Circuit corpus reference

All 17 circuits currently registered in qloop/circuits/. “Stages” lists which of the six pipeline stages actually run for that circuit (based on which optional CircuitSpec methods it implements) — everything not listed skips visibly rather than silently passing.

Base circuits (pre-framework)

Name	Qubits	Stages	Notes
bell	2	2, 3, 4, 5	Bell state; the original minimal example
grover	2	3, 4	2-qubit Grover search, marked bitstring swept via param_space
vqe	2	2, 3, 4	Hardware-efficient ansatz against a toy 2-qubit Hamiltonian

Tier 0 — seed benchmarks

Airtight closed-form oracles, chosen to validate all six stages end to end before later tiers remove that safety net.

Name	Qubits	Stages	Source
ghz	12	2, 3, 4	CNOT-ladder GHZ; textbook
ghz-tree	12	2, 3, 4	Log-depth CNOT-tree GHZ, contrasted against ghz; textbook
qft	4	2, 3, 4	Qiskit’s QFTGate; round-trip-identity and add-1-in-Fourier-basis oracles
tfim-trotter	4	2 (tolerance-band), 3, 4	Trotterized transverse-field Ising evolution; textbook
mermin-bell	3	2, 3, 4	GHZ + Mermin operator, exact quantum violation of the classical bound

Tier 1 — Clifford QEC

Name	Qubits	Stages	Source
bb-code-72	144	3b (Stim), 4	Bravyi, Cross, Gambetta, Maslov, Rall, Yoder, Nature 627, 778 (2024), arXiv:2308.07915 — [[72,12,6]] bivariate-bicycle code

Tier 2 — non-Clifford exact oracles

Name	Qubits	Stages	Source
color-832-ccz	8	2, 3, 4	arXiv:2309.08663 — [[8,3,2]] color code transversal CCZ; logical operators derived from scratch via GF(2) linear algebra
dicke	6	2, 3, 4	Motivated by Yuan & Zhang, arXiv:2505.15413 — implements a simpler, independently-verifiable hypergeometric-recursion construction rather than that paper’s specific algorithm
gqsp	1	3 (via invariants_for), 4	Motivated by Motlagh & Wiebe GQSP, arXiv:2308.01501 — implements classic single-qubit QSP (a strict special case) with phases derived numerically, not reproduced from the paper

Tier 3 — topology + statistics

Name	Qubits	Stages	Source
qaoa-maxcut	6	2, 3, 4, 5	Motivated by PHOENIX, arXiv:2504.03529 — standard p=1 QAOA MaxCut on K₃,₃ (non-local graph)
qaoa-ring	6	2, 3, 4, 5	Motivated by arXiv:2509.17296 — same ansatz on a 6-cycle (local graph), contrasted against qaoa-maxcut
quantum-volume	4	3, 4, 5	Cross et al., Phys. Rev. A 100, 032328 (2019) — fixed-seed random-circuit model; heavy-output criterion checked as an invariant, not against an external oracle

Tier 4 — hard tier

Name	Qubits	Stages	Source
bb-code-144	288	3b (Stim), 4	Same source as bb-code-72, scaled to the [[144,12,12]] gross code (l=12, m=6)
magic-cultivation	3	3, 4	Motivated by Gidney, Shutty & Jones, arXiv:2409.17595 — implements a genuine (but much simpler) postselected stabilizer-based magic-state verification, not the paper’s surface-code cultivation protocol

Where “stages” comes from

Read directly off which optional CircuitSpec methods each circuit implements — see Architecture for the mapping. To check any circuit’s actual behavior yourself:

from qloop.core.registry import registry

spec = registry.get("dicke")
print("exact tier:", spec.reference_state() is not None)
print("noisy tier:", spec.expected_distribution() is not None)
print("stim tier:", spec.stim_program() is not None)