Skip to content

Neural-Operator Forecast Baselines

SPDX-License-Identifier: AGPL-3.0-or-later

scpn_quantum_control.neural_operator_baseline_product composes the existing classical neural-operator and observed-synchronisation forecast surfaces without adding another model or solver.

Bounded product

The report:

  • verifies the committed neural-operator evidence schema, host-independent cost arithmetic, disabled production-claim flag, and payload digest;
  • labels training_flops as a one-time training estimate, surrogate_flops_per_query as a per-inference estimate, and wall-clock values as advisory host-bounded measurements;
  • issues the governed advantage-language no_advantage_default certificate;
  • admits only committed public measurements, source-backed public replays, or explicit synthetic fixtures; unknown/private classifications and unsafe paths are refused;
  • records challenge-oracle registration as fail-closed and descoped because the current oracle has no public classical-baseline registration API; and
  • records multimodal forecasting as a design dependency, not completed wiring.
from scpn_quantum_control.neural_operator_baseline_product import (
    build_neural_operator_baseline_product,
)

report = build_neural_operator_baseline_product(
    "docs/benchmarks/neural_operator_advantage.json"
)
assert report.artifact.valid
assert report.no_advantage.language_status == "no_advantage_default"
assert all(row.allowed for row in report.datasets)

Claim boundary

This is a classical forecast-baseline composition. A held-out fidelity result or arithmetic crossover does not authorise a quantum-advantage claim. It does not accept private datasets, execute hardware forecasts, register a challenge-oracle rank, or complete multimodal forecasting.

Serialized reports use neural_operator_baseline_product.v2 and reject stale schemas or altered claim boundaries.

See Neural-Operator Advantage Study, Real-Data Synchronisation Forecasting, and the Advantage Language Protocol.