Multiplex

Compute systems

Scheduling many persistent agent identities across bounded model residency and physical compute.

Multiplex separates a persistent agent identity from the model process that temporarily performs its work. Its research direction covers scheduling, model residency, routing, and resource accounting so useful work can share constrained compute without losing identity or evidence. The maintained core implements durable queues, bounded reservations, adapter reuse, recovery states, and routing mechanisms. Research evaluation must compare useful outcomes, latency, cost, and failure behavior on real admitted substrates, keeping stored identities and queued jobs separate from simultaneous model execution.

Evidence priorities

Make progress measurable.

The initiative is framed around testable questions. Its research direction should be assessed against observed evidence and the limits of each experiment.

• Useful outcomes, latency, and cost on admitted physical compute
• Queue recovery and behavior under model eviction or failure
• A clear distinction between persistent identities and simultaneous execution

Research goals and proposed mechanisms are not established findings. Product-linked tracks describe the research behind a capability, rather than a claim that its hypotheses have been validated.