Fusion-to-Fire full chain

The full-chain surface requires sibling source checkouts and a real Verilator installation. See the reproduction guide for environment setup, evidence semantics, and the hardware-claim boundary.

full_chain

Public causal full-chain demonstration surface.

__all__ = ['FullChainCaseResult', 'FullChainError', 'FullChainRunResult', 'evaluate_exact_current_aer_diagnostic', 'evaluate_neuro_symbolic_admission', 'evaluate_pulsed_scheduler_admission', 'run_full_chain_demo', 'verify_full_chain_replay'] module-attribute

FullChainCaseResult(payload, fusion_trajectory) dataclass

One complete nominal or adversarial causal-chain result.

FullChainRunResult(output_dir, manifest, nominal, safety_veto) dataclass

Paths and in-memory results emitted by one two-case demonstration.

FullChainError

A fail-closed full-chain dependency or agreement failure.

evaluate_exact_current_aer_diagnostic(*, shot_id)

Run MIF-007 events through MIF-006 mapping and CONTROL's stateful LIF.

This diagnostic never grants actuation authority. The historical deterministic lif_fire gate and the independent machine-safety veto remain the only neuro-symbolic inputs to the trigger decision.

evaluate_neuro_symbolic_admission(report, *, backend_name='auto')

Compile and execute CONTROL's stateless Petri/stochastic permit gate.

CONTROL retains the cross-repository lif_fire API name, but its current transition contract resets the optional neuron before one threshold step. This path therefore carries no membrane state between evaluations and is not evidence of temporal LIF dynamics.

evaluate_pulsed_scheduler_admission(report)

Drive the real CONTROL pulsed-shot scheduler to its compression guard.

run_full_chain_demo(output_dir, *, code_root=None, verilator=None, sc_backend='auto')

Run both causal cases, replay them, and emit a digest-bound bundle.

verify_full_chain_replay(first, second)

Require bit-identical JSON and numerical trajectory replay.