ccqa 1.51.1 → 1.52.0

package/dist/runtime/judge.d.cts

@@ -0,0 +1,68 @@
+//#region src/runtime/judge.d.ts
+/** One decided claim. */
+interface Verdict {
+  ok: boolean;
+  /** Why, in one sentence. Carried into the failure message so a red test says what was wrong. */
+  reason: string;
+}
+/**
+ * The slice of Playwright's `Page` this needs. Structural because ccqa does
+ * not depend on Playwright, which also lets a caller pass anything else that
+ * can hand over text.
+ */
+interface TextSource {
+  innerText(selector: string): Promise<string>;
+}
+/**
+ * The one method this needs from Playwright's `TestInfo` — structural, so
+ * ccqa keeps no Playwright dependency.
+ */
+interface TestInfoLike {
+  attach(name: string, options: {
+    body: string;
+    contentType: string;
+  }): Promise<void>;
+}
+interface JudgeOptions {
+  /** CSS selector whose text is read as the evidence. Default "body". */
+  from?: string;
+  /** Playwright's per-test handle; when given, the verdict is attached to the test's report. */
+  testInfo?: TestInfoLike;
+  /** Model for this judgement. Falls back to CCQA_JUDGE_MODEL, then CCQA_MODEL. */
+  model?: string;
+}
+/**
+ * Fails the test unless a model agrees the claim holds for the text read from
+ * `from` (a selector; omitted, the page's body). The reason the model gave
+ * rides in the failure, so a red test says what was wrong.
+ *
+ * A selector matching several elements judges the first, as Playwright's
+ * page-level `innerText` does — narrow it if that is not what you mean.
+ *
+ * A string third argument is shorthand for `{ from: <string> }` — the form
+ * generated tests already carry. Pass `testInfo` (Playwright's per-test
+ * handle) to attach the verdict to the test's report, pass or fail.
+ */
+declare function judgeByLlm(page: TextSource, claim: string, options?: string | JudgeOptions): Promise<void>;
+interface ClaimInput {
+  /** The claim to decide, as written in the spec's `judgeByLlm`. */
+  claim: string;
+  /** The text it is decided against. */
+  text: string;
+  model?: string;
+  cwd?: string;
+}
+/**
+ * Decides a claim about text a run cannot predict — a generated answer, a
+ * summary. A model that will not answer, or answers something this cannot
+ * read, is an unmade decision rather than a passing one: it throws, so a
+ * claim never goes silently unjudged.
+ */
+declare function decideClaim(input: ClaimInput): Promise<Verdict>;
+/**
+ * The model is asked for bare JSON but sometimes wraps it in prose or a fence.
+ * An answer with no readable verdict is a failure to decide, not a false one.
+ */
+declare function parseVerdict(answer: string): Verdict;
+//#endregion
+export { ClaimInput, JudgeOptions, TestInfoLike, TextSource, Verdict, decideClaim, judgeByLlm, parseVerdict };

package/dist/runtime/judge.d.mts

@@ -13,6 +13,24 @@ interface Verdict {
 interface TextSource {
   innerText(selector: string): Promise<string>;
 }
+/**
+ * The one method this needs from Playwright's `TestInfo` — structural, so
+ * ccqa keeps no Playwright dependency.
+ */
+interface TestInfoLike {
+  attach(name: string, options: {
+    body: string;
+    contentType: string;
+  }): Promise<void>;
+}
+interface JudgeOptions {
+  /** CSS selector whose text is read as the evidence. Default "body". */
+  from?: string;
+  /** Playwright's per-test handle; when given, the verdict is attached to the test's report. */
+  testInfo?: TestInfoLike;
+  /** Model for this judgement. Falls back to CCQA_JUDGE_MODEL, then CCQA_MODEL. */
+  model?: string;
+}
 /**
  * Fails the test unless a model agrees the claim holds for the text read from
  * `from` (a selector; omitted, the page's body). The reason the model gave
@@ -20,8 +38,12 @@ interface TextSource {
  *
  * A selector matching several elements judges the first, as Playwright's
  * page-level `innerText` does — narrow it if that is not what you mean.
+ *
+ * A string third argument is shorthand for `{ from: <string> }` — the form
+ * generated tests already carry. Pass `testInfo` (Playwright's per-test
+ * handle) to attach the verdict to the test's report, pass or fail.
  */
-declare function judgeByLlm(page: TextSource, claim: string, from?: string): Promise<void>;
+declare function judgeByLlm(page: TextSource, claim: string, options?: string | JudgeOptions): Promise<void>;
 interface ClaimInput {
   /** The claim to decide, as written in the spec's `judgeByLlm`. */
   claim: string;
@@ -43,4 +65,4 @@ declare function decideClaim(input: ClaimInput): Promise<Verdict>;
  */
 declare function parseVerdict(answer: string): Verdict;
 //#endregion
-export { ClaimInput, TextSource, Verdict, decideClaim, judgeByLlm, parseVerdict };
+export { ClaimInput, JudgeOptions, TestInfoLike, TextSource, Verdict, decideClaim, judgeByLlm, parseVerdict };