Define Rust persona runtime alpha contract

src/shared/generated/cognition/AdaptiveThroughputPlan.ts

@@ -0,0 +1,4 @@
+// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
+import type { ThroughputJob } from "./ThroughputJob";
+
+export type AdaptiveThroughputPlan = { admitted: Array<ThroughputJob>, deferredMissingDependencies: Array<ThroughputJob>, deferredResourcePressure: Array<ThroughputJob>, droppedStale: Array<ThroughputJob>, droppedSuperseded: Array<ThroughputJob>, };

src/shared/generated/cognition/AdaptiveThroughputRequest.ts

@@ -0,0 +1,5 @@
+// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
+import type { ThroughputJob } from "./ThroughputJob";
+import type { ThroughputLaneBudget } from "./ThroughputLaneBudget";
+
+export type AdaptiveThroughputRequest = { readyArtifactKeys: Array<string>, laneBudgets: Array<ThroughputLaneBudget>, jobs: Array<ThroughputJob>, nowMs: number, };

src/shared/generated/cognition/PersonaTurnPlan.ts

@@ -3,4 +3,4 @@
 /**
  * Persona-specific work item for the turn.
  */
-export type PersonaTurnPlan = { personaId: string, displayName: string, specialty: string, model: string, provider: string, localModel: boolean, generationOrder: number, personaContextKey: string, ragCacheKey: string, inputBudgetTokens: number, maxOutputTokens: number, sourceNames: Array<string>, };
+export type PersonaTurnPlan = { personaId: string, displayName: string, specialty: string, model: string, provider: string, localModel: boolean, generationOrder: number, generationWave: number, personaContextKey: string, ragCacheKey: string, inputBudgetTokens: number, maxOutputTokens: number, estimatedStartMs: number, estimatedFinishMs: number, sourceNames: Array<string>, };

src/shared/generated/cognition/RecipeTurnBatchPlan.ts

@@ -5,4 +5,4 @@ import type { SharedRagSourcePlan } from "./SharedRagSourcePlan";
 /**
  * Result of `cognition/plan-turn-batch`.
  */
-export type RecipeTurnBatchPlan = { turnKey: string, roomId: string, messageId?: string, queryText: string, sharedSources: Array<SharedRagSourcePlan>, personaPlans: Array<PersonaTurnPlan>, skippedDuplicatePersonaIds: Array<string>, maxConcurrentLocalGenerations: number, };
+export type RecipeTurnBatchPlan = { turnKey: string, roomId: string, messageId?: string, queryText: string, sharedSources: Array<SharedRagSourcePlan>, personaPlans: Array<PersonaTurnPlan>, skippedDuplicatePersonaIds: Array<string>, maxConcurrentLocalGenerations: number, estimatedFirstResponseMs: number, estimatedAllResponsesMs: number, meetsFirstResponseBudget: boolean, meetsAllResponsesBudget: boolean, };

src/shared/generated/cognition/RecipeTurnBatchRequest.ts

@@ -11,4 +11,21 @@ export type RecipeTurnBatchRequest = { trigger: RecipeTurnTrigger, personas: Arr
  * Total input-token budget for shared RAG planning. Per-persona
  * generation still uses each candidate's model limits.
  */
-totalInputBudgetTokens: number, };
+totalInputBudgetTokens: number,
+/**
+ * Local inference lanes available for this turn. Zero means unknown,
+ * treated as one lane. The host should pass `inference/capacity` here
+ * so the planner, admission control, and runtime scheduler share the
+ * same source of truth.
+ */
+localInferenceCapacity: number,
+/**
+ * Visible-response budget for the first local persona reply. Zero means
+ * use the alpha gate default.
+ */
+firstResponseBudgetMs: number,
+/**
+ * Visible-response budget for every admitted persona to either respond
+ * or emit a silence reason. Zero means use the alpha gate default.
+ */
+allResponsesBudgetMs: number, };

src/shared/generated/cognition/ResourceClass.ts

@@ -0,0 +1,3 @@
+// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
+
+export type ResourceClass = "CPU" | "DATA" | "GPU" | "EMBEDDING" | "LOCAL_GENERATION" | "CLOUD_PROVIDER" | "IO" | "MEDIA" | "RENDER" | "MEMORY" | "BACKGROUND";

src/shared/generated/cognition/TargetSilicon.ts

@@ -0,0 +1,3 @@
+// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
+
+export type TargetSilicon = "CPU" | "GPU" | "UNIFIED_MEMORY" | "NETWORK" | "DISK" | "CLOUD" | "BACKGROUND";

src/shared/generated/cognition/ThroughputJob.ts

@@ -0,0 +1,9 @@
+// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
+import type { ResourceClass } from "./ResourceClass";
+import type { TargetSilicon } from "./TargetSilicon";
+
+export type ThroughputJob = { jobId: string, artifactKey: string, resourceClass: ResourceClass, targetSilicon: TargetSilicon, priority: number, costUnits: number, dependencyKeys: Array<string>, createdAtMs: number,
+/**
+ * Zero means never stale.
+ */
+staleAfterMs: number, };

src/shared/generated/cognition/ThroughputLaneBudget.ts

@@ -0,0 +1,10 @@
+// This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
+import type { ResourceClass } from "./ResourceClass";
+import type { TargetSilicon } from "./TargetSilicon";
+
+export type ThroughputLaneBudget = {
+/**
+ * Semantic owner for observability. Admission is keyed by target_silicon
+ * so LocalGeneration, Media, and Render can share one physical GPU budget.
+ */
+resourceClass: ResourceClass, targetSilicon: TargetSilicon, maxConcurrency: number, maxCostUnits: number, };