Formal Problem Definition¶
Inputs¶
For each experiment point:
- code family/configuration (distance, rounds, supports),
- noise parameters (
p,sigma, or hybrid parameters), - decoder choice,
- RNG seed and trial count,
- request/response datasets in NDJSON form (for replay workflows).
State Variables¶
- physical error vectors (
e_X,e_Z), - syndromes (
s_X,s_Z), - decoder corrections (
c_X,c_Z), - residual logical predicates.
Output Variables¶
- per-run decoder metrics (flip count, warning rates, event rates),
- logical error estimates,
- confidence intervals,
- threshold/scaling summaries,
- reproducible tables/figures/manifests.
Primary Objective¶
Estimate logical failure behavior under controlled noise and decoding choices, then compare:
- across decoders,
- across software stacks (PennyLane/Qiskit/Cirq vs LiDMaS+ reference),
- across code families (
surface,gkp), - across execution surfaces (CLI vs App).
Constraint Set¶
- deterministic seeds for comparable reruns,
- stable dataset schema for replay compatibility,
- per-step artifact outputs required for auditability,
- explicit failure/warning accounting rather than silent omission.