Skip to content

Generic adjudication scaffold

SV-Gap now includes a generic, deterministic protocol for comparing prerecorded digital observation traces. It exists to make an adjudication method inspectable before any project-specific perturbation mechanism is approved or published.

What is implemented

  • a strict JSON trace contract and CSV normalizer;
  • content digests and candidate-bound deterministic seed derivation;
  • exact or golden-X-wildcard comparison with an explicit bounded global shift;
  • hazard_demonstrated, no_divergence_observed, and inconclusive verdicts;
  • a calibration gate with known-positive, known-negative, and inconclusive prerecorded fixtures; and
  • a machine-readable adjudication report.

Run the public calibration:

svgap calibrate-adjudicator examples/adjudication_calibration/suite.json \
  --max-shift 1

The fixtures are deliberately synthetic and prerecorded. They validate protocol behavior, not a candidate-level hazard claim.

What remains blocked

ResetReleaseSkewInstrumenter is only a capability marker. It has no RTL or netlist rewrite, skew injection, simulation, or candidate execution code. Any call returns BLOCKED_PENDING_PATENT_AND_EMPLOYER_REVIEW. The project must not publish a real perturbation implementation or results produced by one until the appropriate review permits it.

Consequently, the current scaffold supports method legibility and community review of contracts. It does not upgrade prerecorded examples into empirical evidence about generated candidates.

Comparator semantics

A shift of n compares the golden trace with its last n samples removed to the observed trace with its first n samples removed. The remaining windows must have equal length. This permits a declared global observation delay without silently accepting a short matching prefix. Candidate digests and signal sets must match or the result is inconclusive.