Module platform_scheduler 

Platform-aware thread QoS and core-affinity binding.

Binds the calling thread to the most appropriate processor class for its workload so the OS scheduler makes the right placement decision:

Class	macOS QoS	Linux	Windows
UserInteractive	QOS_CLASS_USER_INTERACTIVE	P-cores (affinity)	THREAD_PRIORITY_HIGHEST
UserInitiated	QOS_CLASS_USER_INITIATED	P-cores	THREAD_PRIORITY_ABOVE_NORMAL
Default	QOS_CLASS_DEFAULT	any	THREAD_PRIORITY_NORMAL
Utility	QOS_CLASS_UTILITY	any	THREAD_PRIORITY_BELOW_NORMAL
Background	QOS_CLASS_BACKGROUND	E-cores (affinity)	THREAD_PRIORITY_IDLE

Apple Silicon AMP (P-cores + E-cores) is the primary target. On Intel/AMD/ARM64 the distinction collapses to thread priority.

Enums

QosClass

SchedulerError

Functions

bind_background_thread

Convenience wrapper: bind to Background (E-core / idle priority). Call from WAL flush, Merkle DAG, ambient orchestration threads.

bind_current_thread

Bind the calling thread to class.

bind_inference_thread

Convenience wrapper: bind to UserInteractive (P-core / max priority). Call from the LLM inference thread and active SPARQL query thread.

current_qos_class

Read the QoS class of the calling thread. Returns None on non-macOS.