Passing target: 24/34 (~69%). Every code-output answer below was verified by executing the snippet on Qiskit 2.5.1 / qiskit-ibm-runtime 0.48 / qiskit-aer.

Answer Table

Q	Ans	Q	Ans	Q	Ans	Q	Ans
1	B	10	D	19	A	28	C
2	D	11	A	20	D	29	D
3	A	12	C	21	C	30	A
4	C	13	B	22	A	31	B
5	D	14	A	23	A, C	32	D
6	B	15	D	24	B	33	B
7	A	16	C	25	D	34	C
8	C	17	B	26	A
9	B	18	C	27	B

Explanations

Q1 — B (4 3). size() counts all four operations (h, cx, two measures). Depth is 3: h → cx → the two measures act on different qubits/clbits, so they share the third layer. Verified: prints 4 3.

Q2 — D. x(1) gives |10⟩; h(0) spreads qubit 0: (|10⟩+|11⟩)/√2; cx(0,1) flips qubit 1 only for the |11⟩ branch → (|10⟩+|01⟩)/√2 = |Ψ⁺⟩. A produces |Φ⁺⟩ = (|00⟩+|11⟩)/√2, B a product state, C the deterministic |11⟩. Verified: probabilities {'01': 0.5, '10': 0.5}.

Q3 — A (None). With inplace=True, assign_parameters mutates qc itself and returns None — a very common source of AttributeError: 'NoneType' bugs. Verified.

Q4 — C. The v2.x standard is generate_preset_pass_manager(optimization_level=..., backend=backend) then pm.run(qc) to obtain an ISA circuit. transpile() without a backend/target cannot produce an ISA circuit, backend.run is the removed V1 path, and qc.transpile does not exist.

Q5 — D (['tdg', 'sdg']). inverse() reverses the instruction order and inverts each gate: (S·T)⁻¹ = T†·S†. Verified.

Q6 — B. Preset pass managers run init → layout → routing → translation → optimization → scheduling. Layout picks physical qubits, routing inserts SWAPs, translation rewrites into basis gates, then optimization cleans up.

Q7 — A. |+⟩ is the +1 eigenstate of X (⟨X⟩ = 1) and lies on the Bloch equator, so ⟨Z⟩ = 0. Verified: prints ≈1.0 0.0.

Q8 — C (['XZ']). a.tensor(b) puts a on the higher qubit(s): X ends up leftmost (qubit 1), Z rightmost (qubit 0) → 'XZ'. expand is the method with the reversed convention. Verified.

Q9 — B. (|00⟩+|01⟩)/√2 factors as |0⟩₁ ⊗ (|0⟩+|1⟩)₀/√2 = |0⟩ ⊗ |+⟩ — a product state. A and D are the Φ± Bell states, C is Ψ⁺; none of those factor.

Q10 — D. S only adds a phase i to the |1⟩ amplitude; phases never change measurement probabilities in the Z basis, so the 50/50 split from H survives. Verified: {'0': 0.5, '1': 0.5}.

Q11 — A. T is a π/4 phase gate; two of them give a π/2 phase — exactly S (and four give Z).

Q12 — C (1.0). X then H sends |0⟩ → |1⟩ → H|1⟩ = |−⟩, so the fidelity with Statevector.from_label('-') is exactly 1. Verified.

Q13 — B. Independent jobs that should be scheduled together with parallel/back-to-back execution are exactly what Batch mode is for. Sessions are for iterative feedback loops and are not automatically "fastest" for independent workloads.

Q14 — A. QiskitRuntimeService.save_account(channel=..., token=...) writes credentials to disk so later QiskitRuntimeService() calls need no arguments. The other three methods don't exist.

Q15 — D. ISA (Instruction Set Architecture) circuit = expressed only in the target backend's native basis gates and obeying its coupling map. It has nothing to do with QASM export, measurements, or a particular optimization level.

Q16 — C. qiskit.primitives.StatevectorSampler implements the same V2 interface (PUBs in, result[0].data.<creg> out) with ideal statevector simulation and no account. qiskit.execute was removed in Qiskit 1.0; AerSimulator.run returns Result/counts, not PUB results.

Q17 — B (2048 for both). Precedence: per-PUB shots > run(..., shots=) > options.default_shots. Neither PUB carries its own shots, so the run-level 2048 applies to both.

Q18 — C ({'10': 500}). Qubit 1 is set to 1; little-endian puts qubit 1 as the second character from the right → '10'. measure_all() creates the meas register, so .data.meas is correct. Verified.

Q19 — A (3 256). A BitArray knows its width (num_bits = 3 classical bits) and its shot count (num_shots = 256). Verified.

Q20 — D. DD is nested under its own options group: options.dynamical_decoupling.enable = True (sequence via .sequence_type). A is Estimator-only, B isn't a real Sampler option path, C assigns True to the whole group instead of its enable field.

Q21 — C. Verified behavior: StatevectorSampler emits a UserWarning ("no output classical registers... result will be empty") and the PUB's DataBin has no fields. The Sampler never invents measurements and never returns amplitudes.

Q22 — A ([1. 1. 0.]). For the Bell state Φ⁺: ⟨ZZ⟩ = 1 (perfect correlation), ⟨XX⟩ = 1, and any single-qubit ⟨Z⟩ = 0 (each marginal is maximally mixed). Verified.

Q23 — A and C. ZNE and TREX operate on the measured data after execution — mitigation. Dynamical decoupling (idle-time pulses) and Pauli twirling (randomized compiling) shape the noise before/during execution — suppression.

Q24 — B (1). EstimatorV2 defaults to resilience_level=1: measurement-error mitigation (TREX). Level 0 disables everything; level 2 adds ZNE. (SamplerV2, by contrast, has no resilience level at all.)

Q25 — D ([ 1. 0. -1.]). One observable broadcasts across the three parameter sets: ⟨Z⟩ = cos θ → cos 0 = 1, cos π/2 = 0, cos π = −1. Broadcasting is legal — that's the point of PUB arrays. Verified.

Q26 — A. H sits on the top wire (q_0), the control dot is on q_0, the ⊕/X target on q_1 → h(0); cx(0, 1). Verified against draw('text') output.

Q27 — B (−X). x sends |0⟩ → |1⟩ (−Z pole); h sends |1⟩ → |−⟩, which sits on the −X axis. (+X would be H|0⟩ = |+⟩.)

Q28 — C. number_to_keep=N keeps the N most frequent outcomes as individual bars and lumps everything else into a single "rest" bar — handy for wide distributions.

Q29 — D. plot_histogram accepts a list of counts dicts and draws them side-by-side with a legend — the standard way to compare ideal vs noisy distributions in one figure.

Q30 — A. job.usage() reports the quantum-seconds consumed (billing-relevant QPU time); job.metrics() includes it among broader execution metadata. The other calls don't exist.

Q31 — B (0.68). ZZ eigenvalue is +1 for 00/11 and −1 for 01/10: (420 + 420 − 80 − 80)/1000 = 0.68.

Q32 — D. Jobs live server-side; service.jobs(limit=3) (on a QiskitRuntimeService instance) lists your recent jobs and can filter by backend, session, or pending state. Kernel restarts lose only the local Python handles, never the jobs.

Q33 — B. x q[1] fixes qubit 1 at 1 (left bit), h q[0] randomizes qubit 0 (right bit): the left bit is always 1, giving ≈500/500 over '10' and '11'. Verified via qasm2.loads + Aer: {'10': 503, '11': 497}.

Q34 — C. OpenQASM 3 adds typed classical data (int, float, bit[n], ...), input/output parameters, and rich classical control flow. Custom gates, barriers, and includes (qelib1.inc / stdgates.inc) exist in both versions.

Distractor Analysis — the 10 Trickiest

Q1 (B). Distractor A (4 4) assumes the two measurements must occupy separate layers; they parallelize because they touch disjoint qubit/clbit pairs. C (3 3) comes from not counting measurements as operations at all — size() counts every instruction.
Q3 (A). Distractor B is the near-universal habit answer: people expect assign_parameters to return the new circuit. With inplace=True the return value is None — assigning it (qc = qc.assign_parameters(..., inplace=True)) silently destroys your reference.
Q5 (D). Distractor A (['sdg', 'tdg']) applies the dagger to each gate but forgets that inverse() also reverses the order; B reverses the order but forgets the daggers. Both halves of (AB)⁻¹ = B⁻¹A⁻¹ are tested at once.
Q8 (C). Distractor A (['ZX']) is exactly what you get if you confuse tensor with expand (or read the result big-endian). Remember: a.tensor(b) = a ⊗ b with a on the left (higher qubits).
Q17 (B). Distractor A reflects the wrong-but-common belief that configured options beat call arguments. The specificity ladder — PUB shots, then run(shots=), then default_shots — is a favorite IBM detail.
Q18 (C). Distractor A ('01') is the big-endian misread: x(1) sets the left character, not the right one. B tempts those who mix up which qubit was flipped, D those who think an unmeasured H is hiding somewhere.
Q21 (C). Distractor B feels right because Runtime hardware paths do reject measurement-free Sampler PUBs, but the question names the local StatevectorSampler, which warns and returns an empty DataBin. D confuses the Sampler with Statevector — primitives never hand back amplitudes.
Q23 (A, C). The classic suppression/mitigation blur: twirling (D) "sounds statistical" and DD (B) is configured beside the mitigation options, but both act on the circuit during execution. The discriminator: does the technique reprocess measured data (mitigation) or alter what runs (suppression)?
Q25 (D). Distractor A is the sign-flip trap (⟨Z⟩ after ry(π) is −1, so the sweep ends at −1, not starts). C preys on broadcasting anxiety — a single observable against N parameter sets is a supported, standard PUB shape.
Q31 (B). Distractor A (0.84) is 2×420/1000 — counting only the correlated outcomes and forgetting the −1 contributions of 01/10. C flips the sign convention. Compute (n₊ − n₋)/shots with eigenvalues per bitstring.