Semantic Composition Engine

QDSV Bridge can expose the public status of server-side composite operation generation through the Semantic Composition Engine (SCE). The SCE runs in the Bridge backend; the SDK submits structured requests and receives candidate evidence packages. The SDK does not contain the private compiler, private operation synthesis rules, deployment secrets, or backend runtime.

Public Endpoints

GET /api/product/composition/capabilities

Returns the SCE version, state model, publication policy and semantic cross-check policy.

POST /api/product/composition/generate

Requests a composite operation candidate from a bounded expression and declared finite domains.

Candidate States

AUTO_GENERATED

Internal transient state before evidence gates.

EXPERIMENTAL

The operation has a candidate package but did not pass every publication gate. Bridge must not expose it as public.

OFFICIAL_CANDIDATE

The operation passed automatic evidence gates and may be exposed by Bridge as a controlled public candidate.

OFFICIAL

Reserved for an explicit governance step. The generator never promotes a candidate to OFFICIAL by itself.

Evidence Required For OFFICIAL_CANDIDATE

An OFFICIAL_CANDIDATE requires both:

  • operation-program verification through the QDSV v2 compiler pipeline;

  • semantic cross-check over the declared finite domain.

The semantic cross-check:

  • enumerates the declared finite domain when it is within the configured limit;

  • evaluates the expression with the logical ProblemSpec reference;

  • evaluates the same expression with the canonical QDSV reference;

  • compiles case fixtures using expected outputs only as verification references;

  • executes the canonical reversible machine and checks result correctness;

  • verifies input preservation and clean ancillas.

If the domain is too large to check exhaustively, or any case fails, the candidate remains EXPERIMENTAL and is not Bridge-public.

No Classical Bypass

The evidence package distinguishes verification from materialization:

reference_answers_used_for_semantic_verification = true

Reference outputs may be used to verify a candidate.

reference_answers_used_for_materialization = false

Reference outputs are not used to build or substitute the realization.

This keeps the public claim narrow: SCE candidates are evidence-backed operation packages, not precomputed answers.

SDK Usage

from qdsv_bridge import QDSVBridgeClient

client = QDSVBridgeClient()

capabilities = client.composition_capabilities()
print(capabilities["semantic_composition_engine"]["version"])

result = client.generate_composition_candidate(
    name="policy_gate_score_v1",
    expression={
        "op": "select_if",
        "args": [
            {"op": "gte", "args": [{"var": "risk"}, 5]},
            {
                "op": "add",
                "args": [
                    {"op": "mul", "args": [{"var": "risk"}, 2]},
                    {"var": "impact"},
                ],
            },
            {"var": "impact"},
        ],
    },
    domains=[
        {"type": "int_range", "variable": "risk", "start": 0, "end": 7},
        {"type": "int_range", "variable": "impact", "start": 0, "end": 7},
    ],
)

print(result["candidate"]["state"])
print(result["evidence"]["semantic_cross_check"]["status"])