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.
| 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.
| 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. |
| 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. |
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. |
- 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.
docs/results.mdremains 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.
| 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. |
- 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}.jsonwith 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_benchmarkand 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 2026-05-05 —
docs/campaigns/ibm_phase2_preregistered_manifest_2026-05-05.mdandresults/ibm_phase2_preregistration_2026-05-05.jsonrecord 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 / CANCELLED —
ibm_kingstonjobibm-run-ca8b9612732b84dcwas cancelled after live hardware transpilation exceeded the reduced dry-run depth budget; IBM metadata reported0quantum seconds and0usage seconds. This job is quarantined and not evidence. - 2026-05-05 reduced A+G hardware run status: DONE / PROMOTED —
ibm_kingstonjobsibm-run-7da8644af35021fbandibm-run-6f9990bba1d90a12completed with 612 raw-count circuits. The committed reproducer reports Fisher chi2140.671952, Fisher p3.773718e-20, and 6/10 significant depths atp < 0.05. - 2026-05-05 B-C scaling run status: DONE / PROMOTED AS MIXED —
ibm_kingstonjobibm-run-1f46ebd0da8912ffcompleted with 280 raw-count circuits and IBM-reported usage305quantum seconds. The committed reproducer reportsn=6Fisher p1.883218e-07with negative significant depths andn=8Fisher p2.675193e-04with positive middle-depth sign. - Phase 2 publication package status: DONE / PROMOTED —
docs/publication/publication_phase2_package_2026-05-05.mdrecords the promoted A+G, B-C, and popcount-control artefacts plus conservative claim boundaries.
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.
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.
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.