Struct GornaArbitrator 

pub struct GornaArbitrator { /* private fields */ }

Arbitrates resource allocation between multiple ISAs.

The arbitrator implements a two-pass approach:

• Pass 1 (Negotiation): Collects strategy options from all agents.
• Pass 2 (Fitting): Selects the optimal strategy combination that fits within the global frame budget, respecting priorities and VRAM constraints.

Implementations

impl GornaArbitrator

pub fn new(lock_timeout: Duration) -> Self

Creates a new arbitrator with the specified lock timeout.

The lock timeout determines how long to wait when acquiring locks on agents during negotiation and budget issuance. Agents that cannot be locked within this timeout are skipped.

pub fn set_wave_plan(&mut self, waves: &[Vec<AgentId>])

Sets the scheduler’s latest wave plan (how agents are grouped for concurrent execution). Budget fitting costs each wave by its critical path (max of its members); an empty plan means serial execution, so fitting falls back to summing per-agent costs — bit-identical to the pre-parallel behaviour.

pub fn set_adaptation_mode(&mut self, agent_id: AgentId, mode: AdaptationMode)

Sets the [AdaptationMode] for an agent — the developer-control surface. Manual(strategy) pins the agent; Learning (default) lets GORNA negotiate.

pub fn adaptation_mode(&self, agent_id: AgentId) -> AdaptationMode

Returns the [AdaptationMode] configured for an agent (default Learning).

pub fn arbitrate( &self, context: &Context, report: &AnalysisReport, agents: &mut [Arc<Mutex<dyn Agent>>], measured_costs: &HashMap<AgentId, f64>, replay: Option<&TickDecisions>, hints: &HashMap<AgentId, AgentHints>, ) -> TickDecisions

Performs a full GORNA arbitration round.

Arguments

• context: The current DCC situational model (phase, hardware, multiplier).
• report: The analysis report from the HeuristicEngine.
• agents: The registered ISA agents.
• measured_costs: Per-agent measured execution cost in milliseconds (from the DCC’s empirical cost models). Used to calibrate the agents’ self-quoted strategy estimates against reality; agents without a measurement keep their quotes as-is (cold start).
• replay: When Some, issue the recorded strategy per agent instead of the negotiated fit (deterministic replay — bypasses budget fitting and the per-agent AdaptationMode). None for normal live arbitration.
• hints: Per-agent developer [AgentHints] accumulated by the DCC. Prioritize biases the negotiation priority (which agents the fit upgrades first); Cap clamps the issued strategy to a time ceiling. Advisory only — replay and a Manual pin both override a hint.

Returns the [TickDecisions] actually issued this tick (agent → strategy), so the DCC can record them for later replay.