Part 1 — Results and Jobs (§7)

Q1 — B. RuntimeJobV2 status strings are "INITIALIZING", "QUEUED", "RUNNING", "CANCELLED", "DONE", "ERROR"; the final states are exactly ("DONE", "CANCELLED", "ERROR") (verified in runtime 0.48.0 source). The server-side API status COMPLETED is mapped to "DONE" — "COMPLETED" itself is never returned, and "FINISHED"/"CLOSED" do not exist.

Q2 — A. Results are stored server-side; service.job(job_id) reconstructs the job handle and job.result() fetches the stored result. service.jobs(...) (plural) takes filters (limit, backend_name, pending, session_id), not a single ID. There is no SamplerV2.retrieve.

Q3 — B. job.usage() returns QPU seconds consumed — the number billed against your quota. job.metrics() is the broader dict (timestamps, usage, executions); A gives only timestamps; C and D do not exist.

Q4 — C. measure_all() creates a classical register named meas, and DataBin fields are named after cregs: result[0].data.meas.get_counts(). data.c would be correct only for the default register from QuantumCircuit(n, m). PrimitiveResult has no top-level get_counts() — that was the V1 / backend.run world.

Q5 — B. get_bitstrings() returns one string per shot (1000 entries here). num_shots == 1000, num_bits == 2 (A is backwards); get_counts() returns integer counts, not probabilities; BitArray has no meaningful len() equal to shots.

Q6 — B. ⟨Z⟩ = (n₀ − n₁)/shots = (3072 − 1024)/4096 = 2048/4096 = 0.5. Plus for |0⟩, minus for |1⟩.

Q7 — D. Error ∝ 1/√shots, so shots ∝ 1/error². Going from 0.02 to 0.005 is 4× smaller error → 4² = 16× the shots. This is also why halving precision quadruples QPU time.

Q8 — B. join_data() concatenates the per-register BitArrays into one BitArray (here 2+1 = 3 bits). Verified: with cregs alpha and beta, the later-declared register's bits join to the left (beta+alpha in the bitstring).

Part 2 — Mixed Trap Review

Q9 — A. Pauli strings are little-endian: rightmost character → qubit 0. In 'XZ', Z acts on q0, X on q1. Verified: Pauli('IZ').to_matrix() has diagonal [1, −1, 1, −1] — sign flips with q0's bit, i.e. Z on qubit 0.

Q10 — B. Counts keys are little-endian too: leftmost bit = highest qubit. X on q0 gives q1q0 = '01'. Picking '10' is the classic endianness mistake.

Q11 — B. QuantumCircuit(2, 2) creates the default classical register named c, so the data lives at result[0].data.c. data.meas exists only when measure_all() (or measure_active()) created the meas register. Attribute name always == creg name.

Q12 — B. Transpilation maps virtual qubits onto physical qubits; the observable must be re-indexed to match with obs.apply_layout(isa.layout). Verified on FakeManilaV2: SparsePauliOp('ZZ') became 'IIIZZ' (5 qubits) after apply_layout. Forgetting this either errors (qubit-count mismatch) or silently measures the wrong qubits.

Q13 — A. Sampler samples classical registers, so it requires measurements; Estimator computes ⟨ψ|O|ψ⟩ where the observable defines the "measurement", so the circuit must not contain measurements.

Q14 — C. Since March 2024 / v2 primitives, hardware primitives accept only ISA circuits — already transpiled to the target's basis gates and coupling map. Verified error text: "The instruction h on qubits (0,) is not supported by the target system. Circuits that do not match the target hardware definition are no longer supported…". Fix: generate_preset_pass_manager(optimization_level=..., backend=backend) then pm.run(qc).

Q15 — C. Iteration i+1 depends on iteration i ⇒ you need low-latency sequential access without re-queuing: Session. Batch is for many independent jobs; job mode re-queues every submission.

Q16 — B. Level 0 = nothing; level 1 = default = measurement error mitigation (TREX, with measurement twirling); level 2 = level 1 plus ZNE (and gate twirling). Note the official sample test Q16 dispute: twirling is applied at level 2 as well (Qiskit docs issue #4298) — levels are cumulative. PEC is never enabled by a level; it's opt-in (resilience.pec_mitigation), and Sampler has no resilience levels at all (kills D).

Q17 — B. Dynamic circuits use the context manager: with qc.if_test((clbit_or_creg, value)): — optionally as else_ for the else branch. .c_if() is the removed legacy API; C and D are not real syntax (Python if runs at build time, not run time).

Q18 — B. List binding follows circuit.parameters, which is sorted alphabetically by name, not insertion order: alpha ← 0.5, theta ← 1.5. Verified: a circuit built using beta-then-alpha still reports [Parameter(alpha), Parameter(beta)]. Use a dict to be explicit.

Q19 — B. dumps → string; dump(qc, file) → writes to a file object; loads/load are the import direction. Verified output starts OPENQASM 3.0; include "stdgates.inc"; with bit[n]/qubit[n] declarations and c[0] = measure q[0]; assignments.

Q20 — C. Verified in 0.48.0: setting it raises ValidationError: Object has no attribute 'resilience_level' on SamplerOptions. Mitigation (TREX/ZNE/PEC) reconstructs expectation values, which only makes sense for Estimator; Sampler gets suppression only: dynamical_decoupling and twirling.

Q21 — A. Suppression happens before/during execution at the pulse/circuit level: dynamical decoupling, twirling. Mitigation is classical post-processing of results: TREX (measurement), ZNE (zero-noise extrapolation), PEC (probabilistic error cancellation).

Q22 — B. Verified by instantiation on 0.48.0: options.dynamical_decoupling.enable = True and .sequence_type = "XY4"; valid sequence types are exactly 'XX', 'XpXm', 'XY4' (from the pydantic validation message). The other paths don't exist.

Q23 — A. Little-endian statevector indexing: basis index = binary number q1q0. q0=1 ⇒ index 0b01 = 1 ⇒ [0, 1, 0, 0]. Verified: Statevector([0, 1, 0, 0], dims=(2,2)).

Q24 — C. Estimator PUB order: (circuit, observables[, parameter_values[, precision]]). Sampler PUB order: (circuit[, parameter_values[, shots]]). There is no options.observable.

Scoring: ≥ 18/24 → on track. 14–17 → redo Part 2 misses tomorrow morning. < 14 → fold the weakest section into Day 9 morning before starting papers.