The Audit Methodology.
Public. Versioned. Signed by our technical advisor. Hashed onto every page of every report we deliver.
If our audit cannot withstand scrutiny by your engineering team, your CTO, and your board's technical advisors, it should not exist. This page is how we prove it.
Public
Anyone can read this. Anyone can challenge it. We respond.
Versioned
Every revision is hashed. Reports cite the version that produced them.
Signed
A credentialed HPC operations advisor signs the methodology and every report.
15 archetypes. 7 families. One classifier.
Every workload in your audit is evaluated against this library. Each archetype is selected, not invented — anchored to published literature, with rationale and selection criteria documented here.
Optimization
3 archetypesQuadratic and combinatorial cost functions classical solvers struggle to scale at enterprise instance sizes.
Combinatorial QAOA
Selection criteria: Workload signature matches Max-Cut / MIS / weighted scheduling. Job submits MILP solver with O(N²) or worse scaling. QAOA depth-1 mapping documented in [Farhi 2014].
Portfolio QAOA
Selection criteria: Quadratic objective + budget + return constraints. Run cadence aligned with risk-cycle. Amplitude-estimation interaction.
MILP-mapped routing
Selection criteria: Vehicle / network flow / assignment problems. Integer linear program with branch-and-bound timeout indicators in logs.
Simulation
3 archetypesQuantum systems are the cleanest target for quantum hardware. Hamiltonian-shaped workloads admit known polynomial advantages.
VQE molecular Hamiltonian
Selection criteria: Drug-discovery / catalyst-design workloads invoking Hartree-Fock / coupled-cluster solvers. Electron count + basis-set size in plausible window.
Quantum chemistry
Selection criteria: Electronic structure beyond DFT. Run targets correlation energy at chemical accuracy. Active-space partitioning indicators.
QPE eigenvalue
Selection criteria: High-precision spectral problems. Phase estimation depth requirements documented per [Kitaev 1995]. Currently beyond NISQ but well-characterized.
Search
2 archetypesGrover-class problems admit known sqrt(N) speedup. Honest classification means flagging the well-known practical caveats.
Grover unstructured
Selection criteria: Pre-image / SAT-style search with O(N) classical baseline. Practical advantage only above problem sizes that exceed near-term hardware fidelity.
Grover-mapped CSP
Selection criteria: Constraint satisfaction with oracle-style verification. Reported separately from generic CSP solvers because oracle cost can erase the asymptotic advantage.
Linear Algebra
2 archetypesHHL and amplitude estimation admit polynomial speedups under tight preconditions. Reporting both with their caveats.
HHL
Selection criteria: Sparse, well-conditioned linear systems with quantum state-preparation oracle. We flag the well-known reading-out caveat: quantum solution is amplitudes, not classical vectors.
Amplitude Estimation
Selection criteria: Monte Carlo workloads with statistical convergence. Quadratic speedup well-characterized. Hardware horizon driven by gate-fidelity floor.
Number Theory
1 archetypesCryptographic relevance. We report Shor strictly as speculative until hardware credibly threatens to factor enterprise-relevant integer sizes.
Shor's algorithm
Selection criteria: Factoring workloads with cryptographic relevance. Reported as speculative because credible factoring of RSA-2048 requires resource estimates ([Gidney-Ekerå 2019]) beyond any 2030 roadmap.
ML / Hybrid
2 archetypesMost aggressive marketing claims live here. We deliberately downweight pending replicable evidence.
Quantum kernel methods
Selection criteria: Kernel SVMs with quantum feature maps. Reported as "low" confidence: theoretical advantage is dataset-dependent and current empirical results do not generalize.
Variational gradient
Selection criteria: Quantum-assisted training. Reported as speculative until end-to-end benchmark vs. classical optimizers shows reproducible advantage on production-shaped datasets.
Graph / Stochastic
2 archetypesNewer entries to the literature. Reported with appropriate caveats.
Quantum walks
Selection criteria: Graph traversal / centrality. Polynomial speedup under specific oracle assumptions. Reported "low" until oracle cost is bounded.
Stochastic optimization
Selection criteria: Optimization under uncertainty. Mapped to amplitude estimation + QAOA hybrid. Reported "medium" pending hardware horizon.
How a workload becomes a classification.
Rule-based. Auditable. No ML black box.
Ingest scheduler logs
The CLI reads sacct (Slurm), kube-state-metrics (K8s), or qstat (PBS) for every job in the configured window. Raw logs never leave the head node.
Extract signatures
For each job: name pattern, container image, MPI rank count and topology, memory footprint distribution, runtime distribution, submission cadence, queue wait. These are the inputs to classification.
Match against archetype library
Each signature is pattern-matched against every archetype's canonical form. Multiple matches are allowed; the score reflects the cleanness of the strongest match.
Assign confidence band
High: signature unambiguously matches one archetype with clear evidence. Medium: signature matches an archetype but with caveats (e.g., size out of window). Low: signature partially matches; auxiliary evidence required. Insufficient evidence: report explicitly says no archetype assigned.
No quota, no upcharge
When a workload portfolio contains zero quantum-suitable jobs, we report that and the audit price does not change. The Latency Index value comes from the full inventory, including the no-match class.
The Quantum Suitability Score, formally.
Five inputs. Five weights. One score per workload. Reproducible from your scheduler logs and your compute rates alone.
Archetype Match
Cleanness of the fit between the workload signature and the closest archetype. Composite of pattern-match strength (job name, container image, MPI rank distribution), behavioral fit (memory + runtime distribution), and structural fit (call graph patterns vs. archetype canonical form).
Scale: Score saturates at high confidence. Workloads matching 2+ archetypes are split.
Problem Size
Whether the workload instance is in the size window where quantum could plausibly help. Below the window: classical wins trivially. Above the window: no near-term hardware can hold the state.
Scale: Per-archetype window. Outside the window: score floors near zero.
Classical Cost
Annual dollar cost of this workload on classical infrastructure. Computed from customer-supplied rates ($/CPU-hour, $/GPU-hour, $/node-hour) multiplied by observed runtime and queue-wait amortization.
Scale: Logarithmic. $1M/yr scores meaningfully higher than $10K/yr.
Hardware Horizon
When credible hardware is expected to exist for this archetype. Categorical mapping: 2026 = 1.0 (near-term), 2028 = 0.7 (mid-term), 2030+ = 0.4 (far-term), speculative = 0.1 (post-2030 with significant caveats).
Scale: We do not assign Q-quarter precision to any vendor roadmap.
Translation Difficulty
Cost of translating the classical code to a quantum-integrated workflow. Inverse: clean library boundary scores high; deep entanglement with classical control loops scores low.
Scale: Coarse. Three bands: tight (8–10), moderate (4–7), entangled (0–3).
Banding
Why we report horizons in bands, not quarters.
Today's NISQ-era QPUs run with gate fidelity floors above 0.5% per two-qubit gate and calibration that drifts on the order of hours. The scoring rubric reflects this honestly.
What we score against
- Published vendor roadmaps with referenced device generations
- Peer-reviewed resource estimates per archetype
- Gate-fidelity floor required for production-grade execution
- Calibration drift cadence vs. workload duration
What we will not score against
- Vendor marketing claims of imminent advantage
- Unpublished error-correction claims
- Speedups extrapolated past device generations not yet on a roadmap
- Pre-print results that have not been replicated
The honesty guardrail, enumerated.
These rules are enforced by a linter in the report-generation pipeline. A report that violates any of them does not compile.
No specific speedup multipliers on today's NISQ-era hardware
Reports may cite published asymptotic complexity bounds. They will not extrapolate to wall-clock speedups on hardware that does not yet exist.
No promised quantum-advantage delivery dates beyond banded horizons
We commit to bands (2026 / 2028 / 2030+ / speculative). We do not commit to quarters, vendors, or device generations within those bands.
No vendor performance comparisons
We classify workloads against published archetypes. We do not predict relative performance of IBM Heron vs. IonQ Forte vs. neutral-atom platforms.
No ROI projections beyond classical bottleneck cost
Reports state the dollar cost of the workload on classical infrastructure today. They do not project savings that depend on quantum hardware delivering claims we cannot verify.
No forced classifications
When no archetype credibly fits, the workload is labeled "no quantum candidate" and reported. No quota for quantum-applicable findings.
Methodology linter enforces this
The report-generation pipeline flags forbidden phrases ("speedup", "advantage today", "outperform", "100x", "exponential", "unhackable") and refuses to compile a report that contains them outside a properly cited methodology appendix entry.
What leaves your network. What does not.
Reports stay local. Only the sanitized, customer-approved JSON manifest is uploaded — and only after you review it.
| Field | Treatment | Leaves network |
|---|---|---|
| Job names | SHA-256 hashed before any export | Sanitized |
| Container image names | SHA-256 hashed | Sanitized |
| File paths / source code paths | Stripped entirely | Never |
| User IDs / group IDs | Stripped entirely | Never |
| IP addresses / hostnames | Stripped entirely | Never |
| Archetype counters | Aggregated per family | Sanitized |
| Cost vectors | Bucketed (log scale), customer-relative | Sanitized |
| Scheduler type | Reported as one of {slurm, k8s, pbs} | Sanitized |
| CLI version | Reported (audit version tracking) | Sanitized |
--i-have-reviewed-this flag. No silent transmission, ever.The technical advisor.
Methodology, every revision, every audit report — signed.
Methodology v1.0.0
Signed by [Technical Advisor on file]. A credentialed HPC operations veteran with peer-reviewed publications in the operator's vertical.
Names disclosed to enterprise customers during scoping, under mutual NDA. We do not take an engagement without a credentialed advisor signed for the customer's industry.
Advisor role
- Reviews the methodology before publication and on every revision
- Reviews and signs every audit report before delivery
- Available on the debrief call as a credentialed reference
- Can refuse to sign — refusal is reported to the customer with rationale
Hashed onto every page of every report.
Methodology revisions are tracked. Every report pins to the version that produced it.
Each customer audit report carries the version + hash of the methodology that produced it on every PDF page footer.