A public Manifest over three NETLIB linear programs (afiro, adlittle, and agg). The hybrid quantum interior point method runs in two Newton-system formulations against HiGHS 1.14.0 and Clp 1.17.9 under Tier A. Every route in the chassis returns REDIRECT. The dominant cost driver is full primal-dual vector output extraction, not Newton-system inversion.
sha256:e11a927f69827d5617ee3ce6a8f0838e057f64f7b4786f9ab8746a554129a813
SHA-256 over the canonical JSON serialization of this Manifest with the top-level hash field stripped. Reproducing the audit on the same inputs reproduces the hash exactly.
Inputs are MPS text files decompressed from the public NETLIB EMPS-compressed downloads. The three files are concatenated and hashed together. Precision target 1e-6 relative. Operational latency batch.
| INSTANCE | ROWS | COLUMNS | NONZEROS |
|---|---|---|---|
| afiro | 27 | 32 | 83 |
| adlittle | 56 | 97 | 383 |
| agg | 488 | 163 | 2,410 |
sha256:08a7065a6cfbf7312eed776d3e0c1c8116c6c4c78c3cc27e5046a94dcbfd8c84 SHA-256 over the concatenation of the three MPS files in the order shown. Different instance sets produce different Manifests; identical instance sets reproduce the runtime distribution within the runner's stated tolerances.
A second baseline establishes that the verdict does not hinge on a single solver. Both reach optimality on the largest instance in milliseconds.
| SOLVER | VERSION | CONFIGURATION | RUNTIME (s) | PRECISION | CONFIDENCE |
|---|---|---|---|---|---|
| HiGHS | 1.14.0 | default_dual_simplex, presolve on, Apple silicon, single core | 2.263e-3 | 1e-9 | strong |
| Clp | 1.17.9 | primal_simplex, presolve on, single-core CPU | 5.000e-2 | 1e-7 | adequate |
Wall-clock on agg, the largest of the three instances. afiro and adlittle terminated in 0.0217s and 0.000945s respectively. Captured by baselines/run-baseline.py against highspy 1.14.0.
Estimated runtime for agg under Clp 1.17.9 primal simplex. Source: COIN-OR Clp benchmark notes for NETLIB instances at this scale (small LPs solve in tens of milliseconds). No runtime executed in this Audit; the second baseline establishes that the verdict does not hinge on a single solver.
binkowski_2026_lp_qipm
Modified Newton system (MNES) formulation across the three instances under Tier A benevolent bound. Cycle figure dominated by agg, the largest of the three. Output extraction via tomography over the full primal-dual vector.
binkowski_2026_lp_qipm
Orthogonal subspaces system (OSS) formulation. Lower condition number than MNES on these instances; cycle bound also dominated by output extraction over the full vector.
lp-qipm-mnes-vs-highs-tier-a Scalar utility output via amplitude estimation. See reformulation_log step 0.
VERDICT SENSITIVITYThe verdict reflects that 5e10 logical cycles, normalized by a benevolent 1ns-per-cycle assumption, produce wall times five orders of magnitude greater than HiGHS at the workload scale.
lp-qipm-oss-vs-highs-tier-a Block-structured solver path. See reformulation_log step 1.
VERDICT SENSITIVITYOSS is the structurally cheaper Newton system on these instances but the verdict at Tier A still rests on output extraction.
lp-qipm-mnes-vs-clp-tier-a Scalar utility output via amplitude estimation.
VERDICT SENSITIVITYlp-qipm-oss-vs-clp-tier-a Block-structured solver path.
VERDICT SENSITIVITYTest the scalar utility output redirect path. Defer full-vector LP/QIPM until logical cycle time falls below 1e-13 seconds at the named precision and the output requirement is reduced from a primal-dual vector to a scalar functional.
Verdicts are bounded by this workload, these baselines, and the stated assumption envelope. They do not generalize to all quantum candidates for this problem class.
output extraction
HiGHS 1.14.0
OSS Newton system with scalar utility output via amplitude estimation
Each entry records what was tested, what was promoted to a route, and why. The log is part of the Manifest body and contributes to the canonical hash.
Reduces output extraction from O(n) tomography to a single amplitude estimation. Documented as a redirect path; not promoted to a route in this demo because the source paper's lower bound assumes full-vector extraction.
Considered as a redirect path for the OSS route. Did not change the Tier A verdict because the dominant term remains output extraction, not Newton-system inversion.
Schema 0.2. Every field validated against the JSON Schema at the runner. The downloads section below provides a clean copy for offline review.
sha256:e11a927f69827d5617ee3ce6a8f0838e057f64f7b4786f9ab8746a554129a813 {
"schema_version": "0.2",
"manifest_id": "DEMO-NETLIB-LP-001",
"audit_tier": "Standard",
"workload": {
"identity": "netlib-lp-portfolio-afiro-adlittle-agg",
"problem_class": "linear_programming",
"inputs": {
"format": "MPS",
"instance_count": 3,
"raw_dimensions": {
"afiro": {
"rows": 27,
"columns": 32,
"nonzeros": 83
},
"adlittle": {
"rows": 56,
"columns": 97,
"nonzeros": 383
},
"agg": {
"rows": 488,
"columns": 163,
"nonzeros": 2410
}
},
"sparsity": 0.0303,
"input_hash": "sha256:08a7065a6cfbf7312eed776d3e0c1c8116c6c4c78c3cc27e5046a94dcbfd8c84"
},
"precision_target": "1e-6 relative",
"operational_constraints": {
"latency": "batch",
"explainability": "required for board memo",
"re_run_frequency": "one_off",
"budget_cycle_context": "ad_hoc"
}
},
"classical_baselines": [
{
"solver": "HiGHS",
"version": "1.14.0",
"configuration": {
"presolve": "on",
"solver": "default_dual_simplex",
"platform": "Apple silicon, single core"
},
"presolve": true,
"runtime_seconds": 0.0022633750340901315,
"optimality_status": "optimal",
"precision_achieved": 1e-9,
"baseline_confidence": "strong",
"accelerator": "CPU",
"notes": "Wall-clock on agg, the largest of the three instances. afiro and adlittle terminated in 0.0217s and 0.000945s respectively. Captured by baselines/run-baseline.py against highspy 1.14.0."
},
{
"solver": "Clp",
"version": "1.17.9",
"configuration": {
"presolve": "on",
"solver": "primal_simplex",
"platform": "single-core CPU"
},
"presolve": true,
"runtime_seconds": 0.05,
"optimality_status": "optimal",
"precision_achieved": 1e-7,
"baseline_confidence": "adequate",
"accelerator": "CPU",
"notes": "Estimated runtime for agg under Clp 1.17.9 primal simplex. Source: COIN-OR Clp benchmark notes for NETLIB instances at this scale (small LPs solve in tens of milliseconds). No runtime executed in this Audit; the second baseline establishes that the verdict does not hinge on a single solver."
}
],
"quantum_candidates": [
{
"algorithm": "hybrid_QIPM",
"formal_model": "binkowski_2026_lp_qipm",
"newton_system": "MNES",
"cost_drivers": {
"sparsity": 0.0303,
"condition_number_lower_bound": 10000,
"qlsa_precision": 0.000001,
"output_extraction": "tomography",
"state_preparation_cost": "block-encoded coefficient matrix per IPM iteration"
},
"assumption_tier": "A",
"cycle_lower_bound": 50000000000,
"notes": "Modified Newton system (MNES) formulation across the three instances under Tier A benevolent bound. Cycle figure dominated by agg, the largest of the three. Output extraction via tomography over the full primal-dual vector."
},
{
"algorithm": "hybrid_QIPM",
"formal_model": "binkowski_2026_lp_qipm",
"newton_system": "OSS",
"cost_drivers": {
"sparsity": 0.0303,
"condition_number_lower_bound": 8000,
"qlsa_precision": 0.000001,
"output_extraction": "tomography",
"state_preparation_cost": "block-encoded augmented system per IPM iteration"
},
"assumption_tier": "A",
"cycle_lower_bound": 30000000000,
"notes": "Orthogonal subspaces system (OSS) formulation. Lower condition number than MNES on these instances; cycle bound also dominated by output extraction over the full vector."
}
],
"assumption_tiers": {
"A": {
"name": "Benevolent lower bound",
"noise": "noise-free",
"cycles_per_oracle_call": 1,
"ipm_iterations": "lower-bound count from Binkowski 2026",
"precision": "favorable",
"condition_estimates": "lower-bound from instance structure"
}
},
"reformulation_log": [
{
"step_index": 0,
"reformulation": "Replace full primal-dual vector output with a single scalar utility (objective value with a chosen linear functional of the optimum).",
"outcome": "tested",
"notes": "Reduces output extraction from O(n) tomography to a single amplitude estimation. Documented as a redirect path; not promoted to a route in this demo because the source paper's lower bound assumes full-vector extraction."
},
{
"step_index": 1,
"reformulation": "Block-structured solver path on the constraint matrix, exploiting sparsity below 0.05 across the three instances.",
"outcome": "tested",
"notes": "Considered as a redirect path for the OSS route. Did not change the Tier A verdict because the dominant term remains output extraction, not Newton-system inversion."
}
],
"results": [
{
"route_id": "lp-qipm-mnes-vs-highs-tier-a",
"candidate": "hybrid_QIPM (MNES)",
"baseline": "HiGHS 1.14.0",
"tier": "A",
"cycle_time_to_runtime_seconds": 50000,
"break_even_cycle_time_seconds": 4.5e-14,
"verdict": "REDIRECT",
"verdict_sensitivity": [
{
"assumption": "output_extraction",
"sensitivity": "verdict_flips_if",
"threshold": "scalar output (amplitude estimation) instead of full primal-dual vector"
},
{
"assumption": "condition_number",
"sensitivity": "verdict_invariant_to",
"threshold": "any kappa within one order of magnitude of 1e4 leaves the verdict at REDIRECT"
}
],
"redirect_path": "Scalar utility output via amplitude estimation. See reformulation_log step 0.",
"notes": "The verdict reflects that 5e10 logical cycles, normalized by a benevolent 1ns-per-cycle assumption, produce wall times five orders of magnitude greater than HiGHS at the workload scale."
},
{
"route_id": "lp-qipm-oss-vs-highs-tier-a",
"candidate": "hybrid_QIPM (OSS)",
"baseline": "HiGHS 1.14.0",
"tier": "A",
"cycle_time_to_runtime_seconds": 30000,
"break_even_cycle_time_seconds": 7.5e-14,
"verdict": "REDIRECT",
"verdict_sensitivity": [
{
"assumption": "output_extraction",
"sensitivity": "verdict_flips_if",
"threshold": "scalar output instead of full primal-dual vector"
},
{
"assumption": "condition_number",
"sensitivity": "verdict_flips_if",
"threshold": "kappa below 1e3 changes the dominant term"
}
],
"redirect_path": "Block-structured solver path. See reformulation_log step 1.",
"notes": "OSS is the structurally cheaper Newton system on these instances but the verdict at Tier A still rests on output extraction."
},
{
"route_id": "lp-qipm-mnes-vs-clp-tier-a",
"candidate": "hybrid_QIPM (MNES)",
"baseline": "Clp 1.17.9",
"tier": "A",
"cycle_time_to_runtime_seconds": 50000,
"break_even_cycle_time_seconds": 1e-12,
"verdict": "REDIRECT",
"verdict_sensitivity": [
{
"assumption": "output_extraction",
"sensitivity": "verdict_flips_if",
"threshold": "scalar output instead of full primal-dual vector"
}
],
"redirect_path": "Scalar utility output via amplitude estimation."
},
{
"route_id": "lp-qipm-oss-vs-clp-tier-a",
"candidate": "hybrid_QIPM (OSS)",
"baseline": "Clp 1.17.9",
"tier": "A",
"cycle_time_to_runtime_seconds": 30000,
"break_even_cycle_time_seconds": 1.7e-12,
"verdict": "REDIRECT",
"verdict_sensitivity": [
{
"assumption": "output_extraction",
"sensitivity": "verdict_flips_if",
"threshold": "scalar output instead of full primal-dual vector"
}
],
"redirect_path": "Block-structured solver path."
}
],
"portfolio_summary": {
"instance_count": 3,
"excluded_under_tier_A": 0,
"redirected": 1,
"monitored": 0,
"passed": 0,
"most_sensitive_assumption": "output_extraction",
"worst_classical_competitor": "HiGHS 1.14.0",
"best_case_quantum_scenario": "OSS Newton system with scalar utility output via amplitude estimation",
"recommended_action": "Test the scalar utility output redirect path. Defer full-vector LP/QIPM until logical cycle time falls below 1e-13 seconds at the named precision and the output requirement is reduced from a primal-dual vector to a scalar functional.",
"forbidden_generalization": "Verdicts are bounded by this workload, these baselines, and the stated assumption envelope. They do not generalize to all quantum candidates for this problem class."
},
"citations": {
"primary_methodology": [
{
"citation_key": "binkowski_2026_lp_qipm",
"full_citation": "Binkowski, Practical lower bounds for hybrid quantum interior point methods in linear programming, arXiv:2604.24362, 2026",
"role": "Source paper for the LP/QIPM lower-bound methodology. DeployQuantum did not author this paper. The Audit operationalizes its decision logic into a customer-facing audit workflow.",
"license": "arXiv abstract page references CC BY 4.0; associated GitHub repository licence reviewed at runner-publication time.",
"applies_to_routes": [
"lp-qipm-mnes-vs-highs-tier-a",
"lp-qipm-oss-vs-highs-tier-a",
"lp-qipm-mnes-vs-clp-tier-a",
"lp-qipm-oss-vs-clp-tier-a"
]
}
],
"independence_statement": "Informed by published lower-bound analysis of hybrid quantum interior point methods. DeployQuantum did not author that research."
},
"sign_off": {
"audit_owner": "DeployQuantum",
"publication_release": "named_with_release"
},
"hash": "sha256:e11a927f69827d5617ee3ce6a8f0838e057f64f7b4786f9ab8746a554129a813"
}The full Manifest as it appears in section VII. Schema 0.2. Hash visible at the top of the page and at the top of the JSON viewer.
DOWNLOAD / MARKDOWNThe two-page board memo derived from the Manifest. Executive verdict, the workload, the baselines, the candidates, the verdicts, what would change them, recommended next steps, citation.
The runner is an open-source Python tool that validates a Manifest against the JSON Schema and the twelve field-level publish blockers. Cloning the repository, fetching the Manifest, and running the validator reproduces the canonical hash byte-for-byte.
The repository at github.com/deployquantum/feasibility-manifest is private until publish-gate condition 8 closes. The runner version that produced this Manifest is 0.2.0; reproducing the hash requires the same runner version (or a newer version that maintains canonicalization invariants), the same canonical-hash policy, and the same input-set hash.
github.com/deployquantum/feasibility-manifest at tag v0.2.0 (post publish-gate close). Download manifest.json from the downloads section.
hash field. Recursively sort keys at every depth. Serialize with no insignificant whitespace, preserve Unicode, no trailing newline. Compute SHA-256. Compare against the hash printed at the top of this page.
Informed by published lower-bound analysis of hybrid quantum interior point methods. DeployQuantum did not author that research.
Each route in the Audit chassis is informed by published research with full citation in every Manifest. Lower-bound exclusion methodology covers hybrid quantum interior point methods, dequantization frameworks, and quantum-optimization landscape reviews. Amplitude estimation cites option-pricing resource estimates, threshold-advantage analysis, and quantum signal processing. Classical-baseline reviews cover quantum simulation, chemistry tensor methods, and the QOBLIB Decathlon benchmark.
The Audit operationalizes published decision logic into a customer-facing audit workflow, with citation per route in every Manifest. The independence statement is captured verbatim in this Manifest at citations.independence_statement.
The Manifest schema is open-sourced under CC BY 4.0. The public-benchmark runner is open-sourced under MIT. The Audit's pipeline is reimplemented independently in DQ-owned code; no third-party code is copied.
For the field-by-field schema reference and the full citation list, see the methodology page.