import {z} from 'zod' export type GetApiApiV1ConsoleEvalsPerceptionLeaderboardArgs = {'x-api-key'?: (string | null) | undefined}; export type GetApiApiV1ConsoleEvalsPerceptionLeaderboardOptions = {baseUrl: string, HTTPBearer?: string | undefined, fetch?: typeof fetch | undefined}; export const datasetReference = z.object({repository: z.string(), revision: z.string().nullable().optional(), url: z.string().nullable().optional(), qualification: z.string().nullable().optional()}); export const officialComparator = z.object({provider: z.string(), score: z.number().gte(0).lte(1)}); export const frontierReference = z.object({provider: z.enum(["GPT-6 Astra", "Claude Fable 5.1"]), score: z.number().gte(0).lte(1), median_seconds: z.number().gte(0).nullable().optional(), p95_seconds: z.number().gte(0).nullable().optional(), captured_cost_usd: z.number().gte(0).nullable().optional(), excluded_from_official_sota: z.literal(true).optional()}); export const perceptionLane = z.object({id: z.string(), operation: z.enum(["parse", "ground", "split", "classify", "extract"]), document_type: z.string(), baseline_score: z.number().gte(0).lte(1), ndi_score: z.number().gte(0).lte(1), official_comparator: officialComparator.nullable().optional(), status: z.enum(["verified_leader", "provisional_leader", "improved", "behind", "limited_evidence", "regressed"]), median_seconds: z.number().gte(0).nullable().optional(), p95_seconds: z.number().gte(0).nullable().optional(), usage_units: z.number().int().gte(0).nullable().optional(), estimated_cost_usd: z.number().gte(0).nullable().optional(), evidence: z.string(), frontier_references: z.array(frontierReference).optional()}); export const perceptionImprovement = z.object({title: z.string(), summary: z.string(), result: z.string(), lane_ids: z.array(z.string()).optional(), pr_urls: z.array(z.string()).optional()}); export const providerCampaignResult = z.object({provider: z.string(), score: z.number().gte(0).lte(1).nullable().optional(), median_seconds: z.number().gte(0).nullable().optional(), p95_seconds: z.number().gte(0).nullable().optional(), cost_usd: z.number().gte(0).nullable().optional(), cost_kind: z.enum(["captured", "estimated", "captured_lower_bound", "proxy_upper_bound", "not_captured", "not_run"]), qualification: z.string()}); export const providerCampaign = z.object({id: z.string(), name: z.string(), operation: z.enum(["parse", "ground", "split", "classify", "extract"]), metric: z.string(), scope: z.string(), qualification: z.string(), dataset: datasetReference.nullable().optional(), results: z.array(providerCampaignResult)}); export const perceptionSnapshot = z.object({version: z.string(), published_at: z.string(), title: z.string(), summary: z.string(), dataset: datasetReference, frontier_dataset: datasetReference.nullable().optional(), verified_leaders: z.number().int().gte(0).lte(31), leaders_with_provisional: z.number().int().gte(0).lte(31), close_signals: z.number().int().gte(0).lte(31).nullable().optional(), lanes: z.array(perceptionLane), improvements: z.array(perceptionImprovement).optional(), provider_campaigns: z.array(providerCampaign).optional()}); export const perceptionLeaderboard = z.object({latest_version: z.string(), snapshots: z.array(perceptionSnapshot)}); 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 Perception Leaderboard * * GET /api/v1/console/evals/perception-leaderboard */ export const getApiApiV1ConsoleEvalsPerceptionLeaderboard = async (args: GetApiApiV1ConsoleEvalsPerceptionLeaderboardArgs, options: GetApiApiV1ConsoleEvalsPerceptionLeaderboardOptions) => { const {baseUrl, fetch: fetchFn = fetch, HTTPBearer} = options ?? {} const query = new URLSearchParams() const headers = new Headers() headers.set('accept', "application/json") if (args['x-api-key'] != null) headers.set("x-api-key", String(args['x-api-key'])) if (HTTPBearer !== undefined) headers.set('authorization', `Bearer ${HTTPBearer}`) const url = `${baseUrl.replace(/\/+$/, '')}/api/v1/console/evals/perception-leaderboard${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(perceptionLeaderboard, await res.json()) };