@lunora/testing 1.0.0-alpha.50 → 1.0.0-alpha.52

package/dist/index.d.mts

@@ -110,6 +110,53 @@ declare const agentHarness: (agent: AgentDefinition, options: AgentHarnessOption
 declare const finalTurn: (text: string, extra?: Partial<AgentGenerateResult>) => AgentGenerateResult;
 /** A single-tool-call turn: the loop runs the named tool with `input`. */
 declare const toolCallTurn: (id: string, name: string, input: unknown, text?: string) => AgentGenerateResult;
+/** The primitive an eval attaches to a span: a numeric score or a string label. */
+type EvaluationAttributeValue = number | string;
+/**
+ * Structural slice of the post-hoc span handle `ctx.trace` hands its body (see the
+ * server `SpanHandle`) — enough to attach an eval's attributes. Declared here
+ * rather than imported so `@lunora/testing` takes no dependency on `@lunora/server`
+ * or `@lunora/do`; the real handle is assignable to it.
+ */
+interface EvaluationSpanHandle {
+  setAttributes: (fields: Record<string, EvaluationAttributeValue>) => void;
+}
+/** One eval verdict to emit. */
+interface RecordEvaluationInput {
+  /**
+   * Optional categorical label (e.g. `"pass"` / `"fail"` / a rubric bucket),
+   * emitted as the `.label` attribute. Omitted → no label attribute.
+   */
+  label?: string;
+  /**
+   * The scorer/evaluation name — becomes the key's name segment. Any character
+   * outside `[A-Za-z0-9._-]` is replaced with `_` so a scorer name carrying a
+   * colon (e.g. `"contains:shipped"`) still yields a well-formed attribute key.
+   */
+  name: string;
+  /** The numeric score (typically `[0, 1]`), emitted as the `.score` attribute. */
+  score: number;
+  /**
+   * Optional generation span to attach the attributes to — the post-hoc
+   * `SpanHandle` a `ctx.trace` body receives. Omitted → nothing is attached and
+   * the caller uses the returned bag for a standalone event/metric.
+   */
+  span?: EvaluationSpanHandle;
+}
+/**
+ * Build the `gen_ai.evaluation.NAME.*` attribute bag for one eval verdict — the
+ * `.score` (number) always, the `.label` (string) when a label is given. Exported
+ * so a caller can emit the score as a standalone event/metric without a span.
+ */
+declare const evaluationAttributes: (input: Pick<RecordEvaluationInput, "label" | "name" | "score">) => Record<string, EvaluationAttributeValue>;
+/**
+ * Emit one eval verdict as `gen_ai.evaluation.NAME.*` attributes. Attaches them to
+ * `input.span` when supplied (the post-hoc generation-span handle), and always
+ * returns the attribute bag so a span-less caller can ship it as an eval
+ * event/metric. Privacy-safe: only the name, score, and optional label are
+ * emitted — never the graded prompt or output.
+ */
+declare const recordEvaluation: (input: RecordEvaluationInput) => Record<string, EvaluationAttributeValue>;
 /** A pending job entry in the fake scheduler queue. */
 interface FakeScheduledJob extends ScheduledJob {
   /** The args the job was scheduled with. */
@@ -460,4 +507,4 @@ declare const scoreSample: (sample: ScorerSample, scorers: ReadonlyArray<Scorer>
  * Returns per-case results plus the mean of their averages.
  */
 declare const evaluate: (cases: ReadonlyArray<EvalCase>, produce: (input: string) => Promise<string> | string, scorers: ReadonlyArray<Scorer>) => Promise<EvalResult>;
-export { type AgentHarness, type AgentHarnessOptions, type AgentRunOverrides, type EvalCase, type EvalItemResult, type EvalResult, type FakeScheduledJob, type FakeSchedulerControls, type FunctionRegistry, type HarnessDispatch, type HarnessMessage, type HarnessThread, type LunoraTestOptions, type ScheduledJobFailure, type ScoreResult, type Scorer, type ScorerSample, type SweepOptions, type TestHarness, type TestIdentity, type TestSubscription, agentHarness, containsScorer, evaluate, exactMatchScorer, finalTurn, keywordScorer, llmScorer, lunoraTest, regexScorer, scoreSample, toolCallTurn };
+export { type AgentHarness, type AgentHarnessOptions, type AgentRunOverrides, type EvalCase, type EvalItemResult, type EvalResult, type EvaluationAttributeValue, type EvaluationSpanHandle, type FakeScheduledJob, type FakeSchedulerControls, type FunctionRegistry, type HarnessDispatch, type HarnessMessage, type HarnessThread, type LunoraTestOptions, type RecordEvaluationInput, type ScheduledJobFailure, type ScoreResult, type Scorer, type ScorerSample, type SweepOptions, type TestHarness, type TestIdentity, type TestSubscription, agentHarness, containsScorer, evaluate, evaluationAttributes, exactMatchScorer, finalTurn, keywordScorer, llmScorer, lunoraTest, recordEvaluation, regexScorer, scoreSample, toolCallTurn };

package/dist/index.d.ts

@@ -110,6 +110,53 @@ declare const agentHarness: (agent: AgentDefinition, options: AgentHarnessOption
 declare const finalTurn: (text: string, extra?: Partial<AgentGenerateResult>) => AgentGenerateResult;
 /** A single-tool-call turn: the loop runs the named tool with `input`. */
 declare const toolCallTurn: (id: string, name: string, input: unknown, text?: string) => AgentGenerateResult;
+/** The primitive an eval attaches to a span: a numeric score or a string label. */
+type EvaluationAttributeValue = number | string;
+/**
+ * Structural slice of the post-hoc span handle `ctx.trace` hands its body (see the
+ * server `SpanHandle`) — enough to attach an eval's attributes. Declared here
+ * rather than imported so `@lunora/testing` takes no dependency on `@lunora/server`
+ * or `@lunora/do`; the real handle is assignable to it.
+ */
+interface EvaluationSpanHandle {
+  setAttributes: (fields: Record<string, EvaluationAttributeValue>) => void;
+}
+/** One eval verdict to emit. */
+interface RecordEvaluationInput {
+  /**
+   * Optional categorical label (e.g. `"pass"` / `"fail"` / a rubric bucket),
+   * emitted as the `.label` attribute. Omitted → no label attribute.
+   */
+  label?: string;
+  /**
+   * The scorer/evaluation name — becomes the key's name segment. Any character
+   * outside `[A-Za-z0-9._-]` is replaced with `_` so a scorer name carrying a
+   * colon (e.g. `"contains:shipped"`) still yields a well-formed attribute key.
+   */
+  name: string;
+  /** The numeric score (typically `[0, 1]`), emitted as the `.score` attribute. */
+  score: number;
+  /**
+   * Optional generation span to attach the attributes to — the post-hoc
+   * `SpanHandle` a `ctx.trace` body receives. Omitted → nothing is attached and
+   * the caller uses the returned bag for a standalone event/metric.
+   */
+  span?: EvaluationSpanHandle;
+}
+/**
+ * Build the `gen_ai.evaluation.NAME.*` attribute bag for one eval verdict — the
+ * `.score` (number) always, the `.label` (string) when a label is given. Exported
+ * so a caller can emit the score as a standalone event/metric without a span.
+ */
+declare const evaluationAttributes: (input: Pick<RecordEvaluationInput, "label" | "name" | "score">) => Record<string, EvaluationAttributeValue>;
+/**
+ * Emit one eval verdict as `gen_ai.evaluation.NAME.*` attributes. Attaches them to
+ * `input.span` when supplied (the post-hoc generation-span handle), and always
+ * returns the attribute bag so a span-less caller can ship it as an eval
+ * event/metric. Privacy-safe: only the name, score, and optional label are
+ * emitted — never the graded prompt or output.
+ */
+declare const recordEvaluation: (input: RecordEvaluationInput) => Record<string, EvaluationAttributeValue>;
 /** A pending job entry in the fake scheduler queue. */
 interface FakeScheduledJob extends ScheduledJob {
   /** The args the job was scheduled with. */
@@ -460,4 +507,4 @@ declare const scoreSample: (sample: ScorerSample, scorers: ReadonlyArray<Scorer>
  * Returns per-case results plus the mean of their averages.
  */
 declare const evaluate: (cases: ReadonlyArray<EvalCase>, produce: (input: string) => Promise<string> | string, scorers: ReadonlyArray<Scorer>) => Promise<EvalResult>;
-export { type AgentHarness, type AgentHarnessOptions, type AgentRunOverrides, type EvalCase, type EvalItemResult, type EvalResult, type FakeScheduledJob, type FakeSchedulerControls, type FunctionRegistry, type HarnessDispatch, type HarnessMessage, type HarnessThread, type LunoraTestOptions, type ScheduledJobFailure, type ScoreResult, type Scorer, type ScorerSample, type SweepOptions, type TestHarness, type TestIdentity, type TestSubscription, agentHarness, containsScorer, evaluate, exactMatchScorer, finalTurn, keywordScorer, llmScorer, lunoraTest, regexScorer, scoreSample, toolCallTurn };
+export { type AgentHarness, type AgentHarnessOptions, type AgentRunOverrides, type EvalCase, type EvalItemResult, type EvalResult, type EvaluationAttributeValue, type EvaluationSpanHandle, type FakeScheduledJob, type FakeSchedulerControls, type FunctionRegistry, type HarnessDispatch, type HarnessMessage, type HarnessThread, type LunoraTestOptions, type RecordEvaluationInput, type ScheduledJobFailure, type ScoreResult, type Scorer, type ScorerSample, type SweepOptions, type TestHarness, type TestIdentity, type TestSubscription, agentHarness, containsScorer, evaluate, evaluationAttributes, exactMatchScorer, finalTurn, keywordScorer, llmScorer, lunoraTest, recordEvaluation, regexScorer, scoreSample, toolCallTurn };

package/dist/index.mjs

@@ -1,4 +1,5 @@
 export { agentHarness, finalTurn, toolCallTurn } from './packem_shared/agentHarness-GTPhWdQ8.mjs';
-export { lunoraTest } from './packem_shared/lunoraTest-CHix6VWB.mjs';
+export { evaluationAttributes, recordEvaluation } from './packem_shared/evaluationAttributes-C08Q5MOX.mjs';
+export { lunoraTest } from './packem_shared/lunoraTest-Dv_w_v9z.mjs';
 export { containsScorer, evaluate, exactMatchScorer, keywordScorer, llmScorer, regexScorer, scoreSample } from './packem_shared/containsScorer-DGwtvUNp.mjs';
 export { extractLink, listCapturedMail, waitForMail } from '@lunora/mail/testing';

package/dist/packem_shared/evaluationAttributes-C08Q5MOX.mjs

@@ -0,0 +1,31 @@
+import { LunoraError } from '@lunora/errors';
+
+const SAFE_NAME_CHAR = /[\w.-]/u;
+const sanitizeName = (name) => {
+  let out = "";
+  for (const char of name) {
+    out += SAFE_NAME_CHAR.test(char) ? char : "_";
+  }
+  return out;
+};
+const evaluationAttributes = (input) => {
+  if (typeof input.name !== "string" || input.name.length === 0) {
+    throw new LunoraError("BAD_REQUEST", "@lunora/testing: recordEvaluation requires a non-empty `name`");
+  }
+  if (typeof input.score !== "number" || !Number.isFinite(input.score)) {
+    throw new LunoraError("BAD_REQUEST", "@lunora/testing: recordEvaluation `score` must be a finite number");
+  }
+  const key = sanitizeName(input.name);
+  const attributes = { [`gen_ai.evaluation.${key}.score`]: input.score };
+  if (input.label !== void 0) {
+    attributes[`gen_ai.evaluation.${key}.label`] = input.label;
+  }
+  return attributes;
+};
+const recordEvaluation = (input) => {
+  const attributes = evaluationAttributes(input);
+  input.span?.setAttributes(attributes);
+  return attributes;
+};
+
+export { evaluationAttributes, recordEvaluation };

package/dist/packem_shared/{lunoraTest-CHix6VWB.mjs → lunoraTest-Dv_w_v9z.mjs}

@@ -151,7 +151,11 @@ const stubProxy = (surface) => /* @__PURE__ */ new Proxy((..._args) => unavailab
   apply: () => unavailable(surface),
   get: () => unavailable(surface)
 });
-const passthroughTrace = async (_name, function_) => await function_(passthroughTrace);
+const noopSpan = {
+  setAttribute: () => void 0,
+  setAttributes: () => void 0
+};
+const passthroughTrace = async (_name, function_) => await function_(passthroughTrace, noopSpan);
 const noopMetrics = {
   count: () => void 0,
   gauge: () => void 0,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lunora/testing",
-  "version": "1.0.0-alpha.50",
+  "version": "1.0.0-alpha.52",
   "description": "Testing toolkit for Lunora: an in-memory harness for queries, mutations, and actions",
   "keywords": [
     "cloudflare",
@@ -46,11 +46,11 @@
     "access": "public"
   },
   "dependencies": {
-    "@lunora/agent": "1.0.0-alpha.7",
-    "@lunora/do": "1.0.0-alpha.37",
-    "@lunora/errors": "1.0.0-alpha.6",
-    "@lunora/mail": "1.0.0-alpha.16",
-    "@lunora/server": "1.0.0-alpha.29"
+    "@lunora/agent": "1.0.0-alpha.9",
+    "@lunora/do": "1.0.0-alpha.41",
+    "@lunora/errors": "1.0.0-alpha.7",
+    "@lunora/mail": "1.0.0-alpha.17",
+    "@lunora/server": "1.0.0-alpha.30"
   },
   "engines": {
     "node": "^22.15.0 || >=24.11.0"