Coherent

Circuit-level (non-Pauli) tensor decoder

8/8 passed in 33187ms

Test	What it does	Result
test_beats_trivial_predictor	On sampled coherent-RX syndromes the logical error rate is far below the always-zero predictor's.	✅ pass
test_coherent_surface_builds_at_d3	A d=3 rotated surface memory under RZ builds without blowing up tensor degree, in chi mode: the Z branch flips the undeclared X-checks, which dephase it.	✅ pass
test_damping_no_flip_weight	A damping location's weight on flipping nothing declared is the chance it does nothing visible: 1 on a repetition memory in |0>, 1 - p in |1>, and 1 - p/2 on a surface Z-memory whose readout tells the Z branch apart.	✅ pass
test_knobs_exact	A wide bond cap on coherent RX noise and a memory budget on Pauli noise decode every syndrome like the uncapped decoder.	✅ pass
test_matches_mld_on_pauli	On Pauli noise the circuit-level decoder decodes like exact MLD: every repetition-code syndrome, and sampled rotated-surface shots.	✅ pass
test_posterior_matches_dense_oracle	Class weights match a dense density-matrix posterior to 1e-9 and the argmax is a MAP class, in amplitude mode (coherent RX on a repetition memory) and in chi mode (damping on a logical-|1> repetition memory, RX on a surface memory with undeclared checks), each build exact.	✅ pass
test_warns_outside_its_exact_regime	Damping on a surface Z-memory reads a random data readout, so the build flags itself approximate and warns; on a repetition memory, where every record is pinned, it stays exact and silent.	✅ pass
test_zero_rate_sources_are_scalars	A noise location that cannot fire costs no legs: at rate 0 every source tensor is a scalar, however many locations the circuit declares.	✅ pass