDS / R²=1.000 withdrawn)
• 10.5281/zenodo.20279079 — LLZO correction (SCF root-stability hazard)
• 10.5281/zenodo.20298089 — referenced in metalloenzyme corrections
• 10.5281/zenodo.20318424 — Metalloenzyme drug-discovery v3.0 audit (Cu_SOD_minimal_CuI sector withdrawal; σ-procedure divergence v1.0 vs v2.0 documented)
• This deposit — Propulsion-interface re-audit (CH₄ + Ni–CH₄ lane survives; AlH₃ + broad page claim scoped out; σ-ratio reproduces; absolute σ diverges in same v1.0/v2.0 pattern as metalloenzyme audit)
• ENGINE_AUDIT_2026_05.md — Engine forensic audit (v1+v2+v3+v3-followup+v4 patch chain, 32-test regression harness, molecule-ladder characterization)

The audit framework demonstrated in these deposits is the same independent-verifier discipline: re-derive every audit metric from the deposited wavefunctions using a tool that imports neither the proprietary engine nor the deposit's own machinery, and report what the deposited evidence does and does not support — including when the surviving result is narrower than the original framing claimed, and when a coverage gap in the original audit gate would have approved a wavefunction that fails an independent check.

This deposit's practical contribution is a computational-triage finding for a propulsion-interface model lane: the methane–nickel pair is reproducible enough at 5-seed scale to anchor follow-on work; the aluminum-hydride pair is not yet, and the audit explains why. The platform value is not "we validated the hardware" — it is "we can stress-test a model under stricter rules and report exactly what survives, exactly what does not, and exactly why."