Core Concepts
A short glossary of the ideas the rest of the documentation builds on.
Circuit, KIR, operation
A circuit is an ordered list of operations over numbered qubits and classical bits. An operation is a named gate, a unitary given as an explicit matrix (up to five qubits), or a measurement. See KIR.
Job and output request
A job is a circuit plus what you want out of it: measurement counts (shots > 0), the exact statevector, or an expectation value of an observable. A job must request at least one output. The request shapes the plan: only the dense engine can produce a statevector, and only some engines can produce an expectation value.
Engine and backend
An engine is an algorithm (dense, stabilizer, Pauli propagation, MPS, tensor network). A backend is the hardware it runs on (CPU or CUDA). They are chosen separately: the algorithm by work units, then the backend by calibrated seconds.
Plan, strategy, region
A plan is the fully decided way to run a job. Its strategy is one of direct_dense, direct_stabilizer, direct_pauli, direct_mps, direct_tensor_network, factorized, or cut_and_recombine. A region is a set of qubits run together on one engine. A direct plan has one region; a factorized or cut plan has several.
Candidate
Every strategy the planner considered, feasible or not. The plan report lists all of them. See Plan Explanation.
Work unit
The cost model's shared currency. A work unit is comparable across engines, which is what lets the planner choose between them. It is not seconds; seconds come from calibration. See Cost Model.
Feasible and infeasible
A candidate is infeasible if it cannot serve the request: it would exceed the memory budget, it does not support a gate or an output, or it would exceed the time budget. Infeasible candidates carry a reason string.
Budget
A constraint on a run. A memory budget bounds the bytes a plan may need. A time budget bounds the wall-clock seconds of one execution attempt. Both are checked at plan time; the time budget is also checked during execution. See Memory Constraints and Time Budgets.
Exact
A run is exact if no approximation was made. Every engine reachable from Python is exact, and Result.exact reports it. See Exactness.
Replan
When a run overruns its time budget it is stopped and planned again using what was observed. The abandoned attempt is reported. See Replanning.
Deterministic
The same circuit, options and seed always produce the same counts and the same plan. See Determinism.