QDSV Bridge Developer Preview

Current package version: 0.4.1.

Documentation site: https://qdsvquantum-afk.github.io/qdsv-bridge/

QDSV Bridge is a lightweight Python client SDK for a controlled semantic-to-circuit bridge built on QDSV - Quantum Declarative Semantic Value.

It turns supported semantic problem specifications into typed QDSV operation graphs, reversible IR, executable QASM/Qiskit circuit materializations, oracle specs or expert construction inputs.

Bridge never labels a scaffold as a completed circuit. Circuit exports are produced through the same canonical QDSV materializer used by the Aer and IBM hardware routes. A circuit request must compile to a reversible_semantic_formula without classical scanning, precomputed candidates or expected-value tables. A standalone goal.predicate_ir can still be returned as expert construction input, but it is not converted into a circuit by classically enumerating its answers.

The public developer preview is available without an API key:

from qdsv_bridge import QDSVBridgeClient

client = QDSVBridgeClient()  # Uses https://api.qdsv.cloud/api

Use a private/local Docker node only when you are running QDSV privately:

client = QDSVBridgeClient.local()  # Uses http://localhost:18080/api

QDSV Bridge is not a template selector and not a free arbitrary circuit generator. Its core rule is:

do not force the problem into a prefabricated circuit;
derive the circuit or construction inputs from the semantic problem specification.

Choose A Delivery Mode

Bridge has one SDK with four delivery modes. Choose the mode by the output you need, not by creating a different kind of project.

Choose how much control you want over the returned artifact. A legacy family value may be retained as a descriptive label, but operation capabilities determine whether a circuit can be generated:

If you need...	Use	Python helper	Main output
A simpler starting point without designing circuits	Bridge Use	client.generate(spec)	Canonically materialized circuit, evidence, guidance and ready-to-run example
An inspectable OpenQASM/Qiskit/Braket-oriented artifact	Bridge Build	client.build(spec)	Executable QASM/Qiskit artifact, oracle spec, canonical IR summary, actual metrics and digests
Expert construction inputs before final circuit materialization	Bridge Expert Prepare	client.prepare(spec)	Semantic construction inputs without forcing a final circuit
Expert comparison of possible materializations	Bridge Expert Evaluate	client.evaluate(spec)	Suggested materialization variants and comparison evidence

If you are not sure where to start, use:

result = client.generate(spec)  # basic

If you want to inspect or route the artifact into Qiskit, Braket or another OpenQASM-oriented workflow, use:

result = client.build(spec)  # intermediate/developer

Expert users have two separate routes: prepare when they want semantic construction ingredients, and evaluate when they want to compare possible materializations.

5 Minute Quickstart

Install the SDK:

pip install qdsv-bridge

For the optional Qiskit artifact inspection path:

pip install "qdsv-bridge[qiskit]"

The Qiskit extra is intentionally version-capped to the current supported major series (qiskit>=2,<3) to follow Qiskit Ecosystem compatibility guidance.

Build a QASM3 artifact from a compact semantic spec. No key is required in the public preview:

from qdsv_bridge import QDSVBridgeClient

client = QDSVBridgeClient()

spec = {
    "state_space": {
        "kind": "finite_candidates",
        "candidate_count": 2,
        "candidate_id": "candidate",
    },
    "signals": ["eligibility_score"],
    "prepared_candidates": [
        {"eligibility_score": 0},
        {"eligibility_score": 1},
    ],
    "goal": {
        "kind": "marking",
        "threshold": 1,
        "criteria": [
            {"signal": "eligibility_score", "influence": 1, "priority": 1}
        ],
    },
    "target": {
        "format": "qasm3",
        "backend_family": "qiskit",
    },
    "limits": {
        "max_qubits": 8,
        "max_depth": 160,
    },
}

result = client.generate(spec)  # basic-user mode

print(result["status"])
print(result["bridge_mode"])
print(result["circuit_origin"])
print(result["artifact"]["format"])
print(result["materialization_evidence"]["formula_materialized_in"])
print(result["materialization_evidence"]["candidate_precomputed"])

For a developer-oriented package with oracle spec, IR summary, digests and preservation evidence:

result = client.build(spec)

print(result["editable_artifacts"]["oracle_spec"])
print(result["editable_artifacts"]["ir_summary"])
print(result["digests"])

Colab Notebooks

Use these notebooks when you want to try Bridge without setting up a local project:

Notebook	Flow
Open 01 Semantic Candidate Marking	bounded_semantic_marking -> QASM3 artifact -> Bridge Report
Open 02 Predicate Construction Inputs	predicate_marking -> expert construction inputs -> Bridge Report
Open 03 Semantic Signal Classification	semantic_signal_classification -> QASM3 artifact -> Bridge Report
Open 04 IBM/Qiskit Artifact Demo	Problem-first spec -> QASM3 artifact -> Qiskit inspection/simulation -> Bridge Report
Open 05 AWS Braket OpenQASM Demo	Problem-first spec -> OpenQASM artifact -> Braket LocalSimulator -> Bridge Report

Each notebook follows the same route:

problem spec -> Bridge artifact -> shareable report

IBM/Qiskit Workflow

For Qiskit-oriented users, see the public integration guide:

• QDSV Bridge for IBM/Qiskit Workflows
• Open 04 IBM/Qiskit Artifact Demo

Amazon Braket OpenQASM Workflow

For Braket-oriented users, see the public OpenQASM integration guide:

• QDSV Bridge for Amazon Braket OpenQASM Workflows
• Amazon Braket Community Submission Path
• Open 05 AWS Braket OpenQASM Demo

QDSV Bridge exports OpenQASM artifacts and includes to_braket_openqasm() for the bounded gate-name/definition conversion used by the tested Amazon Braket LocalSimulator workflow. This is not an official Amazon Braket integration and does not include managed AWS hardware execution.

OpenQASM-First Interoperability

For a provider-neutral view of the Bridge artifact handoff, see:

• QDSV Bridge OpenQASM-First Interoperability

Bridge uses OpenQASM as a public artifact boundary between controlled semantic problem specifications and framework-specific workflows such as Qiskit or Amazon Braket.

Default cloud endpoint:

https://api.qdsv.cloud/api

Private/local Docker endpoint:

client = QDSVBridgeClient.local()

What This Is

• A client SDK for QDSV Bridge.
• A controlled interface to the QDSV Operation Compiler and canonical circuit materializer.
• A way to request QDSV IR, oracle specs, QASM/Qiskit artifacts or expert construction inputs.
• A way to generate shareable Bridge Reports in JSON, Markdown or HTML.
• A bridge for users who need circuit ecosystems but do not want to start from prebuilt templates.

What This Is Not

• It is not the private QDSV Runtime.
• It is not a bulk data processing SDK.
• It is not a hardware execution SDK.
• It is not an arbitrary circuit generator.
• It is not a selector of prefabricated circuit templates.
• It does not expose CAP, backend selection heuristics, private lowering internals, QuEST/Aer/IBM adapters, secrets or production configuration.

Why This Exists

Traditional quantum workflows often start by asking users to choose a circuit, encoding, ansatz or measurement pattern. That can force the data to adapt to a circuit.

QDSV Bridge starts from a bounded semantic problem. The QDSV Operation Compiler builds a typed operation graph and checks every required operation before any circuit is claimed.

semantic problem spec
-> typed operation graph
-> reversible IR and capability validation
-> oracle specification
-> materialization policy
-> generated circuit artifact or expert construction inputs

For users who want a simpler starting point, Bridge can generate executable problem-derived circuits when the specification compiles completely to a reversible semantic formula. It refuses to present incomplete blueprints or classically enumerated answer oracles as completed circuits.

For expert constructors, Bridge can also return the key semantic inputs needed to design a custom circuit without forcing a final circuit.

Bridge Modes

QDSV Bridge is one SDK with multiple output depths. The modes do not change the core principle: a circuit is delivered only after canonical QDSV ProblemSpec/IR materialization succeeds.

API	Commercial name	Intended user	What Bridge returns
generate() / use	Bridge Use	Basic user who wants to solve without designing circuits.	Canonical materialized circuit, explanation, materialization evidence, actual metrics and ready-to-run example.
build() / build	Bridge Build	Intermediate user who understands Qiskit, QASM or quantum workflows.	Canonical materialized circuit plus executable QASM/Qiskit, oracle spec, IR summary, actual qubits/depth and digests.
prepare() / expert_prepare	Bridge Expert Prepare	Expert constructor who wants to design a custom circuit.	Validated family, ProblemSpec/IR, oracle spec, predicates, target state/goal, constraints, variables, information-loss risk, encoding/measurement suggestions, estimated limits and evidence. It does not force a final circuit.
evaluate() / expert_evaluate	Bridge Expert Evaluate	Expert evaluator who wants to compare QDSV materializations.	Canonical QDSV circuit artifact, materialization variants, resource comparison and reproducible evidence.

Python helpers:

client.generate(spec)  # mode="use"
client.build(spec)     # mode="build"
client.prepare(spec)   # mode="expert_prepare"
client.evaluate(spec)  # mode="expert_evaluate"

Explicit mode:

client.export(spec, mode="build")
client.explain(spec, mode="expert_prepare")

CLI:

qdsv-bridge export spec.json --mode use
qdsv-bridge export spec.json --mode build
qdsv-bridge export spec.json --mode expert_prepare
qdsv-bridge export spec.json --mode expert_evaluate
qdsv-bridge report spec.json --mode build --format markdown --output bridge_report.md

Every export response should include traceability metadata:

{
  "mode": "use | build | expert_prepare | expert_evaluate",
  "artifact_type": "qasm3",
  "circuit_origin": "qdsv_canonical_problem_ir_materializer",
  "materialization_evidence": {
    "formula_materialized_in": "qpu_circuit",
    "classical_scan": false,
    "candidate_precomputed": false,
    "answer_leakage": false,
    "precomputed_result_table": false
  },
  "warnings": [],
  "digests": {}
}

Examples By User Type

Basic user:

result = client.generate(spec)
print(result["circuit_origin"])          # qdsv_canonical_problem_ir_materializer
print(result["circuit"]["status"])       # materialized_from_canonical_problem_ir
print(result["ready_to_run_example"])

Intermediate developer:

result = client.build(spec)
print(result["artifact"]["format"])
print(result["editable_artifacts"]["oracle_spec"])
print(result["editable_artifacts"]["ir_summary"])
print(result["digests"])

Expert constructor:

inputs = client.prepare(spec)
print(inputs["expert_inputs"]["encoding_suggestions"])
print(inputs["expert_inputs"]["measurement_suggestions"])

Expert evaluator:

comparison = client.evaluate(spec)
print(comparison["materialization_variants"])
print(comparison["comparison"])

Operation Capabilities

Materialization is selected from the typed operation graph, not from a problem-family name. Query the current catalog with:

catalog = client.capabilities()
print(catalog["operation_capabilities"])

client.capabilities() uses /api/bridge/capabilities. The previous client.families() helper remains only as a compatibility alias; family labels do not select a compiler, set physical capacity or promise circuit delivery.

The compiler v1 executable slice supports bounded prepared numeric data and predicates composed from:

• fields and constants;
• addition and subtraction;
• multiplication and division within the certified bounded profile;
• absolute and squared differences plus bounded similarity expressions;
• eq, ne, lt, lte, gt and gte decisions;
• ScoreModel v2 flat and hierarchical importance-priority aggregation;
• signed contextual adjustments, normalization and penalties;
• reversible formula and decision operators with explicit uncompute evidence.

ScoreModel v2 Decision Capability

ScoreModel v2 is the broadest decision-oriented circuit path currently exposed through Bridge. It is not tied to one industry or to similarity alone. It materializes bounded multi-criteria decision predicates over finite candidates through the canonical QDSV operation compiler.

ScoreModel v2 accepts two complementary value routes:

Value route	What the user provides	Typical sources
Prepared numeric metric	Any finite numeric value that fits the declared numeric contract, together with optional provenance	Probability, normalized distance, risk metric, correlation, externally calculated similarity, model output or domain measurement
QDSV numeric expression	Prepared numeric fields plus a bounded operation graph supported by the ScoreModel v2 physical profile	Direct values, bounded arithmetic, absolute or squared difference and scalar numeric similarity

An externally produced metric remains an input: Bridge does not claim to have computed it. This lets users bring a value from another model or domain process while QDSV materializes the bounded decision operation. QDSV can also compute scalar numeric similarity inside the circuit profile. Arbitrary vector or cosine similarity is not currently computed by this physical profile, but a separately computed finite similarity value can be supplied as a prepared numeric metric.

The implemented ScoreModel v2 capability includes:

• flat multi-term decisions;
• hierarchical decisions with independent importance and priority at each block;
• non-negative term and block importance and priority values;
• signed contextual adjustments over bounded adjustment values;
• term aggregation, normalization and zero-mass protection;
• term-level, block-level and global penalty handling;
• exact rational semantic evaluation followed by declared fixed-point output, deterministic rounding and overflow rejection;
• eq, ne, lt, lte, gt and gte threshold decisions;
• bounded value expressions using fields, constants, addition, subtraction, multiplication, division, safe division, modulo, absolute value, absolute difference, squared difference and scalar numeric similarity when accepted by the current capability assessment;
• candidate-independent reversible formula synthesis, decision marking, measurement and explicit uncompute evidence;
• optional bounded Grover amplification without inferring the number of winning candidates through a classical answer scan.

Public ScoreModel specifications use user-oriented decision names:

term = {
    "value": prepared_metric,
    "importance": 2,
    "priority": 3,
}

importance expresses how much the factor contributes to the decision model. priority expresses its urgency, severity or decision precedence. The legacy names weight and criticality remain accepted for compatibility, but new SDK code and public Bridge responses use importance and priority. QDSV keeps any canonical mathematical naming internal to the operation compiler.

This supports practical finite-candidate workflows such as:

• eligibility, approval and compliance screening;
• multi-criteria candidate selection by threshold;
• risk-benefit and cost-value assessment;
• project, supplier, customer, hypothesis or alternative evaluation;
• contextual decisions where the same observed value must be adjusted by declared circumstances;
• hierarchical decisions that combine local block assessments into one global acceptance predicate.

These are threshold-based decision and selection workflows. The current physical profile does not claim complete ranking, Top-K, argmax/argmin, unrestricted optimization, arbitrary vector similarity or automatic model calibration.

ScoreModel v2 is available through a canonical problem_spec. Bridge returns the materialized circuit and a public evidence passport, but not the private lowering, bounded function rows, candidate-score tables or precomputed answers. The public evidence states whether the formula was materialized in the circuit, whether candidate answers were precomputed, which lowering profile was used and the actual qubit and depth metrics.

The current physical synthesis is bounded by the declared max_input_qubits and max_function_states, as well as Bridge payload, QASM, qubit and depth limits. Resource cost can grow rapidly with the number and precision of prepared fields. The flat and hierarchical profiles are verified on Aer and QuEST; ScoreModel v2 IBM hardware execution remains preflight-ready but pending a dedicated hardware run. See examples/score_model_v2.py for a bounded end-to-end example.

Other Problem IR operations can still be represented semantically. If any graph node lacks a certified recursive lowering, Bridge returns exact missing_capabilities for expert construction instead of claiming a circuit.

Legacy family values such as bounded_semantic_marking remain accepted as compatibility labels. They do not choose the compiler, alter the generated circuit or promise support.

Compatibility example:

spec = {
    "family": "bounded_semantic_marking",
    "state_space": {
        "kind": "finite_candidates",
        "candidate_count": 2,
        "candidate_id": "candidate",
    },
    "signals": ["eligibility_score"],
    "prepared_candidates": [
        {"eligibility_score": 0},
        {"eligibility_score": 1},
    ],
    "goal": {
        "kind": "marking",
        "threshold": 1,
        "criteria": [
            {"signal": "eligibility_score", "influence": 1, "priority": 1}
        ],
    },
    "target": {
        "format": "qasm3",
        "backend_family": "qiskit",
    },
    "limits": {
        "max_qubits": 8,
        "max_depth": 160,
    },
}

result = client.build(spec)
print(result["family"])
print(result["circuit_origin"])
print(result["warnings"])

What It Exports

Depending on target.format, QDSV Bridge can export:

• problem_spec
• ir
• oracle_spec
• qasm2
• qasm3
• qiskit_blueprint

Bridge Report

Bridge Report turns the same auditable export package into a document users can share, attach to technical notes or keep as reproducibility evidence.

Supported report formats:

• json
• markdown
• html

Python:

report = client.report(spec, mode="build", format="markdown")
print(report["content"])

CLI:

qdsv-bridge report spec.json --mode build --format markdown --output bridge_report.md
qdsv-bridge report spec.json --mode build --format html --output bridge_report.html
qdsv-bridge report spec.json --mode build --format json --output bridge_report.json

Each Bridge Report includes:

1. Problem and deliverable contract.
2. Family and mode used.
3. Validations and canonical materialization evidence.
4. Generated executable QASM/Qiskit content when requested.
5. Circuit policy and template-use status.
6. Digests.
7. Resource limits and warnings.
8. Reproducibility metadata.

Bridge Report is the public trust document for Bridge: it makes clear what was derived, what was preserved, what was warned about and which digests identify the exported artifacts.

For use and build, circuit-oriented targets return a circuit only when canonical QDSV materialization succeeds, together with formula/oracle digests and actual circuit metrics.

Bridge does not insert placeholder oracle comments and call them completed circuits. Incomplete specifications must use expert_prepare or add computable semantic inputs.

For expert_prepare, Bridge may intentionally return construction inputs instead of forcing a final circuit. This is useful when an expert wants to design a custom circuit with the correct semantic ingredients.

Public Endpoints

The SDK calls the QDSV API:

• GET /api/bridge/families
• POST /api/bridge/validate
• POST /api/bridge/compile
• POST /api/bridge/explain
• POST /api/bridge/export
• POST /api/bridge/report

Default cloud endpoint:

client = QDSVBridgeClient(api_url="https://api.qdsv.cloud/api")

Local/private Docker endpoint:

client = QDSVBridgeClient.local()

client.capabilities() is the recommended name for the informational catalog. families() and /bridge/families remain compatibility aliases. Developer Preview value-producing operations can be used without an API key and are limited by rate limits, payload limits and a demo quota per caller bucket.

Resource And Multi-User Limits

QDSV Bridge accepts compact semantic problem specifications and small prepared numeric signal tables, not raw datasets.

Do not send:

• full CSV files;
• raw datasets;
• thousands of rows;
• sensitive records;
• training data;
• hardware execution requests.

Send:

• semantic problem_spec or bounded state space;
• finite state-space size;
• prepared signal or variable names;
• goal / predicate / ranking intent;
• target format;
• qubit/depth limits;
• materialization policy.
• compact prepared_candidates containing only the bounded numeric signals required by the formula, when requesting a coherent circuit;
• or a canonical problem_spec that the Operation Compiler can certify as reversible_semantic_formula.
• standalone predicates may be sent to expert_prepare; they are not synthesized into circuits from precomputed satisfying states.

Public API minimum limits:

Limit	Default
Max spec payload	64 KB
Max QASM / circuit artifact payload	256 KB
Max compile time	5 seconds
Max export time	10 seconds
API key required for	Not required by default in Developer Preview
Monthly demo quota	100 compilations/exports/reports per caller bucket
Raw data payloads	Not allowed; compact prepared numeric signals are allowed
Hardware execution	Not available from Bridge SDK
Semantic service boundary	4096 finite candidates / 64 named signals
Compiler v1 coherent boundary	16 candidates / 20 prepared-signal qubits
Default requested circuit ceiling	256 qubits / depth 1,000,000

The semantic service boundary is not a circuit-size promise. Actual circuit resources depend on the complete operation graph, precision, registers, reversible arithmetic, ancillas and uncompute.

Default API rate limits may be configured by the QDSV deployment:

• families: 60/minute
• validate: 30/minute
• compile: 20/minute
• explain: 20/minute
• export: 10/minute
• report: 10/minute

Rate limits use Authorization, x-api-key or x-license-key when present. Otherwise they fall back to IP plus SDK name.

API keys are optional for the open Developer Preview path. They become useful for higher quotas, partner access, private pilots, usage attribution or deployments configured with QDSV_BRIDGE_API_KEY_REQUIRED=true.

In the current public deployment, rate limiting is in-memory per API instance. If Cloud Run scales to multiple instances, the effective limit may multiply until a distributed rate-limit store is enabled.

The SDK itself is stateless and does not block multiple users. Multi-user control, quotas, optional API keys, license checks and history belong to the QDSV/Qruba API deployment, not to the local Python package.

Raw dataset payloads are rejected:

{
  "detail": {
    "error_code": "E_BRIDGE_RAW_DATA_NOT_ALLOWED",
    "message": "Raw datasets are not allowed in Bridge specs. Provide semantic signals, predicates, candidates, or summarized problem structure instead.",
    "detail": {
      "forbidden_fields": ["rows"]
    }
  }
}

Examples And Roadmap

• Examples
• Roadmap
• QDSV model site
• Qruba Cloud
• PyPI

Current public artifact targets are problem_spec, ir, oracle_spec, qasm2, qasm3 and qiskit_blueprint.

PennyLane, Azure Quantum/Q#, Cirq, VS Code, TypeScript clients, Amazon Braket managed hardware execution and packaged NISQ recipes such as QUBO/QAOA/VQE/Grover are planned integrations or higher-level examples. They should not be described as current public Bridge capabilities.

Important Boundaries

QDSV Bridge does not expose:

• QDSV Runtime internals;
• CAP internals;
• backend selection heuristics;
• private QuEST/Aer/IBM adapters;
• native lowering internals;
• unrestricted circuit generation;
• arbitrary Python execution;
• prefabricated circuit selection as the main product;
• user-supplied free circuits as family contracts.

Product principle:

semantic problem specification
-> typed operation graph
-> reversible IR and capability assessment
-> executable circuit materialization or expert inputs
-> auditable export

Not:

any input -> arbitrary circuit

Open SDK, Private Runtime

This repository is open-core:

• Open under MIT: Python client SDK, examples, docs and tests.
• Not included: QDSV Runtime, ontology compiler internals, CAP, lowering, advanced materialization, backend adapters, private endpoints, secrets or production configuration.

QDSV, QIntent and Qruba names and marks are project marks of their respective owners. The MIT License for this repository does not grant trademark rights.