opensearch-genai-sdk 0.1.0

package/dist/decorators.d.ts

@@ -0,0 +1,111 @@
+/**
+ * Wrapper functions for tracing custom functions as OTEL spans.
+ *
+ * Provides traceWorkflow, traceTask, traceAgent, and traceTool
+ * that create standard OpenTelemetry spans. These are the user-facing
+ * API for tracing custom application logic -- the gap that pure
+ * auto-instrumentors don't cover.
+ *
+ * All wrappers produce standard OTEL spans with gen_ai semantic
+ * convention attributes. Zero lock-in: remove the wrapper and
+ * your code still works.
+ *
+ * TypeScript does not have Python-style decorators that work on plain
+ * functions, so we use higher-order wrapper functions instead:
+ *
+ * @example
+ * ```ts
+ * const search = traceTool("web_search", async (query: string) => {
+ *   return [{ title: `Result: ${query}` }];
+ * });
+ *
+ * const pipeline = traceWorkflow("qa_pipeline", async (question: string) => {
+ *   const results = await search(question);
+ *   return summarize(results);
+ * });
+ * ```
+ */
+export interface TraceOptions {
+    /** Optional version number for tracking changes. */
+    version?: number;
+}
+/**
+ * Trace a function as a workflow span.
+ *
+ * Use for top-level orchestration functions that coordinate
+ * multiple tasks, agents, or tool calls.
+ *
+ * @param name - Span name for the workflow.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const pipeline = traceWorkflow("qa_pipeline", async (question: string) => {
+ *   const plan = await planSteps(question);
+ *   return await execute(plan);
+ * });
+ * ```
+ */
+export declare function traceWorkflow<TArgs extends unknown[], TReturn>(name: string, fn: (...args: TArgs) => TReturn, options?: TraceOptions): (...args: TArgs) => TReturn;
+/**
+ * Trace a function as a task span.
+ *
+ * Use for individual units of work within a workflow.
+ *
+ * @param name - Span name for the task.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const summarize = traceTask("summarize", async (text: string) => {
+ *   return llm.generate(`Summarize: ${text}`);
+ * });
+ * ```
+ */
+export declare function traceTask<TArgs extends unknown[], TReturn>(name: string, fn: (...args: TArgs) => TReturn, options?: TraceOptions): (...args: TArgs) => TReturn;
+/**
+ * Trace a function as an agent span.
+ *
+ * Use for autonomous agent logic that makes decisions and
+ * invokes tools.
+ *
+ * @param name - Span name for the agent.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const agent = traceAgent("research_agent", async (query: string) => {
+ *   while (!done) {
+ *     const action = await decideNextAction(query);
+ *     result = await executeAction(action);
+ *   }
+ *   return result;
+ * });
+ * ```
+ */
+export declare function traceAgent<TArgs extends unknown[], TReturn>(name: string, fn: (...args: TArgs) => TReturn, options?: TraceOptions): (...args: TArgs) => TReturn;
+/**
+ * Trace a function as a tool span.
+ *
+ * Use for tool/function calls invoked by agents.
+ *
+ * @param name - Span name for the tool.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const search = traceTool("web_search", async (query: string) => {
+ *   return searchApi.query(query);
+ * });
+ * ```
+ */
+export declare function traceTool<TArgs extends unknown[], TReturn>(name: string, fn: (...args: TArgs) => TReturn, options?: TraceOptions): (...args: TArgs) => TReturn;
+//# sourceMappingURL=decorators.d.ts.map

package/dist/decorators.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"decorators.d.ts","sourceRoot":"","sources":["../src/decorators.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAeH,MAAM,WAAW,YAAY;IAC3B,oDAAoD;IACpD,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,aAAa,CAAC,KAAK,SAAS,OAAO,EAAE,EAAE,OAAO,EAC5D,IAAI,EAAE,MAAM,EACZ,EAAE,EAAE,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,EAC/B,OAAO,CAAC,EAAE,YAAY,GACrB,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,CAE7B;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,SAAS,CAAC,KAAK,SAAS,OAAO,EAAE,EAAE,OAAO,EACxD,IAAI,EAAE,MAAM,EACZ,EAAE,EAAE,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,EAC/B,OAAO,CAAC,EAAE,YAAY,GACrB,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,CAE7B;AAED;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,wBAAgB,UAAU,CAAC,KAAK,SAAS,OAAO,EAAE,EAAE,OAAO,EACzD,IAAI,EAAE,MAAM,EACZ,EAAE,EAAE,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,EAC/B,OAAO,CAAC,EAAE,YAAY,GACrB,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,CAE7B;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,SAAS,CAAC,KAAK,SAAS,OAAO,EAAE,EAAE,OAAO,EACxD,IAAI,EAAE,MAAM,EACZ,EAAE,EAAE,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,EAC/B,OAAO,CAAC,EAAE,YAAY,GACrB,CAAC,GAAG,IAAI,EAAE,KAAK,KAAK,OAAO,CAE7B"}

package/dist/decorators.js

@@ -0,0 +1,217 @@
+/**
+ * Wrapper functions for tracing custom functions as OTEL spans.
+ *
+ * Provides traceWorkflow, traceTask, traceAgent, and traceTool
+ * that create standard OpenTelemetry spans. These are the user-facing
+ * API for tracing custom application logic -- the gap that pure
+ * auto-instrumentors don't cover.
+ *
+ * All wrappers produce standard OTEL spans with gen_ai semantic
+ * convention attributes. Zero lock-in: remove the wrapper and
+ * your code still works.
+ *
+ * TypeScript does not have Python-style decorators that work on plain
+ * functions, so we use higher-order wrapper functions instead:
+ *
+ * @example
+ * ```ts
+ * const search = traceTool("web_search", async (query: string) => {
+ *   return [{ title: `Result: ${query}` }];
+ * });
+ *
+ * const pipeline = traceWorkflow("qa_pipeline", async (question: string) => {
+ *   const results = await search(question);
+ *   return summarize(results);
+ * });
+ * ```
+ */
+import { trace, SpanStatusCode } from "@opentelemetry/api";
+/** Span kind values following OpenLLMetry/OTEL GenAI conventions. */
+const SPAN_KIND_WORKFLOW = "workflow";
+const SPAN_KIND_TASK = "task";
+const SPAN_KIND_AGENT = "agent";
+const SPAN_KIND_TOOL = "tool";
+const TRACER_NAME = "opensearch-genai-sdk";
+/** Maximum size (in characters) for serialized input/output attributes. */
+const MAX_ATTRIBUTE_LENGTH = 10_000;
+/**
+ * Trace a function as a workflow span.
+ *
+ * Use for top-level orchestration functions that coordinate
+ * multiple tasks, agents, or tool calls.
+ *
+ * @param name - Span name for the workflow.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const pipeline = traceWorkflow("qa_pipeline", async (question: string) => {
+ *   const plan = await planSteps(question);
+ *   return await execute(plan);
+ * });
+ * ```
+ */
+export function traceWorkflow(name, fn, options) {
+    return makeWrapper(name, fn, SPAN_KIND_WORKFLOW, options);
+}
+/**
+ * Trace a function as a task span.
+ *
+ * Use for individual units of work within a workflow.
+ *
+ * @param name - Span name for the task.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const summarize = traceTask("summarize", async (text: string) => {
+ *   return llm.generate(`Summarize: ${text}`);
+ * });
+ * ```
+ */
+export function traceTask(name, fn, options) {
+    return makeWrapper(name, fn, SPAN_KIND_TASK, options);
+}
+/**
+ * Trace a function as an agent span.
+ *
+ * Use for autonomous agent logic that makes decisions and
+ * invokes tools.
+ *
+ * @param name - Span name for the agent.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const agent = traceAgent("research_agent", async (query: string) => {
+ *   while (!done) {
+ *     const action = await decideNextAction(query);
+ *     result = await executeAction(action);
+ *   }
+ *   return result;
+ * });
+ * ```
+ */
+export function traceAgent(name, fn, options) {
+    return makeWrapper(name, fn, SPAN_KIND_AGENT, options);
+}
+/**
+ * Trace a function as a tool span.
+ *
+ * Use for tool/function calls invoked by agents.
+ *
+ * @param name - Span name for the tool.
+ * @param fn - The function to wrap.
+ * @param options - Optional trace settings.
+ * @returns A wrapped function with the same signature.
+ *
+ * @example
+ * ```ts
+ * const search = traceTool("web_search", async (query: string) => {
+ *   return searchApi.query(query);
+ * });
+ * ```
+ */
+export function traceTool(name, fn, options) {
+    return makeWrapper(name, fn, SPAN_KIND_TOOL, options);
+}
+/**
+ * Create a wrapper function that wraps the original function in an OTEL span.
+ *
+ * Handles both sync and async functions transparently. When the wrapped
+ * function returns a Promise, the span is ended when the Promise settles.
+ */
+function makeWrapper(spanName, fn, spanKind, options) {
+    const wrappedFn = (...args) => {
+        const tracer = trace.getTracer(TRACER_NAME);
+        return tracer.startActiveSpan(spanName, (span) => {
+            setSpanAttributes(span, spanKind, options?.version, args);
+            try {
+                const result = fn(...args);
+                // Handle async functions: result is a Promise
+                if (result instanceof Promise) {
+                    return result
+                        .then((resolved) => {
+                        setOutput(span, resolved);
+                        span.end();
+                        return resolved;
+                    })
+                        .catch((err) => {
+                        span.setStatus({ code: SpanStatusCode.ERROR, message: String(err) });
+                        span.recordException(err);
+                        span.end();
+                        throw err;
+                    });
+                }
+                // Sync function
+                setOutput(span, result);
+                span.end();
+                return result;
+            }
+            catch (err) {
+                span.setStatus({ code: SpanStatusCode.ERROR, message: String(err) });
+                span.recordException(err);
+                span.end();
+                throw err;
+            }
+        });
+    };
+    // Preserve function name for debugging
+    Object.defineProperty(wrappedFn, "name", { value: fn.name || spanName });
+    return wrappedFn;
+}
+/**
+ * Set standard attributes on a span.
+ */
+function setSpanAttributes(span, spanKind, version, args) {
+    // GenAI / traceloop span kind convention
+    span.setAttribute("traceloop.span.kind", spanKind);
+    span.setAttribute("gen_ai.operation.name", spanKind);
+    if (version !== undefined) {
+        span.setAttribute("traceloop.entity.version", version);
+    }
+    // Capture input (best-effort, don't fail if serialization fails)
+    setInput(span, args);
+}
+/**
+ * Attempt to capture function input as a span attribute.
+ */
+function setInput(span, args) {
+    try {
+        if (args.length === 0)
+            return;
+        const value = args.length === 1 ? args[0] : args;
+        let serialized = JSON.stringify(value, (_key, val) => typeof val === "bigint" ? val.toString() : val);
+        if (serialized.length > MAX_ATTRIBUTE_LENGTH) {
+            serialized = serialized.slice(0, MAX_ATTRIBUTE_LENGTH) + "...(truncated)";
+        }
+        span.setAttribute("traceloop.entity.input", serialized);
+    }
+    catch {
+        // Silently ignore serialization errors
+    }
+}
+/**
+ * Attempt to capture function output as a span attribute.
+ */
+function setOutput(span, result) {
+    try {
+        if (result === undefined || result === null)
+            return;
+        let serialized = JSON.stringify(result, (_key, val) => typeof val === "bigint" ? val.toString() : val);
+        if (serialized.length > MAX_ATTRIBUTE_LENGTH) {
+            serialized = serialized.slice(0, MAX_ATTRIBUTE_LENGTH) + "...(truncated)";
+        }
+        span.setAttribute("traceloop.entity.output", serialized);
+    }
+    catch {
+        // Silently ignore serialization errors
+    }
+}
+//# sourceMappingURL=decorators.js.map

package/dist/decorators.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"decorators.js","sourceRoot":"","sources":["../src/decorators.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,OAAO,EAAE,KAAK,EAAE,cAAc,EAAQ,MAAM,oBAAoB,CAAC;AAEjE,qEAAqE;AACrE,MAAM,kBAAkB,GAAG,UAAU,CAAC;AACtC,MAAM,cAAc,GAAG,MAAM,CAAC;AAC9B,MAAM,eAAe,GAAG,OAAO,CAAC;AAChC,MAAM,cAAc,GAAG,MAAM,CAAC;AAE9B,MAAM,WAAW,GAAG,sBAAsB,CAAC;AAE3C,2EAA2E;AAC3E,MAAM,oBAAoB,GAAG,MAAM,CAAC;AAOpC;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,UAAU,aAAa,CAC3B,IAAY,EACZ,EAA+B,EAC/B,OAAsB;IAEtB,OAAO,WAAW,CAAC,IAAI,EAAE,EAAE,EAAE,kBAAkB,EAAE,OAAO,CAAC,CAAC;AAC5D,CAAC;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,SAAS,CACvB,IAAY,EACZ,EAA+B,EAC/B,OAAsB;IAEtB,OAAO,WAAW,CAAC,IAAI,EAAE,EAAE,EAAE,cAAc,EAAE,OAAO,CAAC,CAAC;AACxD,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,UAAU,UAAU,CACxB,IAAY,EACZ,EAA+B,EAC/B,OAAsB;IAEtB,OAAO,WAAW,CAAC,IAAI,EAAE,EAAE,EAAE,eAAe,EAAE,OAAO,CAAC,CAAC;AACzD,CAAC;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,SAAS,CACvB,IAAY,EACZ,EAA+B,EAC/B,OAAsB;IAEtB,OAAO,WAAW,CAAC,IAAI,EAAE,EAAE,EAAE,cAAc,EAAE,OAAO,CAAC,CAAC;AACxD,CAAC;AAED;;;;;GAKG;AACH,SAAS,WAAW,CAClB,QAAgB,EAChB,EAA+B,EAC/B,QAAgB,EAChB,OAAsB;IAEtB,MAAM,SAAS,GAAG,CAAC,GAAG,IAAW,EAAW,EAAE;QAC5C,MAAM,MAAM,GAAG,KAAK,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC;QAE5C,OAAO,MAAM,CAAC,eAAe,CAAC,QAAQ,EAAE,CAAC,IAAU,EAAE,EAAE;YACrD,iBAAiB,CAAC,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,CAAC;YAE1D,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;gBAE3B,8CAA8C;gBAC9C,IAAI,MAAM,YAAY,OAAO,EAAE,CAAC;oBAC9B,OAAO,MAAM;yBACV,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE;wBACjB,SAAS,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;wBAC1B,IAAI,CAAC,GAAG,EAAE,CAAC;wBACX,OAAO,QAAQ,CAAC;oBAClB,CAAC,CAAC;yBACD,KAAK,CAAC,CAAC,GAAU,EAAE,EAAE;wBACpB,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,cAAc,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;wBACrE,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC;wBAC1B,IAAI,CAAC,GAAG,EAAE,CAAC;wBACX,MAAM,GAAG,CAAC;oBACZ,CAAC,CAAY,CAAC;gBAClB,CAAC;gBAED,gBAAgB;gBAChB,SAAS,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;gBACxB,IAAI,CAAC,GAAG,EAAE,CAAC;gBACX,OAAO,MAAM,CAAC;YAChB,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,cAAc,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBACrE,IAAI,CAAC,eAAe,CAAC,GAAY,CAAC,CAAC;gBACnC,IAAI,CAAC,GAAG,EAAE,CAAC;gBACX,MAAM,GAAG,CAAC;YACZ,CAAC;QACH,CAAC,CAAC,CAAC;IACL,CAAC,CAAC;IAEF,uCAAuC;IACvC,MAAM,CAAC,cAAc,CAAC,SAAS,EAAE,MAAM,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,IAAI,QAAQ,EAAE,CAAC,CAAC;IACzE,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;GAEG;AACH,SAAS,iBAAiB,CACxB,IAAU,EACV,QAAgB,EAChB,OAA2B,EAC3B,IAAe;IAEf,yCAAyC;IACzC,IAAI,CAAC,YAAY,CAAC,qBAAqB,EAAE,QAAQ,CAAC,CAAC;IACnD,IAAI,CAAC,YAAY,CAAC,uBAAuB,EAAE,QAAQ,CAAC,CAAC;IAErD,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,IAAI,CAAC,YAAY,CAAC,0BAA0B,EAAE,OAAO,CAAC,CAAC;IACzD,CAAC;IAED,iEAAiE;IACjE,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;AACvB,CAAC;AAED;;GAEG;AACH,SAAS,QAAQ,CAAC,IAAU,EAAE,IAAe;IAC3C,IAAI,CAAC;QACH,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO;QAE9B,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QACjD,IAAI,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,EAAE,CACnD,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,GAAG,CAC/C,CAAC;QAEF,IAAI,UAAU,CAAC,MAAM,GAAG,oBAAoB,EAAE,CAAC;YAC7C,UAAU,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,oBAAoB,CAAC,GAAG,gBAAgB,CAAC;QAC5E,CAAC;QAED,IAAI,CAAC,YAAY,CAAC,wBAAwB,EAAE,UAAU,CAAC,CAAC;IAC1D,CAAC;IAAC,MAAM,CAAC;QACP,uCAAuC;IACzC,CAAC;AACH,CAAC;AAED;;GAEG;AACH,SAAS,SAAS,CAAC,IAAU,EAAE,MAAe;IAC5C,IAAI,CAAC;QACH,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,IAAI;YAAE,OAAO;QAEpD,IAAI,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,EAAE,CACpD,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,GAAG,CAC/C,CAAC;QAEF,IAAI,UAAU,CAAC,MAAM,GAAG,oBAAoB,EAAE,CAAC;YAC7C,UAAU,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,oBAAoB,CAAC,GAAG,gBAAgB,CAAC;QAC5E,CAAC;QAED,IAAI,CAAC,YAAY,CAAC,yBAAyB,EAAE,UAAU,CAAC,CAAC;IAC3D,CAAC;IAAC,MAAM,CAAC;QACP,uCAAuC;IACzC,CAAC;AACH,CAAC"}

package/dist/evals/evaluate.d.ts

@@ -0,0 +1,97 @@
+/**
+ * Evaluation orchestrator for OpenSearch AI observability.
+ *
+ * Runs a task function across a dataset, applies scorers to each output,
+ * creates OTEL spans for the entire eval run, and stores scores in OpenSearch.
+ */
+import { Score, Scorer } from "./protocol.js";
+/**
+ * Result of a single data point evaluation.
+ */
+export interface EvalResult {
+    /** The input given to the task. */
+    input: unknown;
+    /** The task's output. */
+    output?: unknown;
+    /** The expected output (if provided). */
+    expected?: unknown;
+    /** Dict of scorer name to Score. */
+    scores: Record<string, Score>;
+    /** Error message if the task failed. */
+    error?: string;
+}
+/**
+ * Summary of an evaluation run.
+ */
+export interface EvalSummary {
+    /** The evaluation name. */
+    name: string;
+    /** Per-data-point results. */
+    results: EvalResult[];
+    /** Average score per scorer. */
+    averages: Record<string, number>;
+    /** Number of data points. */
+    total: number;
+    /** Number of failed data points. */
+    errors: number;
+}
+/**
+ * A single data point in the evaluation dataset.
+ */
+export interface EvalDatum {
+    /** The input to the task. */
+    input: unknown;
+    /** The expected output (optional). */
+    expected?: unknown;
+    /** Allow additional fields. */
+    [key: string]: unknown;
+}
+export interface EvaluateOptions {
+    /** Name for this evaluation run. */
+    name: string;
+    /** List of dicts with "input" and optional "expected" keys, or a callable that returns such a list. */
+    data: EvalDatum[] | (() => EvalDatum[]);
+    /** Function that takes an input and returns output. Can be sync or async. */
+    task: (input: unknown) => unknown | Promise<unknown>;
+    /** List of scorer instances. */
+    scores: Scorer[];
+    /** Whether to emit scores as separate OTEL spans for Data Prepper routing. Defaults to true. */
+    emitScores?: boolean;
+}
+/**
+ * Run an evaluation: dataset -> task -> scorers -> OTEL spans.
+ *
+ * For each data point, runs the task function to produce output,
+ * then applies all scorers. Creates OTEL spans for the entire flow.
+ * Scores are emitted as OTEL spans through the same exporter pipeline.
+ *
+ * @example
+ * ```ts
+ * import { evaluate } from "@opensearch-project/genai-sdk";
+ * import { Score } from "@opensearch-project/genai-sdk/evals";
+ *
+ * const exactMatch: Scorer = {
+ *   name: "exact_match",
+ *   score({ output, expected }) {
+ *     const match = output.trim().toLowerCase() === (expected ?? "").trim().toLowerCase();
+ *     return { name: "exact_match", value: match ? 1.0 : 0.0 };
+ *   },
+ * };
+ *
+ * const results = evaluate({
+ *   name: "qa-eval",
+ *   data: [
+ *     { input: "Capital of France?", expected: "Paris" },
+ *     { input: "2+2?", expected: "4" },
+ *   ],
+ *   task: (input) => myLlmCall(input as string),
+ *   scores: [exactMatch],
+ * });
+ * ```
+ */
+export declare function evaluate(options: EvaluateOptions): EvalSummary;
+/**
+ * Format an EvalSummary as a human-readable string.
+ */
+export declare function formatEvalSummary(summary: EvalSummary): string;
+//# sourceMappingURL=evaluate.d.ts.map

package/dist/evals/evaluate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluate.d.ts","sourceRoot":"","sources":["../../src/evals/evaluate.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,KAAK,EAAE,MAAM,EAAc,MAAM,eAAe,CAAC;AAK1D;;GAEG;AACH,MAAM,WAAW,UAAU;IACzB,mCAAmC;IACnC,KAAK,EAAE,OAAO,CAAC;IACf,yBAAyB;IACzB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,yCAAyC;IACzC,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,oCAAoC;IACpC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IAC9B,wCAAwC;IACxC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B,2BAA2B;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,8BAA8B;IAC9B,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,gCAAgC;IAChC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,6BAA6B;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,oCAAoC;IACpC,MAAM,EAAE,MAAM,CAAC;CAChB;AAED;;GAEG;AACH,MAAM,WAAW,SAAS;IACxB,6BAA6B;IAC7B,KAAK,EAAE,OAAO,CAAC;IACf,sCAAsC;IACtC,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,+BAA+B;IAC/B,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,oCAAoC;IACpC,IAAI,EAAE,MAAM,CAAC;IACb,uGAAuG;IACvG,IAAI,EAAE,SAAS,EAAE,GAAG,CAAC,MAAM,SAAS,EAAE,CAAC,CAAC;IACxC,6EAA6E;IAC7E,IAAI,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IACrD,gCAAgC;IAChC,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,gGAAgG;IAChG,UAAU,CAAC,EAAE,OAAO,CAAC;CACtB;AAaD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AACH,wBAAgB,QAAQ,CAAC,OAAO,EAAE,eAAe,GAAG,WAAW,CAiL9D;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,WAAW,GAAG,MAAM,CAM9D"}

package/dist/evals/evaluate.js

@@ -0,0 +1,202 @@
+/**
+ * Evaluation orchestrator for OpenSearch AI observability.
+ *
+ * Runs a task function across a dataset, applies scorers to each output,
+ * creates OTEL spans for the entire eval run, and stores scores in OpenSearch.
+ */
+import { trace, SpanStatusCode } from "@opentelemetry/api";
+import { adaptScore } from "./protocol.js";
+import { score as submitScore } from "../score.js";
+const TRACER_NAME = "opensearch-genai-sdk-evals";
+/**
+ * Format a span context ID as a hex string.
+ */
+function formatTraceId(traceId) {
+    return traceId;
+}
+function formatSpanId(spanId) {
+    return spanId;
+}
+/**
+ * Run an evaluation: dataset -> task -> scorers -> OTEL spans.
+ *
+ * For each data point, runs the task function to produce output,
+ * then applies all scorers. Creates OTEL spans for the entire flow.
+ * Scores are emitted as OTEL spans through the same exporter pipeline.
+ *
+ * @example
+ * ```ts
+ * import { evaluate } from "@opensearch-project/genai-sdk";
+ * import { Score } from "@opensearch-project/genai-sdk/evals";
+ *
+ * const exactMatch: Scorer = {
+ *   name: "exact_match",
+ *   score({ output, expected }) {
+ *     const match = output.trim().toLowerCase() === (expected ?? "").trim().toLowerCase();
+ *     return { name: "exact_match", value: match ? 1.0 : 0.0 };
+ *   },
+ * };
+ *
+ * const results = evaluate({
+ *   name: "qa-eval",
+ *   data: [
+ *     { input: "Capital of France?", expected: "Paris" },
+ *     { input: "2+2?", expected: "4" },
+ *   ],
+ *   task: (input) => myLlmCall(input as string),
+ *   scores: [exactMatch],
+ * });
+ * ```
+ */
+export function evaluate(options) {
+    const { name, data, task, scores: scorers, emitScores = true } = options;
+    const tracer = trace.getTracer(TRACER_NAME);
+    // Resolve data if callable
+    const dataset = typeof data === "function" ? data() : data;
+    const summary = {
+        name,
+        results: [],
+        averages: {},
+        total: dataset.length,
+        errors: 0,
+    };
+    tracer.startActiveSpan("evaluate", {
+        attributes: {
+            "eval.name": name,
+            "eval.dataset_size": dataset.length,
+            "eval.scorer_count": scorers.length,
+        },
+    }, (evalSpan) => {
+        for (let i = 0; i < dataset.length; i++) {
+            const datum = dataset[i];
+            const inputVal = datum.input;
+            const expectedVal = datum.expected;
+            const evalResult = {
+                input: inputVal,
+                expected: expectedVal,
+                scores: {},
+            };
+            tracer.startActiveSpan("eval_item", {
+                attributes: {
+                    "eval.item.index": i,
+                    "eval.item.input": String(inputVal).slice(0, 1000),
+                },
+            }, (itemSpan) => {
+                // Run the task
+                let output;
+                let taskFailed = false;
+                tracer.startActiveSpan("eval_task", (taskSpan) => {
+                    try {
+                        const result = task(inputVal);
+                        // Handle async tasks by noting we don't support them in sync evaluate()
+                        // For a sync evaluate, the task should be sync. Async tasks would need
+                        // an evaluateAsync() variant.
+                        if (result instanceof Promise) {
+                            console.warn("[opensearch-genai-sdk] Async task detected in sync evaluate(). " +
+                                "The promise will not be awaited. Use sync tasks or implement evaluateAsync().");
+                        }
+                        output = result;
+                        evalResult.output = output;
+                        taskSpan.setAttribute("eval.task.output", String(output).slice(0, 1000));
+                    }
+                    catch (err) {
+                        evalResult.error = String(err);
+                        summary.errors++;
+                        taskSpan.setStatus({ code: SpanStatusCode.ERROR, message: String(err) });
+                        taskSpan.recordException(err);
+                        taskFailed = true;
+                    }
+                    taskSpan.end();
+                });
+                if (!taskFailed) {
+                    // Run each scorer
+                    for (const scorer of scorers) {
+                        const scorerName = scorer.name;
+                        tracer.startActiveSpan(`eval_score.${scorerName}`, { attributes: { "eval.scorer.name": scorerName } }, (scoreSpan) => {
+                            try {
+                                const rawResult = scorer.score({
+                                    input: String(inputVal),
+                                    output: String(output),
+                                    expected: expectedVal !== undefined ? String(expectedVal) : undefined,
+                                });
+                                const scoreObj = adaptScore(scorerName, rawResult);
+                                evalResult.scores[scorerName] = scoreObj;
+                                if (scoreObj.value !== undefined) {
+                                    scoreSpan.setAttribute("eval.score.value", scoreObj.value);
+                                }
+                                if (scoreObj.label) {
+                                    scoreSpan.setAttribute("eval.score.label", scoreObj.label);
+                                }
+                                if (scoreObj.rationale) {
+                                    scoreSpan.setAttribute("eval.score.rationale", scoreObj.rationale.slice(0, 500));
+                                }
+                            }
+                            catch (err) {
+                                console.warn(`[opensearch-genai-sdk] Scorer ${scorerName} failed on item ${i}: ${err}`);
+                                scoreSpan.setStatus({ code: SpanStatusCode.ERROR, message: String(err) });
+                                scoreSpan.recordException(err);
+                            }
+                            scoreSpan.end();
+                        });
+                    }
+                    // Emit scores as OTEL spans
+                    if (emitScores) {
+                        const spanContext = itemSpan.spanContext();
+                        const traceIdHex = formatTraceId(spanContext.traceId);
+                        const spanIdHex = formatSpanId(spanContext.spanId);
+                        for (const [scorerName, scoreObj] of Object.entries(evalResult.scores)) {
+                            try {
+                                submitScore({
+                                    name: scorerName,
+                                    value: scoreObj.value,
+                                    traceId: traceIdHex,
+                                    spanId: spanIdHex,
+                                    label: scoreObj.label,
+                                    rationale: scoreObj.rationale,
+                                    source: "eval",
+                                    metadata: { eval_name: name, item_index: i },
+                                });
+                            }
+                            catch (err) {
+                                console.warn(`[opensearch-genai-sdk] Failed to emit score ${scorerName}: ${err}`);
+                            }
+                        }
+                    }
+                }
+                summary.results.push(evalResult);
+                itemSpan.end();
+            });
+        }
+        // Compute averages
+        const scoreTotals = {};
+        for (const result of summary.results) {
+            for (const [scorerName, scoreObj] of Object.entries(result.scores)) {
+                if (scoreObj.value !== undefined) {
+                    if (!scoreTotals[scorerName]) {
+                        scoreTotals[scorerName] = [];
+                    }
+                    scoreTotals[scorerName].push(scoreObj.value);
+                }
+            }
+        }
+        for (const [scorerName, values] of Object.entries(scoreTotals)) {
+            const avg = values.reduce((a, b) => a + b, 0) / values.length;
+            summary.averages[scorerName] = avg;
+            evalSpan.setAttribute(`eval.avg.${scorerName}`, avg);
+        }
+        evalSpan.setAttribute("eval.errors", summary.errors);
+        evalSpan.end();
+    });
+    return summary;
+}
+/**
+ * Format an EvalSummary as a human-readable string.
+ */
+export function formatEvalSummary(summary) {
+    const parts = [`Eval: ${summary.name} (${summary.total} samples, ${summary.errors} errors)`];
+    for (const [scorerName, avg] of Object.entries(summary.averages)) {
+        parts.push(`  ${scorerName}: ${avg.toFixed(3)}`);
+    }
+    return parts.join("\n");
+}
+//# sourceMappingURL=evaluate.js.map

package/dist/evals/evaluate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluate.js","sourceRoot":"","sources":["../../src/evals/evaluate.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,KAAK,EAAE,cAAc,EAAQ,MAAM,oBAAoB,CAAC;AACjE,OAAO,EAAiB,UAAU,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,EAAE,KAAK,IAAI,WAAW,EAAE,MAAM,aAAa,CAAC;AAEnD,MAAM,WAAW,GAAG,4BAA4B,CAAC;AA2DjD;;GAEG;AACH,SAAS,aAAa,CAAC,OAAe;IACpC,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,YAAY,CAAC,MAAc;IAClC,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AACH,MAAM,UAAU,QAAQ,CAAC,OAAwB;IAC/C,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,GAAG,IAAI,EAAE,GAAG,OAAO,CAAC;IAEzE,MAAM,MAAM,GAAG,KAAK,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC;IAE5C,2BAA2B;IAC3B,MAAM,OAAO,GAAgB,OAAO,IAAI,KAAK,UAAU,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IAExE,MAAM,OAAO,GAAgB;QAC3B,IAAI;QACJ,OAAO,EAAE,EAAE;QACX,QAAQ,EAAE,EAAE;QACZ,KAAK,EAAE,OAAO,CAAC,MAAM;QACrB,MAAM,EAAE,CAAC;KACV,CAAC;IAEF,MAAM,CAAC,eAAe,CACpB,UAAU,EACV;QACE,UAAU,EAAE;YACV,WAAW,EAAE,IAAI;YACjB,mBAAmB,EAAE,OAAO,CAAC,MAAM;YACnC,mBAAmB,EAAE,OAAO,CAAC,MAAM;SACpC;KACF,EACD,CAAC,QAAc,EAAE,EAAE;QACjB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACxC,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;YACzB,MAAM,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC;YAC7B,MAAM,WAAW,GAAG,KAAK,CAAC,QAAQ,CAAC;YAEnC,MAAM,UAAU,GAAe;gBAC7B,KAAK,EAAE,QAAQ;gBACf,QAAQ,EAAE,WAAW;gBACrB,MAAM,EAAE,EAAE;aACX,CAAC;YAEF,MAAM,CAAC,eAAe,CACpB,WAAW,EACX;gBACE,UAAU,EAAE;oBACV,iBAAiB,EAAE,CAAC;oBACpB,iBAAiB,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;iBACnD;aACF,EACD,CAAC,QAAc,EAAE,EAAE;gBACjB,eAAe;gBACf,IAAI,MAAe,CAAC;gBACpB,IAAI,UAAU,GAAG,KAAK,CAAC;gBAEvB,MAAM,CAAC,eAAe,CAAC,WAAW,EAAE,CAAC,QAAc,EAAE,EAAE;oBACrD,IAAI,CAAC;wBACH,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC;wBAE9B,wEAAwE;wBACxE,uEAAuE;wBACvE,8BAA8B;wBAC9B,IAAI,MAAM,YAAY,OAAO,EAAE,CAAC;4BAC9B,OAAO,CAAC,IAAI,CACV,iEAAiE;gCAC/D,+EAA+E,CAClF,CAAC;wBACJ,CAAC;wBAED,MAAM,GAAG,MAAM,CAAC;wBAChB,UAAU,CAAC,MAAM,GAAG,MAAM,CAAC;wBAC3B,QAAQ,CAAC,YAAY,CAAC,kBAAkB,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;oBAC3E,CAAC;oBAAC,OAAO,GAAG,EAAE,CAAC;wBACb,UAAU,CAAC,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;wBAC/B,OAAO,CAAC,MAAM,EAAE,CAAC;wBACjB,QAAQ,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,cAAc,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;wBACzE,QAAQ,CAAC,eAAe,CAAC,GAAY,CAAC,CAAC;wBACvC,UAAU,GAAG,IAAI,CAAC;oBACpB,CAAC;oBACD,QAAQ,CAAC,GAAG,EAAE,CAAC;gBACjB,CAAC,CAAC,CAAC;gBAEH,IAAI,CAAC,UAAU,EAAE,CAAC;oBAChB,kBAAkB;oBAClB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;wBAC7B,MAAM,UAAU,GAAG,MAAM,CAAC,IAAI,CAAC;wBAE/B,MAAM,CAAC,eAAe,CACpB,cAAc,UAAU,EAAE,EAC1B,EAAE,UAAU,EAAE,EAAE,kBAAkB,EAAE,UAAU,EAAE,EAAE,EAClD,CAAC,SAAe,EAAE,EAAE;4BAClB,IAAI,CAAC;gCACH,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC;oCAC7B,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC;oCACvB,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC;oCACtB,QAAQ,EAAE,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,SAAS;iCACtE,CAAC,CAAC;gCACH,MAAM,QAAQ,GAAG,UAAU,CAAC,UAAU,EAAE,SAAS,CAAC,CAAC;gCACnD,UAAU,CAAC,MAAM,CAAC,UAAU,CAAC,GAAG,QAAQ,CAAC;gCAEzC,IAAI,QAAQ,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;oCACjC,SAAS,CAAC,YAAY,CAAC,kBAAkB,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;gCAC7D,CAAC;gCACD,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;oCACnB,SAAS,CAAC,YAAY,CAAC,kBAAkB,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;gCAC7D,CAAC;gCACD,IAAI,QAAQ,CAAC,SAAS,EAAE,CAAC;oCACvB,SAAS,CAAC,YAAY,CACpB,sBAAsB,EACtB,QAAQ,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CACjC,CAAC;gCACJ,CAAC;4BACH,CAAC;4BAAC,OAAO,GAAG,EAAE,CAAC;gCACb,OAAO,CAAC,IAAI,CACV,iCAAiC,UAAU,mBAAmB,CAAC,KAAK,GAAG,EAAE,CAC1E,CAAC;gCACF,SAAS,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,cAAc,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gCAC1E,SAAS,CAAC,eAAe,CAAC,GAAY,CAAC,CAAC;4BAC1C,CAAC;4BACD,SAAS,CAAC,GAAG,EAAE,CAAC;wBAClB,CAAC,CACF,CAAC;oBACJ,CAAC;oBAED,4BAA4B;oBAC5B,IAAI,UAAU,EAAE,CAAC;wBACf,MAAM,WAAW,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC;wBAC3C,MAAM,UAAU,GAAG,aAAa,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;wBACtD,MAAM,SAAS,GAAG,YAAY,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;wBAEnD,KAAK,MAAM,CAAC,UAAU,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC;4BACvE,IAAI,CAAC;gCACH,WAAW,CAAC;oCACV,IAAI,EAAE,UAAU;oCAChB,KAAK,EAAE,QAAQ,CAAC,KAAK;oCACrB,OAAO,EAAE,UAAU;oCACnB,MAAM,EAAE,SAAS;oCACjB,KAAK,EAAE,QAAQ,CAAC,KAAK;oCACrB,SAAS,EAAE,QAAQ,CAAC,SAAS;oCAC7B,MAAM,EAAE,MAAM;oCACd,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE;iCAC7C,CAAC,CAAC;4BACL,CAAC;4BAAC,OAAO,GAAG,EAAE,CAAC;gCACb,OAAO,CAAC,IAAI,CACV,+CAA+C,UAAU,KAAK,GAAG,EAAE,CACpE,CAAC;4BACJ,CAAC;wBACH,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;gBACjC,QAAQ,CAAC,GAAG,EAAE,CAAC;YACjB,CAAC,CACF,CAAC;QACJ,CAAC;QAED,mBAAmB;QACnB,MAAM,WAAW,GAA6B,EAAE,CAAC;QACjD,KAAK,MAAM,MAAM,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;YACrC,KAAK,MAAM,CAAC,UAAU,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;gBACnE,IAAI,QAAQ,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;oBACjC,IAAI,CAAC,WAAW,CAAC,UAAU,CAAC,EAAE,CAAC;wBAC7B,WAAW,CAAC,UAAU,CAAC,GAAG,EAAE,CAAC;oBAC/B,CAAC;oBACD,WAAW,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;gBAC/C,CAAC;YACH,CAAC;QACH,CAAC;QAED,KAAK,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,CAAC;YAC/D,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;YAC9D,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,GAAG,GAAG,CAAC;YACnC,QAAQ,CAAC,YAAY,CAAC,YAAY,UAAU,EAAE,EAAE,GAAG,CAAC,CAAC;QACvD,CAAC;QAED,QAAQ,CAAC,YAAY,CAAC,aAAa,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QACrD,QAAQ,CAAC,GAAG,EAAE,CAAC;IACjB,CAAC,CACF,CAAC;IAEF,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAoB;IACpD,MAAM,KAAK,GAAG,CAAC,SAAS,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,KAAK,aAAa,OAAO,CAAC,MAAM,UAAU,CAAC,CAAC;IAC7F,KAAK,MAAM,CAAC,UAAU,EAAE,GAAG,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QACjE,KAAK,CAAC,IAAI,CAAC,KAAK,UAAU,KAAK,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IACnD,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/evals/index.d.ts

@@ -0,0 +1,12 @@
+/**
+ * Evaluation framework for OpenSearch AI observability.
+ *
+ * Provides the evaluate() orchestrator and Scorer interface for running
+ * evaluations on LLM outputs. Compatible with autoevals, phoenix-evals,
+ * or any custom scorer that matches the Scorer interface.
+ */
+export type { Score, Scorer } from "./protocol.js";
+export { adaptScore } from "./protocol.js";
+export type { EvalResult, EvalSummary, EvalDatum, EvaluateOptions } from "./evaluate.js";
+export { evaluate, formatEvalSummary } from "./evaluate.js";
+//# sourceMappingURL=index.d.ts.map

package/dist/evals/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/evals/index.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,YAAY,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,eAAe,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC3C,YAAY,EAAE,UAAU,EAAE,WAAW,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AACzF,OAAO,EAAE,QAAQ,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC"}