显式配置的原生调度¶

本例把贝尔电路转换成 L2 原生调度记录，再从 JSON 恢复、认证、投影并关联 L5 输出。所有时长和资源均为教学用的**合成配置**；4、8、10、6 ns 不代表任何真实设备。运行结果是测量事件引用，不是采样值；shot_count 只描述实验请求。

配置必须列出实际展开所需的每个有序 (kind, qubits)。虚拟相位操作也需要配置，时长为零；测量延迟从测量结束时刻起算。这里让两路读出使用各自独占资源。

from qaiji import from_qasm3
from qaiji.core.native import (
    NativeScheduleIR, OperationSpec, QubitResource, ScheduleConfig,
    lower_to_native, qubic_mapping, validate_native_schedule,
)
from qaiji.core.program import (
    ExperimentMetadata, ProgramIR, QuantumInvocation, ResultOutput,
    compute_kernel_ref,
)
from qaiji.core.program_native import NativeBinding, bridge_program_native

circuit = from_qasm3("""OPENQASM 3.0;
include "stdgates.inc";
qubit[2] q;
bit[2] c;
h q[0];
cx q[0], q[1];
c[0] = measure q[0];
c[1] = measure q[1];
""")
config = ScheduleConfig(
    qubit_resources=(
        QubitResource(qubit=0, resource="q0"),
        QubitResource(qubit=1, resource="q1"),
    ),
    operation_specs=(
        OperationSpec(kind="RZ", qubits=(0,), duration_ns=0,
                      resources=("q0",), result_latency_ns=0),
        OperationSpec(kind="RX90", qubits=(0,), duration_ns=4,
                      resources=("q0",), result_latency_ns=0),
        OperationSpec(kind="RZ", qubits=(1,), duration_ns=0,
                      resources=("q1",), result_latency_ns=0),
        OperationSpec(kind="RX90", qubits=(1,), duration_ns=4,
                      resources=("q1",), result_latency_ns=0),
        OperationSpec(kind="CZ", qubits=(0, 1), duration_ns=8,
                      resources=("q0", "q1"), result_latency_ns=0),
        OperationSpec(kind="MEASURE", qubits=(0,), duration_ns=10,
                      resources=("q0",), result_latency_ns=6),
        OperationSpec(kind="MEASURE", qubits=(1,), duration_ns=10,
                      resources=("q1",), result_latency_ns=6),
    ),
    cz_couplings=((0, 1),),
)
schedule = lower_to_native(circuit, config)
restored = NativeScheduleIR.from_json(schedule.to_json())
assert restored == schedule
report = validate_native_schedule(circuit, restored, config)
assert report.valid
assert restored.end_ns == 40
assert report.all_results_ready_ns == 46
projection = qubic_mapping(circuit, restored, config)
assert projection.events == restored.events
assert projection.all_results_ready_ns == max(e.ready_ns for e in restored.events)

kernel_ref = compute_kernel_ref(circuit)
program = ProgramIR(
    quantum_invocations=(QuantumInvocation(kernel_ref),),
    experiment_metadata=ExperimentMetadata(shot_count=100),
    result_outputs=(ResultOutput(0, "c"),),
)
mapping = bridge_program_native(
    program, {kernel_ref: circuit},
    {0: NativeBinding(schedule=restored, config=config)},
)
output = mapping.outputs[0]
assert [(e.invocation_index, e.event_id) for e in output.history] == [
    (0, "m0"), (0, "m1"),
]
assert [(b.bit_index, b.event.event_id) for b in output.final_bits] == [
    (0, "m0"), (1, "m1"),
]

from_json 只检查固定格式版本和内部结构；独立认证还需要原始电路和原始配置。不要根据待认证调度记录的时长反向生成配置。qubic_mapping 和桥接函数 bridge_program_native 会再次认证；先前的 report.valid 不是可复用的授权令牌。

如果将测量后的 Conditional 加入电路，条件寄存器的每一位都必须已经测量，条件区域等待这些位的最新结果就绪；空条件体仍保留区域与屏障。条件只描述寄存器整体相等比较，不执行真实分支。完整限制见原生调度的语义与边界；字段与异常见 L2 原生调度接口、L5 / L2 桥接接口。