Exact Methods and Proof Obligations¶
LiDMaS+ uses exact or explicitly-defined computational procedures at each stage. This section documents method-level proof obligations for engineering trust.
Method Classes¶
- Syndrome construction
- Exact parity evaluation over defined check supports.
- Decoder optimization/inference
- MWPM objective minimization on constructed defect graphs.
- Union-Find + peeling graph procedure.
- BP message-passing approximations with explicit update rules.
- Statistical inference
- Closed-form estimators (means/rates/Wilson) and bootstrap CIs.
- Cross-source comparators
- Explicit delta metrics, TV distance, and Jensen-Shannon divergence.
Proof Obligations (Engineering Form)¶
| Obligation | Statement |
|---|---|
| O1: Syndrome consistency | Generated/corrected syndromes must satisfy parity relations in the selected code geometry. |
| O2: Decoder output validity | Corrections must be schema-valid and auditable per request line. |
| O3: Deterministic replay | Fixed seeds/configs produce equivalent artifacts on rerun. |
| O4: Comparator correctness | Source-vs-reference deltas are computed directly from exported tables, not ad hoc UI transforms. |
| O5: Surface parity | CLI and App use equivalent workflow semantics for paper_04 parity checks. |
Theorem-Style Claims (Current Scope)¶
Claim A: Reproducible Metric Extraction¶
Given fixed artifacts and deterministic parsing rules, replay metrics are deterministic functions of input NDJSON.
Evidence paths:
examples/paper_runs/paper_04/scripts/analyze_pennylane_vs_lidmas.pyexamples/paper_runs/paper_04/scripts/analyze_bootstrap_ci.py
Claim B: Request-Set Equivalence Distances Are Well-Defined¶
TV and JS statistics are computed from empirical PMFs over extracted event counts.
Evidence path:
examples/paper_runs/paper_04/scripts/analyze_request_equivalence.py
Claim C: Geometry-Constrained Generator Consistency¶
Request generation uses explicit distance-driven formulas \(n, m_X, m_Z\) and support maps.
Evidence path:
examples/paper_runs/paper_04/scripts/generate_comparison_requests.py
Boundary¶
These are implementation-level correctness obligations, not machine-checked formal proofs. The trust mechanism is transparent computation + artifact traceability.