Execute
The runtime executes a plan. It resolves each region to an engine and backend, runs it, and recombines the regional results.
Direct plans
A direct_* plan has one region covering every qubit. The runtime gives that engine the region's operations, the output request and the seed, and returns its result. Before running it computes which measured qubit feeds which classical bit, so that a classical bit written more than once takes its last write.
Factorized plans
A factorized plan has one region per group of non-interacting qubits, each already assigned its own cheapest feasible engine. Regions run one at a time, and each result is folded into the answer immediately:
| Output | Recombination |
|---|---|
| Statevector | Exact tensor product of the regional states, built into one global buffer with one region live at a time |
| Counts | Independent per-shot draws from each region's outcome distribution, combined bit-wise. Exact: the regions are independent |
| Expectation | For each observable term, the product of the regional values, then the coefficient-weighted sum (the mixed-product property) |
A region with no operations and no measurement is an idle region: a qubit that stays |0⟩. It needs no allocation and no engine.
Global phase
A circuit's global_phase multiplies the whole statevector. For a factorized plan it is applied once, as a scalar over the recombined buffer, never per region. It cancels in counts and in expectation values, so it affects only the statevector.
Cut plans
A cut_and_recombine plan runs each fragment once for every distinct assignment of its cut wires, on the cheapest engine for that fragment, and reconstructs the expectation value exactly from a signed, weighted sum. See Cut and Recombine.
Invariants are checked
Before executing a factorized plan the runtime verifies that its regions are consistent with the circuit, and fails loudly if not. There is no silent fallback to a different plan.
Timing and memory
The result records execute_seconds and sample_seconds of the attempt that succeeded, and actual_state_bytes, which is a genuine measurement (for a factorized run, the peak over regions plus the global buffer) rather than a copy of the estimate.
With a time budget
If the job carries a time budget, the same engines report safe boundaries to a monitor while running. See Monitoring.