import {z} from 'zod' export type GetApiV1PublicEvaluationRunsEvalRunIdArgs = {eval_run_id: string}; export type GetApiV1PublicEvaluationRunsEvalRunIdOptions = {baseUrl: string, fetch?: typeof fetch | undefined}; export const evaluationRunStatus = z.enum(["CLAIMED", "RUNNING", "SUCCESS", "FAILED", "TIMED_OUT", "STALE", "CANCELLED"]); export const safeEnvironmentMetadata = z.object({display_name: z.string(), catalog_epoch: z.string(), contract_version: z.string(), runtime_version: z.string(), result_schema_version: z.string()}); export const checkSummary = z.object({passed: z.number().int().optional(), failed: z.number().int().optional(), unknown: z.number().int().optional()}); export const rewardSummary = z.object({component_count: z.number().int().optional()}); export const evaluationItemRead = z.object({item_id: z.string(), item_kind: z.enum(["legacy_problem", "envpack_episode"]), family: z.string().nullable().optional(), status: z.string(), score: z.number().nullable().optional(), verdict_status: z.enum(["passed", "failed"]).nullable().optional(), paid_reward: z.number().nullable().optional(), check_summary: checkSummary.nullable().optional(), reward_summary: rewardSummary.nullable().optional(), step_count: z.number().int().nullable().optional(), eval_run_id: z.string(), validator_hotkey: z.string(), episode_result_id: z.string().nullable().optional()}); export const evaluationItemsRead = z.object({eval_run_id: z.string(), execution_kind: z.enum(["legacy_shoppingbench", "envpack"]), detail_released: z.boolean(), expected_item_count: z.number().int().nullable().optional(), persisted_item_count: z.number().int(), items: z.array(evaluationItemRead)}); export const evaluationRunDetail = z.object({eval_run_id: z.string(), agent_version_id: z.string(), agent_name: z.string(), validator_hotkey: z.string(), status: evaluationRunStatus, score: z.number().nullable().optional(), version_number: z.number().int(), claimed_at: z.string(), completed_at: z.string().nullable().optional(), failure_reason: z.string().nullable().optional(), is_included: z.boolean().optional(), invalidated_at: z.string().nullable().optional(), invalidation_reason: z.string().nullable().optional(), sandbox_metadata: z.record(z.string(), z.unknown()).nullable().optional(), inference_failure_count: z.number().int().nullable().optional(), inference_total: z.number().int().nullable().optional(), score_components_summary: z.record(z.string(), z.unknown()).nullable().optional(), execution_kind: z.enum(["legacy_shoppingbench", "envpack"]).optional(), environment: safeEnvironmentMetadata.nullable().optional(), items: evaluationItemsRead.nullable().optional()}); export const validateResponse = async ( schema: { readonly '~standard': { readonly validate: (value: unknown) => ({ readonly value: Output; readonly issues?: undefined } | { readonly issues: ReadonlyArray<{ readonly message: string; readonly path?: ReadonlyArray | undefined }> }) | Promise<{ readonly value: Output; readonly issues?: undefined } | { readonly issues: ReadonlyArray<{ readonly message: string; readonly path?: ReadonlyArray | undefined }> }> } }, value: unknown ): Promise => { const result = await schema['~standard'].validate(value) if (result.issues !== undefined) { const detail = result.issues .map(({ message, path }) => path?.length ? `${path.map((segment) => String(typeof segment === 'object' ? segment.key : segment)).join('.')}: ${message}` : message ) .join('; ') throw new Error(`Response does not match the schema: ${detail}`, { cause: { issues: result.issues } }) } return result.value }; /** * Get evaluation run details * * GET /v1/public/evaluation-runs/{eval_run_id} */ export const getApiV1PublicEvaluationRunsEvalRunId = async (args: GetApiV1PublicEvaluationRunsEvalRunIdArgs, options: GetApiV1PublicEvaluationRunsEvalRunIdOptions) => { const {baseUrl, fetch: fetchFn = fetch} = options ?? {} const query = new URLSearchParams() const headers = new Headers() headers.set('accept', "application/json") const url = `${baseUrl.replace(/\/+$/, '')}/v1/public/evaluation-runs/${encodeURIComponent(String(args.eval_run_id))}${query.size > 0 ? `?${query}` : ''}` const res = await fetchFn(url, {method: "GET", headers}) if (!res.ok) throw new Error(`Request failed with status ${res.status}`, {cause: {status: res.status, body: await res.text()}}) return validateResponse(evaluationRunDetail, await res.json()) };