Quantum Data Lineage, Catalogs, and Quality Gates

Quantum data products are not raw counts. They are structured artifacts with lineage: problem instance, circuit generation, compilation target, execution baseline, result counts, mitigation model, statistical analysis, and reviewed claim. Treating those artifacts as anonymous files destroys reproducibility.

OpenLineage models job, run, and dataset metadata with extensible facets [R216]. A quantum data catalog can use the same pattern: attach quantum-specific facets to experiments, circuits, calibration snapshots, result datasets, and claims.

Diagram loads as you read

Quantum Data Lineage, Catalogs, and Quality Gates · Figure 1

View diagram source

flowchart LR
    Problem[Problem instance] --> Circuit[Circuit dataset]
    Circuit --> Compiled[Compiled circuit]
    Compiled --> Execution[Execution dataset]
    Execution --> Counts[Counts dataset]
    Counts --> Mitigated[Mitigated estimate]
    Mitigated --> Claim[Claim dataset]

Dataset classes

Quantum platforms should distinguish dataset classes.

Dataset class	Contents	Quality gate
problem instance	molecule, graph, Hamiltonian, objective	schema and domain validation
circuit artifact	logical circuit and parameterization	deterministic hash
compiled artifact	target-bound circuit	compiler target compatibility
execution dataset	job metadata and raw counts	shot completeness and baseline binding
mitigation dataset	transformed counts or estimates	mitigation profile and uncertainty
claim dataset	conclusion plus evidence	review and retention policy

Diagram loads as you read

Dataset classes · Figure 2

View diagram source

flowchart TB
    Catalog[Quantum data catalog] --> Problem[Problem instances]
    Catalog --> Circuits[Circuit artifacts]
    Catalog --> Compiled[Compiled artifacts]
    Catalog --> Counts[Raw counts]
    Catalog --> Mitigated[Mitigated estimates]
    Catalog --> Claims[Claims]

Lineage graph

Lineage should answer what changed between two results.

Diagram loads as you read

Lineage graph · Figure 3

View diagram source

flowchart TB
    P1[Problem v1] --> C1[Circuit v1]
    C1 --> T1[Target A]
    T1 --> R1[Run 1]
    R1 --> M1[Mitigation v1]
    M1 --> Claim1[Claim A]
    C1 --> T2[Target B]
    T2 --> R2[Run 2]
    R2 --> M2[Mitigation v2]
    M2 --> Claim2[Claim B]

If two claims differ, the catalog should reveal whether the difference came from problem encoding, circuit synthesis, compiler target, hardware baseline, shot policy, mitigation, or statistical interpretation.

Quality gates

Data quality gates should run before artifacts move between classes.

Diagram loads as you read

Quality gates · Figure 4

View diagram source

flowchart LR
    Artifact[Artifact] --> Schema[Schema validation]
    Schema --> Hash[Hash check]
    Hash --> Completeness[Completeness check]
    Completeness --> Physics[Physics sanity checks]
    Physics --> Policy[Policy check]
    Policy --> Publish[Publish to catalog]

Physics sanity checks include impossible counts, missing classical registers, unexpected shot totals, invalid parameter bounds, stale calibration references, and inconsistent basis-gate declarations.

A catalog facet is a small metadata object attached to a run, job, or dataset.

Illustrative listing · yaml

quantum_execution_facet:
  qpu_id: qpu-redacted-17
  backend_family: superconducting-transmon
  calibration_baseline_id: cal-2026-04-20T0900Z
  compiler_target_id: target-2026-04-20T0940Z
  circuit_hash: sha256:4e7c...
  shots_requested: 10000
  shots_completed: 10000
  mitigation_profile: measurement-mitigation-v4
  evidence_id: qev-2026-04-20-219

Diagram loads as you read

Catalog facets · Figure 5

View diagram source

flowchart TB
    Run[OpenLineage-style run] --> Facet1[Quantum execution facet]
    Run --> Facet2[Calibration facet]
    Run --> Facet3[Compiler facet]
    Run --> Facet4[Mitigation facet]
    Run --> Facet5[Evidence facet]

Lineage-aware deletion

Quantum data retention is not uniform. Some artifacts must be kept for audit; some sensitive inputs must be minimized.

Diagram loads as you read

Lineage-aware deletion · Figure 6

View diagram source

flowchart LR
    Dataset[Dataset] --> Retention{Retention class}
    Retention -- ephemeral --> TTL[Delete after TTL]
    Retention -- reproducibility --> Archive[Archive with hash]
    Retention -- claim-bearing --> Preserve[Preserve evidence chain]
    Retention -- sensitive --> Minimize[Redact or tokenize]

Lineage lets the platform delete safely. If a dataset supports a public claim, it cannot be silently removed without preserving the chain needed to verify the claim.

Ownership

Every quantum data product needs an owner.

Diagram loads as you read

Ownership · Figure 7

View diagram source

flowchart TB
    DataProduct[Quantum data product] --> Owner[Accountable owner]
    DataProduct --> Schema[Schema steward]
    DataProduct --> Producer[Producer system]
    DataProduct --> Consumer[Consumer systems]
    DataProduct --> SLA[Quality SLA]
    DataProduct --> Retention[Retention rule]

The owner is responsible for schema evolution, quality gates, lineage completeness, access policy, and deprecation.