sandialabs / LoQS / 41202741864

90%

push

github

Fix STIM-circuit and `DictNoiseModel` bugs on the pyGSTi-free path (#168)

* Allow idle STIM qubit labels and split layers on TICKs

STIMPhysicalCircuit now keeps labels its circuit doesn't use, and its
layer edits read circuits with or without a final TICK alike (#138).

- Labels past the highest index are idle qubits; a stim.Circuit is
  compacted only when label and used-qubit counts match.
- Too few labels, or an index past the end, raise in one check in
  set_qubit_labels_inplace; copy() no longer swaps integer labels.
- One TICK splitter (tagged and REPEAT TICKs too) feeds every layer
  edit; each keeps its input's terminator and empty layers.
- Merge checks collisions on parsed targets, and all layer edits now
  expand REPEAT blocks.
- Minor: insert's idx follows list slicing; idles are added in label
  order; shared circuit tests run STIM with and without a final TICK;
  new tests for idle labels, copies and tagged/REPEAT layers.

Co-authored-by: claude-sonnet-5-5 <noreply@anthropic.com>
Co-authored-by: claude-opus-5-5 <noreply@anthropic.com>

* Run DictNoiseModel STIM circuits on every state backend

DictNoiseModel converts its reps to the types a state asks for and reads
STIM circuits target by target, so one STIM circuit and model now agree
on the STIM, statevector and QuantumSim states (#138).

- get_reps converts reps to the first requested type that works, cached
  per entry; output_*_reps also list every type the model can reach.
- STIM lookup keeps label types and gives one rep per target group in
  target order; STIM-text reps still merge over disjoint groups.
- STIMPhysicalCircuit._parsed_lines owns STIM line parsing (line kinds,
  target labels, pair groups), so the model keeps only its lookup.
- Annotations reach only states taking STIM text; noise-channel lines
  run on a STIM state and are refused elsewhere.
- Tagged or unknown instructions, rec/Pauli targets and stringified keys
  for integer labels raise; a failed lookup raises... (continued)

3693 of 4422 branches covered (83.51%)

Branch coverage included in aggregate %.

11961 of 13151 relevant lines covered (90.95%)

0.91 hits per line