Skip to content

Latest commit

 

History

History
239 lines (206 loc) · 28.8 KB

File metadata and controls

239 lines (206 loc) · 28.8 KB

Hardware Status Ledger

Result-pack integrity layer: promoted IBM raw-count datasets are now indexed in hardware_result_packs.md, with the canonical manifest at data/hardware_result_packs/manifest.json and an offline verifier at scripts/verify_hardware_result_packs.py. The result-pack layer preserves artefact integrity and claim boundaries; it does not broaden any claim in this ledger.

This page is the public index for hardware and simulator evidence. Summary pages may quote selected results, but every quoted campaign should point back here or to a campaign-specific artefact with backend, date, mitigation state, job count, and raw-data path.

Status Snapshot — 2026-05-06

Area Current public status Canonical source
Package line Version 1.1.0, Python >=3.11, Qiskit >=2.2,<3.0. pyproject.toml, CHANGELOG.md
Generic compiler entry point scpn_quantum_control.kuramoto_core validates arbitrary K_nm/omega problems and compiles Hamiltonians, dense matrices, Trotter circuits, and order-parameter measurements. docs/kuramoto_core_facade.md
Core-package licence boundary Possible future lightweight core split is documented, but no permissive relicensing has occurred. docs/core_package_boundary.md
Baseline hardware campaign ibm_fez Heron r2 baseline artefacts are retained as legacy QPU evidence. Quote only values that name a committed raw artefact or retrieval file; do not use this campaign as proof of broad advantage. results/ibm_hardware_2026-03-28/, results/march_2026/, results/IBM_HARDWARE_COMPLETE_AUDIT_2026-03-30.md
DLA parity campaigns ibm_kingston Heron r2 Phase 1, Phase 2 A+G, Phase 2 B-C, and popcount-control campaigns are promoted hardware datasets: raw counts, public run labels, integrity checks, and reproduction harnesses are committed. Raw IBM job identifiers are retained only in the private mapping. data/phase1_dla_parity/, data/phase2_dla_parity/, data/phase2_scaling_bc/, data/phase2_popcount_control/, docs/publication/publication_phase2_package_2026-05-05.md, scripts/analyse_phase2_dla_parity.py, scripts/analyse_phase2_scaling_bc.py, scripts/analyse_phase2_popcount_control.py
SCPN/FIM hardware campaign ibm_kingston Heron r2 pilot and repeated follow-up are promoted as a negative/falsification result for the simple digital lambda=4 hardware-protection hypothesis on the tested circuit family. data/scpn_fim_hamiltonian/, docs/campaigns/scpn_fim_claim_boundary_2026-05-05.md, scripts/analyse_fim_ibm_pilot.py, scripts/analyse_fim_ibm_repeated_followup.py, scripts/analyse_fim_readout_matrix_mitigation.py
Simulator claims BKT, OTOC, Floquet, MBL, FIM, and classical comparison material remain simulator or classical-baseline claims unless a hardware artefact is named. results/SIMULATOR_RESULTS.md, results/classical_baselines_2026-03-30.json
Quarantined / unpromoted IBM output Any frontier, V2, queued-job, placeholder, or aggregate-only IBM output is not promoted until it has raw counts, private retrieval map, analysis code, and an explicit ledger row. results/ibm_hardware_v2_2026-03-29/, results/ibm_runs/jobs.json, <private-internal-record>

README, docs/index.md, and docs/results.md should treat this dated snapshot as the source of truth for public status wording. If a campaign value changes, update this table first, then refresh the summary pages.

The dated snapshot reflects the last promotion review, not a freshness guarantee for every row. The Package-line release version above is not left to manual refresh: scripts/check_version_consistency.py (a pre-commit and CI gate) holds it in lock-step with pyproject.toml, so it cannot silently lag the release between snapshots. Campaign rows change only when a new dataset is promoted, so an unchanged date here means no new promotion has occurred.

Claim Classes

Class Meaning Required evidence
Theory Analytic statement or theorem. Derivation, assumptions, and testable prediction.
Simulator Classical, tensor-network, statevector, or noisy simulator result. Script/notebook path, seed policy, package versions, and output artefact.
Hardware, unmitigated Raw QPU measurement without post-run mitigation beyond standard transpilation. Backend, public run labels, shots, circuit family, and raw counts.
Hardware, mitigated QPU result after mitigation such as ZNE, symmetry checks, or dynamical decoupling. Raw counts, mitigation parameters, and unmitigated comparator.
Falsification or noise-limited Negative or bounded result that constrains a claim. Same evidence as the corresponding simulator or hardware class.

Public Hardware Campaigns

Campaign Backend Date Evidence class Public artefacts Current use
Baseline IBM roadmap experiments ibm_fez Heron r2 2026-03 Legacy hardware artefacts; quote artefact-backed rows only results/ibm_hardware_2026-03-28/, results/march_2026/, results/IBM_HARDWARE_COMPLETE_AUDIT_2026-03-30.md Baseline Bell, QKD, VQE, ZNE, Trotter, and UPDE observations, with no broad-advantage promotion.
DLA parity Phase 1 ibm_kingston Heron r2 2026-04 Promoted hardware dataset with raw-count reproducer data/phase1_dla_parity/, paper/submissions/submission_002_phase1_dla_parity/phase1_dla_parity_short_paper.md, docs/dla_parity.md, scripts/run_dla_parity_suite.py DLA parity asymmetry paper and result figures.
DLA parity Phase 2 reduced A+G ibm_kingston Heron r2 2026-05 Promoted hardware replication dataset with raw-count reproducer data/phase2_dla_parity/, scripts/analyse_phase2_dla_parity.py, results/ibm_phase2_preregistration_2026-05-05.json High-statistics n=4 DLA parity replication plus readout baseline only; no n=6-12, GUESS, or broad-advantage claim.
DLA parity Phase 2 B-C scaling ibm_kingston Heron r2 2026-05 Promoted mixed scaling dataset with raw-count reproducer data/phase2_scaling_bc/, scripts/analyse_phase2_scaling_bc.py, docs/campaigns/ibm_phase2_scaling_bc_manifest_2026-05-05.md Mixed n=6,8 scaling evidence: n=8 positive middle-depth sign, n=6 negative at significant depths; no monotone scaling claim.
DLA parity Phase 2 popcount control ibm_kingston Heron r2 2026-05 Promoted hardware control dataset with raw-count reproducer data/phase2_popcount_control/, scripts/analyse_phase2_popcount_control.py, docs/publication/publication_phase2_package_2026-05-05.md Excitation count and state choice materially contribute; no DLA-parity-only causal claim.
SCPN/FIM pilot and repeated follow-up ibm_kingston Heron r2 2026-05 Promoted negative/falsification hardware result data/scpn_fim_hamiltonian/, docs/campaigns/scpn_fim_claim_boundary_2026-05-05.md, scripts/analyse_fim_ibm_repeated_followup.py Digital lambda=4 implementation increases leakage/decreases retention for the tested circuit family; no coherence-protection claim.
Mitigated Bell re-run (KIMI-9) ibm_fez Heron r2 2026-07 Promoted preregistered hardware dataset with raw-count reproducer data/bell_rerun_mitigated/ (readiness, raw counts, analysis; SHA-256 in files), docs/campaigns/bell_rerun_mitigated_prereg_2026-07-16.md, scripts/submit_bell_rerun_mitigated.py Per pair with exact σ: S = 2.6953 ± 0.0231 (q0–q1), S = 2.6807 ± 0.0232 (q2–q3) unmitigated; 2.7517/2.7510 mitigated. March setting-1 anomaly not reproduced (correct sign, near-band magnitude); small residual q2–q3 asymmetry documented. 40 usage seconds; no loophole-free, device-independent, or QKD claim.
Maximum-width Kuramoto-XY sweep (WIDTH-1) ibm_fez Heron r2 2026-07 Promoted preregistered hardware dataset with raw-count reproducer and exact MPS baseline data/max_width_kuramoto_xy/ (readiness, raw counts, analysis; SHA-256 in files), docs/campaigns/max_width_kuramoto_xy_prereg_2026-07-16.md, scripts/submit_max_width_kuramoto_xy.py Workload-width engineering and noise characterisation only: R(n) vs the exact MPS baseline (≈0.61) at n = 32/64/104/122 gives mitigated R = 0.459/0.342/0.346/0.376 (Trotter reps 1; reps 2 similar), error-aware DFS chain layouts, per-qubit readout mitigation on/off. Widest executed workload in this repository (122 qubits, transpiled depth 2103). Shallow 1-D circuits are classically simulable by construction — NO advantage claim. 38 usage seconds.
On-QPU dynamic-circuit feedback demo (RC-1) ibm_fez Heron r2 2026-07 Promoted preregistered hardware dataset with raw-count reproducer data/onqpu_dynamic_feedback/ (readiness incl. the superseded miscalibrated-gate run, raw counts, analysis; SHA-256 in files), docs/campaigns/onqpu_dynamic_feedback_prereg_2026-07-16.md, scripts/submit_onqpu_dynamic_feedback.py Vendor-executed dynamic circuits (mid-circuit measurement, conditional reset and correction) accepted and executed; feedback-on vs matched open-loop TVD = 0.0693 ± 0.0086 (2 rounds) and 0.1477 ± 0.0093 (3 rounds, independent-bin multinomial σ), per-round monitor trigger rates 3–8%. Controller decisions run inside the vendor runtime: NOT external-FPGA control, and no latency claim is derivable from the outputs. 11 usage seconds.
Matched-basis QBER re-run with basis metadata (KIMI-8) ibm_fez Heron r2 2026-07 Promoted preregistered hardware dataset with raw-count reproducer and committed per-pub basis metadata data/qkd_qber_basis_metadata/ (readiness, raw counts, analysis; SHA-256 in files), docs/campaigns/qkd_qber_basis_metadata_prereg_2026-07-17.md, scripts/submit_qkd_qber_basis_metadata.py Matched-basis mismatch of two Bell pairs, job d9cosfcjeosc73fgikgg: raw 1.39%/2.71% (ZZ, q0–q1/q2–q3) and 1.68%/3.00% (XX); 0.51%/1.09% and 0.74%/1.33% with full-basis readout mitigation (binomial σ 0.18–0.27% at 4096 shots). The committed naive sift of the March counts under explicitly assumed bases gives 1.95–3.70%; the new mitigated rates fall below that band and far below the published 5.5%/5.8%, whose non-derivability caveat stands. Fixed per-pub bases, no per-shot randomness — NO QKD security, key-rate, or viability claim. 38 usage seconds.
March campaign job-id verifiability (KIMI-7) ibm_fez Heron r2 2026-07 Published hash commitments plus dated IBM retrieval receipts for the March 2026 job identifiers data/march_flagship_verifiability/march_job_id_commitments.json (sha256 cdeba2530ce892377dd3d9902d6faf19e96cc92bdb781f67da7c2bc5721c5309), data/march_flagship_verifiability/march_retrieval_receipts_20260717T011035Z.json (sha256 55169b36d75877e6dc54e1ce2cc4141d6ef38faa12a72e34e3b56e665accd360), scripts/build_march_job_id_commitments.py, scripts/retrieve_march_flagship_receipts.py Every March artefact label (24 unique job identifiers across results/ibm_hardware_2026-03-18/, results/ibm_hardware_2026-03-28/, results/ibm_hardware_2026-03-29/, results/march_2026/) carries a published SHA-256 commitment of its raw IBM job identifier (nonce-blinded; the disclosable (raw id, nonce) pair verifies it), and a 2026-07-17 read-only retrieval receipt: 24/24 jobs report DONE on ibm_fez with creation dates 2026-03-18 to 2026-03-29 and 160.0 IBM-reported usage seconds in total. Verifiability evidence only — no new physics claim; 0 quantum seconds consumed. Full disclosure 2026-07-18 (AUD-10): all 24 commitments are now OPENED — data/march_flagship_verifiability/march_job_id_disclosure_2026-07-18.json publishes each (raw IBM job id, nonce) pair, and scripts/disclose_ibm_job_ids.py (+ its CI test) re-verifies SHA-256("<raw>:<nonce>") == commitment_sha256 for all 24. The nonce-blinding is retired for the flagship; the HMAC salt and API token stay private (they grant nothing once the IDs are open).
Flagship TN classical baselines (KIMI-11) local CPU 2026-07 Committed classical-simulability quantification for the flagship workloads data/tn_baseline_flagship/tn_baseline_flagship_20260717T024000Z.json (sha256 f46f61e6336f9eda03e9e5b077995f90c1cdf80f1f5c3ed39aa9f28cd8bb18cd), scripts/tn_baseline_flagship_workloads.py Bounded-χ Aer MPS on the identical logical Trotter bodies of the 16-qubit UPDE (t=0.05, 1 rep) and 8-oscillator Kuramoto (t=0.1, 2 reps) workloads versus the exact statevector reference: UPDE-16 reproduces R to 6.3e-4 already at χ=2 (χ=32 → 1.2e-9, <1 s wall) and Kuramoto-8 converges at χ=4 (1.4e-4). The repository has never claimed quantum advantage for these workloads; this row is the committed gPEPS-style-rebuttal defence stating their classical simulability with concrete numbers. No hardware-quality statement follows.
Recovered IBM v2 fair experiments ibm_fez Heron r2 2026-03 run / 2026-07 recovered Promoted recovered raw-count hardware dataset (full disclosure) data/ibm_hardware_v2_recovered_2026-07-18/ (raw counts, real IBM job identifiers, dated calibration; SHA-256 in manifest.json), scripts/recover_ibm_v2_raw_counts.py, scripts/analyse_ibm_v2_recovered.py Read-only re-retrieval (0 QPU s) of the nine 2026-03-29 ibm_fez jobs originally committed aggregate-only. Full public disclosure: raw per-pub counts + real IBM job identifiers + the dated calibration snapshot. The committed reproducer recomputes 8/9 experiment aggregates to |Δmean|<1e-4 from the raw counts (A_odd to ~3.7%, original mitigation not in-pack). F_FIM=0.9158 > F_XY=0.8484 (all-zero survival) is a genuine observation but carries NO coherence-protection claim — that hypothesis is the falsified promoted ibm_kingston SCPN/FIM result. Supersedes the quarantined results/ibm_hardware_v2_2026-03-29/.
Simulator and classical baselines local CPU/GPU where noted 2026-03 onward Simulator results/*_2026-03-*.json, results/SIMULATOR_RESULTS.md, results/classical_baselines_2026-03-30.json BKT, OTOC, Floquet, MBL, FIM, and classical comparison material.

Quarantined / Unpromoted IBM Artifacts

The April 2026 internal incident trail records placeholder and fake-count handling in frontier IBM workflows; the public incident note is Count-Integrity Incident, April 2026. Therefore the files below must not be used as public proof until they are independently re-retrieved or reproduced from raw IBM counts and promoted here.

Artifact family Current status Reason
results/ibm_hardware_v2_2026-03-29/ RECOVERED 2026-07-18 → superseded by the promoted data/ibm_hardware_v2_recovered_2026-07-18/. The original pack was aggregate-only. The nine jobs have now been re-retrieved read-only from IBM (0 QPU s): raw counts + real IBM job identifiers + dated calibration are public, and a committed reproducer recomputes 8/9 rows to |Δ|<1e-4. The data is genuine; only the "DUAL PROTECTION" coherence-protection interpretation stays retired (falsified on the promoted ibm_kingston SCPN/FIM campaign).
results/ibm_hardware_2026-03-29/dla_parity_*.json Superseded / unpromoted. The audit identifies a circuit-depth artefact in this March DLA parity attempt; use the April data/phase1_dla_parity/ dataset instead.
results/ibm_runs/jobs.json and frontier queue outputs Quarantined. Internal logs document queued-job placeholders and fake all-zero fallback counts in related workflows.
Any "400 jobs", large-N frontier, multi-QPU, or live-loop claim Not promoted. Requires raw counts, private retrieval map, analysis script, and a new ledger row before citation.

Evidence Rules

  • Do not quote a numerical result unless the source artefact is named.
  • Do not mix simulator and hardware numbers in one table without the evidence class column.
  • Do not promote internal or pending campaign output into public claims until its raw counts, private retrieval map, and analysis script are committed.
  • If a later mitigation pass changes a conclusion, keep the older value as a historical row and add the newer value with its mitigation state.
  • Negative results stay in the ledger. They are evidence, not cleanup debt.

Current Gaps

  • docs/results.md remains a public gallery and technical summary. This ledger is the canonical status index; further result-page edits should keep detailed provenance here rather than duplicating claim-state decisions.
  • Later frontier batches must be added only after their raw artefacts, private retrieval maps, and analysis scripts are reviewed and committed.
  • Do not cite aggregate-only IBM JSON, queued-job JSON, or placeholder-derived results as hardware validation.
  • Exact repository-wide test counts belong in CI summaries or release notes. Public overview pages should describe the CI-gated suite and coverage target rather than carrying static counts that drift between commits.

Roadmap State — 2026-04-30

Queue State Next gate
High-impact execution record Complete locally. Dependency hygiene, core facade, documentation ergonomics, baselines, maintenance, frontier-track scaffolding, and CI timing-gate stabilization are checked off. Keep CI green on main; add only scoped follow-up work.
Scientific gaps Partially closed. The EEG PLV K_nm validation artefacts now cover the full 109-subject PhysioNet EEGMMIDB baseline eyes-open and eyes-closed cohorts, with a derived condition comparison. The first physical-unit measured-system control, IEEE 5-bus, is committed and does not close K_nm physical validation; broad quantum advantage remains open; p_h1 = 0.72 is an explicit open empirical/theoretical parameter. Additional physical-unit measured coupling candidates with null models; provenance-rich advantage benchmark tables; TCBO or first-principles p_h1 reproduction.
Hardware experiments March/April/May evidence is narrowed to legacy ibm_fez baseline artefacts, promoted raw-count ibm_kingston DLA Phase 1/2 datasets, and the promoted SCPN/FIM negative result. Further QPU work needs a preregistered manifest, depth/shot gates, QPU-time estimate, and explicit approval; no frontier promotion without raw-count review.
Strategic roadmap All 53 post-v1.0 differentiation tracks remain deferred / CEO-gated. Activate one track explicitly before implementation.

2026-05-03 Scientific-Gap Roadmap Progress

  • Gap A item 1 status: DONE — added a measured non-EEG physical-unit coupling candidate (IEEE 5-bus power grid) with uncertainty propagation and null-model comparison in the existing candidate control pipeline.
  • Gap A item 2 status: DONE — applied the measured-system comparison scan to all current candidate systems in data/public_application_benchmarks/{eeg_alpha_plv_8ch,friston_fep_6node,ieee5bus_power_grid,iter_mhd_8mode}.json with topology/magnitude/decision outputs and no closed candidate.
  • Gap A item 3 status: DONE — claim scope is constrained to condition-specific EEG PLV evidence for any closure that is not backed by a locked physical-unit magnitude match plus null-model pass.
  • Gap B item 1 status: DONE — scaling benchmark rows now carry backend, machine, command, dependency, and commit provenance in the committed run_scaling_benchmark and crossover-point loaders.
  • Gap B item 2 status: DONE — separate exact-simulation crossover from observable-level broad-advantage claims in the public benchmark wording.
  • Gap B item 3 status: DONE — ran the classical/Rust/GPU benchmark matrix for the quantum-advantage comparison and saved it as results/classical_rust_gpu_matrix_2026-05-03.json.
  • Gap B close-out review status: DONE 2026-05-05 — the validation readiness gate records the Phase 1 raw-count reproducer pass, a current-commit classical/Rust matrix smoke check, and the promotion rules for future broad-advantage or hardware claims.
  • Phase 2 preregistration status: DONE 2026-05-05docs/campaigns/ibm_phase2_preregistered_manifest_2026-05-05.md and results/ibm_phase2_preregistration_2026-05-05.json record the QPU-minimised A+G first live command, dry-run circuit inventory, abort criteria, evidence path, and promotion gates.
  • 2026-05-05 live attempt status: ABORTED / CANCELLEDibm_kingston job ibm-run-ca8b9612732b84dc was cancelled after live hardware transpilation exceeded the reduced dry-run depth budget; IBM metadata reported 0 quantum seconds and 0 usage seconds. This job is quarantined and not evidence.
  • 2026-05-05 reduced A+G hardware run status: DONE / PROMOTEDibm_kingston jobs ibm-run-7da8644af35021fb and ibm-run-6f9990bba1d90a12 completed with 612 raw-count circuits. The committed reproducer reports Fisher chi2 140.671952, Fisher p 3.773718e-20, and 6/10 significant depths at p < 0.05.
  • 2026-05-05 B-C scaling run status: DONE / PROMOTED AS MIXEDibm_kingston job ibm-run-1f46ebd0da8912ff completed with 280 raw-count circuits and IBM-reported usage 305 quantum seconds. The committed reproducer reports n=6 Fisher p 1.883218e-07 with negative significant depths and n=8 Fisher p 2.675193e-04 with positive middle-depth sign.
  • Phase 2 publication package status: DONE / PROMOTEDdocs/publication/publication_phase2_package_2026-05-05.md records the promoted A+G, B-C, and popcount-control artefacts plus conservative claim boundaries.

EEG Condition Comparison — 2026-04-30

The K_nm physical-validation data now includes a matched 109-subject PhysioNet EEGMMIDB baseline eyes-closed cohort (S001R02 through S109R02) and a derived eyes-closed-minus-eyes-open comparison artefact.

Metric Value Source
Eyes-open mean edge PLV 0.545586 data/knm_physical_validation/baseline_open_closed_comparison.json
Eyes-closed mean edge PLV 0.600050 data/knm_physical_validation/baseline_open_closed_comparison.json
Mean closed-minus-open delta 0.054463 data/knm_physical_validation/baseline_open_closed_comparison.json
Median closed-minus-open delta 0.061094 data/knm_physical_validation/baseline_open_closed_comparison.json
Mean absolute edge delta 0.066249 data/knm_physical_validation/baseline_open_closed_comparison.json
Largest absolute edge delta 0.135901 on edge (2, 3) data/knm_physical_validation/baseline_open_closed_comparison.json
Pearson r across edge medians 0.963961 data/knm_physical_validation/baseline_open_closed_comparison.json
EEG PLV versus eight-layer K_nm Spearman 0.767804 data/knm_physical_validation/eeg_alpha_plv_knm_comparison.json
EEG PLV versus eight-layer K_nm Pearson 0.715303 data/knm_physical_validation/eeg_alpha_plv_knm_comparison.json

This closes the condition-control EEG PLV comparison gate. It does not close the physical-unit measured-coupling gate: PLV is an association observable, and the measured-system promotion audit now blocks phase_locking_value units from promotion as calibrated K_nm magnitudes. Public claims must continue to describe this as condition-specific alpha-band EEG coupling evidence rather than measured physical K_nm magnitudes.

Measured-System Control — 2026-04-30

The first physical-unit measured-system artefact is the IEEE 5-bus power-grid swing-equation coupling matrix. It records raw public benchmark constants, conversion units, and propagated input-rounding uncertainty.

Metric Value Source
Matched edges 10 data/knm_physical_validation/measured_couplings_power_grid_ieee5bus.json
Spearman topology correlation vs five-layer K_nm 0.190394 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Pearson topology correlation vs five-layer K_nm 0.226144 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Direct RMSE 0.219843 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Direct relative RMSE versus mean absolute measured coupling 261.335679 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Best scale through origin 0.003925 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Scaled RMSE 0.000894 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Maximum direct absolute error 0.300747 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Weighted adjacency spectrum Pearson 0.826329 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Weighted adjacency spectrum RMSE 0.439359 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Weighted Laplacian spectrum Pearson 0.829195 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Weighted Laplacian spectrum RMSE 0.966608 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Kuramoto threshold-proxy ratio measured/canonical 244.239277 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Node-label null Spearman empirical p 0.363636 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Node-label null Pearson empirical p 0.264463 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Node-label null RMSE empirical p 0.264463 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Edge-value null Spearman empirical p 0.291189 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Edge-value null Pearson empirical p 0.254577 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json
Edge-value null RMSE empirical p 0.254577 data/knm_physical_validation/power_grid_ieee5bus_knm_comparison.json

Decision: this is a useful measured-system control and a negative result for the exact-magnitude K_nm promotion gate. It also does not beat the node-label or edge-value permutation null gates, and the spectral/critical-response diagnostics do not rescue the match. Physical validation remains open until a measured-system candidate with units, uncertainty, and preregistered null models passes the promotion criteria.

Measured-System Control Expansion — 2026-05-18

The physical-unit candidate set now also includes the IEEE 14-bus public benchmark as a voltage-weighted branch-admittance control. The artefact records the public branch reactances, solved voltage magnitudes, derived K_ij = V_i V_j / X_ij matrix, and propagated input-rounding uncertainty for all 91 pairwise bus edges.

Metric Value Source
Matched pairwise bus edges 91 data/knm_physical_validation/measured_couplings_power_grid_ieee14bus.json
Non-zero public branch edges 20 data/knm_physical_validation/measured_couplings_power_grid_ieee14bus.json
Spearman topology correlation vs fourteen-layer K_nm 0.406186 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Pearson topology correlation vs fourteen-layer K_nm 0.346893 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Direct RMSE 4.248824 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Direct relative RMSE versus mean absolute measured coupling 2.571947 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Best scale through origin 12.545508 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Scaled RMSE 3.759167 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Maximum direct absolute error 24.528498 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Weighted adjacency spectrum Pearson 0.832010 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Weighted adjacency spectrum RMSE 15.106532 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Weighted Laplacian spectrum Pearson 0.774822 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Weighted Laplacian spectrum RMSE 27.322728 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Critical-response relative difference 0.942248 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Node-label null mode seeded sampled permutations (4096) data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Node-label null Spearman empirical p 0.000244 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json
Edge-value null Spearman empirical p 0.000244 data/knm_physical_validation/power_grid_ieee14bus_knm_comparison.json

Decision: this expands the measured-system candidate inventory, but it does not close physical K_nm validation. Public case14 supplies branch reactances and voltage magnitudes, not measured per-bus inertia constants for every load bus, so the artefact is a non-promotional control candidate unless a future gate adds the missing dynamic-system measurements and passes the null-model, uncertainty, magnitude, and critical-response checks.