@nirmos/sdk 0.2.0 → 0.3.2

package/src/management/execution-types.ts

@@ -0,0 +1,10 @@
+// Public execution DTOs, generated from verified runtime wire contracts.
+export type EvaluationRunCaseDetail = { "case": { "id": string; "rowKey": string; "data": { [key: string]: unknown; }; "metadata": { [key: string]: unknown; } | null; "tags": Array<string>; }; "trials": Array<{ "id": string; "caseId": string; "targetId": string; "attempt": number; "status": string; "output"?: unknown | null; "error"?: unknown | null; "provider": string | null; "model": string | null; "routeId": string | null; "metrics": { "latencyMs": number | null; "ttftMs": number | null; "inputTokens": number | null; "outputTokens": number | null; "totalTokens": number | null; "cost": number | null; "cacheHit": boolean; "retryCount": number; "fallbackCount": number; }; "traceId": string | null; "scores": Array<{ "id": string; "evaluatorVersionId": string; "evaluationColumnId": string | null; "status": string; "value"?: unknown; "label": string | null; "passed": boolean | null; "reasoning": string | null; "error": unknown | null; "reviews": Array<{ "id": string; "status": string; "value": boolean | number | string | null; "label": string | null; "comment": string | null; "updatedAt": string; }>; }>; }>; };
+export type EvaluationRunCases = { "data": Array<{ "id": string; "rowKey": string; "position"?: number; "data": { [key: string]: unknown; }; "metadata": { [key: string]: unknown; } | null; "tags": Array<string>; }>; "pagination": { "page": number; "pageSize": number; "total": number; "hasMore": boolean; }; };
+export type EvaluationRunResults = { "data": Array<{ "id": string; "caseId": string; "targetId": string; "attempt": number; "status": string; "output"?: unknown | null; "error"?: unknown | null; "provider": string | null; "model": string | null; "routeId": string | null; "metrics": { "latencyMs": number | null; "ttftMs": number | null; "inputTokens": number | null; "outputTokens": number | null; "totalTokens": number | null; "cost": number | null; "cacheHit": boolean; "retryCount": number; "fallbackCount": number; }; "traceId": string | null; "scores": Array<{ "id": string; "evaluatorVersionId": string; "evaluationColumnId": string | null; "status": string; "value"?: unknown; "label": string | null; "passed": boolean | null; "reasoning": string | null; "error": unknown | null; "reviews": Array<{ "id": string; "status": string; "value": boolean | number | string | null; "label": string | null; "comment": string | null; "updatedAt": string; }>; }>; }>; "pagination": { "page": number; "pageSize": number; "total": number; "hasMore": boolean; }; };
+export type EvaluationRun = { "id": string; "experimentId": string; "status": "QUEUED" | "RUNNING" | "COMPLETED" | "FAILED" | "CANCELLED"; "progress": { "plannedTrials": number; "completedTrials": number; "failedTrials": number; "percentage": number; }; "cost": { "current": number; "currency": string; }; "summary": { [key: string]: unknown; } | null; "policyResult": unknown | null; "releaseDecision": string; "configuration": { "suiteId": string; "datasetVersionId": string; "trialCount": number; "estimates": { "caseCount": number; "targetCalls": number; "judgeCalls": number; "totalModelCalls": number; }; "targets": Array<{ "experimentTargetId": string; "targetVersionId": string; "targetId": string; "type": string; "alias": string; "role": string; "sourceVersionId"?: string; }>; "evaluators": Array<{ "evaluationColumnId": string; "evaluatorVersionId": string; "evaluatorId": string; "slug": string; "type": string; }>; "releasePolicyVersionId"?: string; }; "error": unknown | null; "heartbeatAt": string | null; "startedAt": string | null; "completedAt": string | null; "createdAt": string; };
+export type ExecutionRunStatus = "QUEUED" | "RUNNING" | "COMPLETED" | "FAILED" | "CANCELLED";
+export type ExecutionTrial = { "id": string; "caseId": string; "targetId": string; "attempt": number; "status": string; "output"?: unknown | null; "error"?: unknown | null; "provider": string | null; "model": string | null; "routeId": string | null; "metrics": { "latencyMs": number | null; "ttftMs": number | null; "inputTokens": number | null; "outputTokens": number | null; "totalTokens": number | null; "cost": number | null; "cacheHit": boolean; "retryCount": number; "fallbackCount": number; }; "traceId": string | null; "scores": Array<{ "id": string; "evaluatorVersionId": string; "evaluationColumnId": string | null; "status": string; "value"?: unknown; "label": string | null; "passed": boolean | null; "reasoning": string | null; "error": unknown | null; "reviews": Array<{ "id": string; "status": string; "value": boolean | number | string | null; "label": string | null; "comment": string | null; "updatedAt": string; }>; }>; };
+export type RunMutationResponse = { "id": string; "status": "QUEUED" | "RUNNING" | "COMPLETED" | "FAILED" | "CANCELLED"; };
+export type StartEvaluationRunRequest = { "experimentId": string; "cases": "all" | Array<string>; "targetIds": Array<string>; "evaluatorIds"?: Array<string>; "trialCount": number; "releasePolicyVersionId"?: string; };
+export interface EvaluationResultsQuery { page?: number; pageSize?: number; caseId?: string; targetId?: string; evaluatorId?: string; status?: string; passed?: boolean; }