Skip to content

GOTM-SCPN Paper 0 K_nm measured-coupling evidence checklist

This checklist defines the minimum evidence package required before the first Paper 0 K_nm replay can be reconsidered for measured-system promotion. It is deliberately stricter than the current replay output because the current EEG candidate is dimensionless PLV and does not contain per-edge uncertainty.

Current status

  • Current replay status: blocked_non_closing_preregistered_replay
  • Current promotion decision: do_not_promote
  • Hardware submission: blocked
  • Claim promotion: blocked
  • Primary blocker: EEG alpha PLV is a synchronisation observable, not a calibrated coupling magnitude.
  • Secondary blocker: the primary candidate lacks per-edge uncertainty.

Required evidence before reconsideration

Evidence item Acceptance condition Failure mode
Named measured system Source dataset identifies the biological or physical system, channel map, acquisition context, and preprocessing provenance. Anonymous or partially described matrix cannot be promoted.
Coupling units Each K_nm edge has source units or a documented transformation into a calibrated coupling scale. Dimensionless association matrix remains a candidate observable only.
Unit-class gate The measured-system promotion audit classifies the unit as a calibrated or model-derived coupling magnitude. PLV, coherence, correlation, mutual information, transfer entropy, or unknown unit classes remain non-promotional observables.
Per-edge uncertainty Each retained edge has uncertainty, standard error, posterior interval, or equivalent uncertainty metadata. Point estimates without uncertainty keep the measured-system gate closed.
Normalisation lock Normalisation is specified before replay and cannot be adjusted after seeing diagnostics. Post-hoc normalisation invalidates the preregistration.
Negative controls Sparse and dense controls are defined with the same analysis code path and frozen thresholds. Topology-only agreement can be mistaken for physical validation.
Null-model battery Permutation or resampling tests are frozen with seeds, counts, and interpretation thresholds. Diagnostic significance cannot be audited.
Frozen manifest Dataset paths, digests, seeds, environment assumptions, and scripts are recorded before promotion review. Input drift or hidden replay changes invalidate comparison.
Claim boundary Promotion text states exactly what is validated and what remains unvalidated. Broad K_nm or hardware claims remain blocked.

Minimum promotion review packet

A future promotion review must include:

  • Updated preregistration document.
  • Machine-readable replay JSON with digest-locked inputs.
  • Human-readable replay report.
  • Contract-check output for the replay schema version.
  • Comparator output showing no drift.
  • Null-model summary for primary and control systems.
  • Explicit falsifier evaluation.
  • Signed decision preserving the no-QPU boundary unless hardware gates are separately authorised.

The contract-check command for the current replay is:

PYTHONPATH=src ./.venv-linux/bin/python scripts/export_paper0_knm_replay_contract.py \
  --check-replay data/paper0_knm_preregistered_replay.json

The measured-candidate release gate for the committed EEG and power-grid audit artefacts is:

scpn-bench knm-measured-candidate-gate

It fails if any committed candidate is promoted, changes its expected edge count, drops the strict unit-class decision, or records K_nm physical validation as closed.

Non-promotional outputs

The following outputs are useful but not sufficient for measured-system promotion:

  • Matrix correlation against the built-in 16-layer candidate.
  • A stable SHA-256 input manifest.
  • A passing no-QPU replay gate.
  • A dimensionless PLV matrix.
  • A power-grid negative control with uncertainties.
  • A report that reproduces exactly from committed artefacts.

These outputs improve reproducibility. They do not close the measured-system gate by themselves.