fix: reject out-of-range embedding-similarity thresholds

src/scorers/embedding-similarity.ts

@@ -1,6 +1,6 @@
 import type { Scorer, ScoreContext, ScorerSpec } from "../types.js";
 import { localEmbedding } from "../providers/mock.js";
-import { result } from "./util.js";
+import { parseThreshold, result } from "./util.js";
 
 /** Cosine similarity between two equal-length vectors. */
 export function cosineSimilarity(a: number[], b: number[]): number {
@@ -38,7 +38,14 @@ export const embeddingSimilarityScorer: Scorer = {
     if (expected === undefined) {
       return result(spec, { score: 0, passed: false, reason: "no expected reference provided" });
     }
-    const threshold = typeof spec.threshold === "number" ? spec.threshold : 0.8;
+    const threshold = parseThreshold(spec, 0.8);
+    if (threshold === null) {
+      return result(spec, {
+        score: 0,
+        passed: false,
+        reason: "invalid threshold (must be a finite number in [0, 1])",
+      });
+    }
 
     const embed = ctx.provider.embed
       ? (t: string) => ctx.provider.embed!(t)

src/scorers/util.ts

@@ -6,6 +6,13 @@ export function clamp01(n: number): number {
   return Math.min(1, Math.max(0, n));
 }
 
+/** Parse a pass threshold in [0, 1], or null when missing/invalid. */
+export function parseThreshold(spec: ScorerSpec, defaultVal: number): number | null {
+  const t = typeof spec.threshold === "number" ? spec.threshold : defaultVal;
+  if (!Number.isFinite(t) || t < 0 || t > 1) return null;
+  return t;
+}
+
 /** Read the weight from a spec, defaulting to 1 and rejecting negatives. */
 export function specWeight(spec: ScorerSpec): number {
   const w = typeof spec.weight === "number" ? spec.weight : 1;

tests/scorers.test.ts

@@ -145,6 +145,33 @@ describe("embedding-similarity", () => {
   it("cosineSimilarity of orthogonal vectors is 0", () => {
     expect(cosineSimilarity([1, 0], [0, 1])).toBe(0);
   });
+  it("rejects NaN threshold", async () => {
+    const r = await run(
+      embeddingSimilarityScorer,
+      { type: "embedding-similarity", expected: "the quick brown fox", threshold: NaN },
+      ctx("the quick brown fox"),
+    );
+    expect(r.passed).toBe(false);
+    expect(r.reason).toMatch(/invalid threshold/i);
+  });
+  it("rejects negative threshold", async () => {
+    const r = await run(
+      embeddingSimilarityScorer,
+      { type: "embedding-similarity", expected: "x", threshold: -5 },
+      ctx("totally unrelated garbage"),
+    );
+    expect(r.passed).toBe(false);
+    expect(r.reason).toMatch(/invalid threshold/i);
+  });
+  it("rejects threshold above 1", async () => {
+    const r = await run(
+      embeddingSimilarityScorer,
+      { type: "embedding-similarity", expected: "the quick brown fox", threshold: 2 },
+      ctx("the quick brown fox"),
+    );
+    expect(r.passed).toBe(false);
+    expect(r.reason).toMatch(/invalid threshold/i);
+  });
 });
 
 describe("llm-judge (mock)", () => {