@absolutejs/absolute 0.19.0-beta.522 → 0.19.0-beta.524

package/dist/types/ai.d.ts

@@ -959,6 +959,51 @@ export type RAGAnswerGroundingEvaluationCaseDifficultyEntry = {
     rank: number;
     totalEvaluations: number;
 };
+export type RAGAnswerGroundingCaseDifficultyRun = {
+    id: string;
+    suiteId: string;
+    label: string;
+    startedAt: number;
+    finishedAt: number;
+    entries: RAGAnswerGroundingEvaluationCaseDifficultyEntry[];
+    metadata?: Record<string, unknown>;
+};
+export type RAGAnswerGroundingCaseDifficultyHistoryStore = {
+    saveRun: (run: RAGAnswerGroundingCaseDifficultyRun) => Promise<void> | void;
+    listRuns: (input?: {
+        suiteId?: string;
+        limit?: number;
+    }) => Promise<RAGAnswerGroundingCaseDifficultyRun[]> | RAGAnswerGroundingCaseDifficultyRun[];
+};
+export type RAGAnswerGroundingCaseDifficultyDiffEntry = {
+    caseId: string;
+    label?: string;
+    query?: string;
+    previousRank?: number;
+    currentRank: number;
+    previousPassRate?: number;
+    currentPassRate: number;
+    previousFailRate?: number;
+    currentFailRate: number;
+    previousAverageCitationF1?: number;
+    currentAverageCitationF1: number;
+};
+export type RAGAnswerGroundingCaseDifficultyRunDiff = {
+    suiteId: string;
+    currentRunId: string;
+    previousRunId?: string;
+    harderCases: RAGAnswerGroundingCaseDifficultyDiffEntry[];
+    easierCases: RAGAnswerGroundingCaseDifficultyDiffEntry[];
+    unchangedCases: RAGAnswerGroundingCaseDifficultyDiffEntry[];
+};
+export type RAGAnswerGroundingCaseDifficultyHistory = {
+    suiteId: string;
+    suiteLabel?: string;
+    runs: RAGAnswerGroundingCaseDifficultyRun[];
+    latestRun?: RAGAnswerGroundingCaseDifficultyRun;
+    previousRun?: RAGAnswerGroundingCaseDifficultyRun;
+    diff?: RAGAnswerGroundingCaseDifficultyRunDiff;
+};
 export type RAGAnswerGroundingEvaluationCaseDiff = {
     caseId: string;
     label?: string;
@@ -969,10 +1014,22 @@ export type RAGAnswerGroundingEvaluationCaseDiff = {
     currentCoverage: RAGAnswerGroundingEvaluationCaseResult['coverage'];
     previousCitationF1?: number;
     currentCitationF1: number;
+    previousCitedIds: string[];
+    currentCitedIds: string[];
     previousMatchedIds: string[];
     currentMatchedIds: string[];
     previousMissingIds: string[];
     currentMissingIds: string[];
+    previousExtraIds: string[];
+    currentExtraIds: string[];
+    previousReferenceCount?: number;
+    currentReferenceCount: number;
+    previousResolvedCitationCount?: number;
+    currentResolvedCitationCount: number;
+    previousUnresolvedCitationCount?: number;
+    currentUnresolvedCitationCount: number;
+    previousUngroundedReferenceNumbers: number[];
+    currentUngroundedReferenceNumbers: number[];
     previousAnswer?: string;
     currentAnswer: string;
     answerChanged: boolean;
@@ -985,9 +1042,15 @@ export type RAGAnswerGroundingEvaluationCaseSnapshot = {
     coverage: RAGAnswerGroundingEvaluationCaseResult['coverage'];
     citationF1: number;
     resolvedCitationRate: number;
+    citationCount: number;
+    referenceCount: number;
+    resolvedCitationCount: number;
+    unresolvedCitationCount: number;
+    citedIds: string[];
     matchedIds: string[];
     missingIds: string[];
     extraIds: string[];
+    ungroundedReferenceNumbers: number[];
     answer: string;
     previousAnswer?: string;
     answerChange: 'new' | 'changed' | 'unchanged';

package/dist/vue/ai/index.js

@@ -1806,10 +1806,49 @@ var buildRAGAnswerGroundingCaseDifficultyLeaderboard = (entries) => {
     totalEvaluations: entry.totalEvaluations
   }));
 };
+var buildGroundingDifficultyDiffEntry = (current, previous) => ({
+  caseId: current.caseId,
+  currentAverageCitationF1: current.averageCitationF1,
+  currentFailRate: current.failRate,
+  currentPassRate: current.passRate,
+  currentRank: current.rank,
+  label: current.label,
+  previousAverageCitationF1: previous?.averageCitationF1,
+  previousFailRate: previous?.failRate,
+  previousPassRate: previous?.passRate,
+  previousRank: previous?.rank,
+  query: current.query
+});
+var buildRAGAnswerGroundingCaseDifficultyRunDiff = ({
+  current,
+  previous
+}) => {
+  const previousEntries = new Map((previous?.entries ?? []).map((entry) => [entry.caseId, entry]));
+  const diffs = current.entries.map((entry) => buildGroundingDifficultyDiffEntry(entry, previousEntries.get(entry.caseId)));
+  return {
+    currentRunId: current.id,
+    easierCases: diffs.filter((entry) => {
+      const previousRank = entry.previousRank ?? entry.currentRank;
+      return entry.currentRank > previousRank;
+    }),
+    harderCases: diffs.filter((entry) => {
+      const previousRank = entry.previousRank ?? Number.MAX_SAFE_INTEGER;
+      return entry.currentRank < previousRank;
+    }),
+    previousRunId: previous?.id,
+    suiteId: current.suiteId,
+    unchangedCases: diffs.filter((entry) => {
+      const previousRank = entry.previousRank ?? entry.currentRank;
+      return entry.currentRank === previousRank;
+    })
+  };
+};
 var toHistorySortOrder = (left, right) => right.finishedAt - left.finishedAt;
 var normalizeHistoryRuns = (runs) => [...runs].sort(toHistorySortOrder);
 var toGroundingHistorySortOrder = (left, right) => right.finishedAt - left.finishedAt;
 var normalizeGroundingHistoryRuns = (runs) => [...runs].sort(toGroundingHistorySortOrder);
+var toGroundingDifficultyHistorySortOrder = (left, right) => right.finishedAt - left.finishedAt;
+var normalizeGroundingDifficultyHistoryRuns = (runs) => [...runs].sort(toGroundingDifficultyHistorySortOrder);
 var buildCaseDiff = (currentCase, previousCase) => ({
   caseId: currentCase.caseId,
   currentF1: currentCase.f1,
@@ -1827,18 +1866,30 @@ var buildGroundingCaseDiff = (currentCase, previousCase) => ({
   answerChanged: typeof previousCase?.answer === "string" ? previousCase.answer !== currentCase.answer : true,
   caseId: currentCase.caseId,
   currentCitationF1: currentCase.citationF1,
+  currentCitedIds: currentCase.citedIds,
   currentCoverage: currentCase.coverage,
+  currentExtraIds: currentCase.extraIds,
   currentMatchedIds: currentCase.matchedIds,
   currentMissingIds: currentCase.missingIds,
+  currentReferenceCount: currentCase.referenceCount,
+  currentResolvedCitationCount: currentCase.resolvedCitationCount,
   currentAnswer: currentCase.answer,
   currentStatus: currentCase.status,
+  currentUngroundedReferenceNumbers: currentCase.groundedAnswer.ungroundedReferenceNumbers,
+  currentUnresolvedCitationCount: currentCase.unresolvedCitationCount,
   label: currentCase.label,
   previousAnswer: previousCase?.answer,
   previousCitationF1: previousCase?.citationF1,
+  previousCitedIds: previousCase?.citedIds ?? [],
   previousCoverage: previousCase?.coverage,
+  previousExtraIds: previousCase?.extraIds ?? [],
   previousMatchedIds: previousCase?.matchedIds ?? [],
   previousMissingIds: previousCase?.missingIds ?? [],
+  previousReferenceCount: previousCase?.referenceCount,
+  previousResolvedCitationCount: previousCase?.resolvedCitationCount,
   previousStatus: previousCase?.status,
+  previousUngroundedReferenceNumbers: previousCase?.groundedAnswer.ungroundedReferenceNumbers ?? [],
+  previousUnresolvedCitationCount: previousCase?.unresolvedCitationCount,
   query: currentCase.query
 });
 var buildGroundingCaseSnapshots = ({
@@ -1855,7 +1906,9 @@ var buildGroundingCaseSnapshots = ({
       answer: entry.answer,
       answerChange: typeof previousCase?.answer === "string" ? previousCase.answer === entry.answer ? "unchanged" : "changed" : "new",
       caseId: entry.caseId,
+      citationCount: entry.citationCount,
       citationF1: entry.citationF1,
+      citedIds: entry.citedIds,
       coverage: entry.coverage,
       extraIds: entry.extraIds,
       label: entry.label,
@@ -1863,8 +1916,12 @@ var buildGroundingCaseSnapshots = ({
       missingIds: entry.missingIds,
       previousAnswer: previousCase?.answer,
       query: entry.query,
+      referenceCount: entry.referenceCount,
+      resolvedCitationCount: entry.resolvedCitationCount,
       resolvedCitationRate: entry.resolvedCitationRate,
-      status: entry.status
+      status: entry.status,
+      ungroundedReferenceNumbers: entry.groundedAnswer.ungroundedReferenceNumbers,
+      unresolvedCitationCount: entry.unresolvedCitationCount
     };
   });
 };
@@ -1995,6 +2052,42 @@ var createRAGFileAnswerGroundingEvaluationHistoryStore = (path) => ({
     }, null, 2));
   }
 });
+var createRAGFileAnswerGroundingCaseDifficultyHistoryStore = (path) => ({
+  async listRuns(input) {
+    try {
+      const raw = await readFile(path, "utf8");
+      const data = JSON.parse(raw);
+      const runs = Array.isArray(data.runs) ? data.runs : [];
+      const filtered = input?.suiteId ? runs.filter((run) => run.suiteId === input.suiteId) : runs;
+      return normalizeGroundingDifficultyHistoryRuns(filtered).slice(0, input?.limit ?? DEFAULT_HISTORY_LIMIT);
+    } catch (error) {
+      if (error && typeof error === "object" && "code" in error && error.code === "ENOENT") {
+        return [];
+      }
+      throw error;
+    }
+  },
+  async saveRun(run) {
+    let runs = [];
+    try {
+      const raw = await readFile(path, "utf8");
+      const data = JSON.parse(raw);
+      runs = Array.isArray(data.runs) ? data.runs : [];
+    } catch (error) {
+      if (!error || typeof error !== "object" || !("code" in error) || error.code !== "ENOENT") {
+        throw error;
+      }
+    }
+    const nextRuns = normalizeGroundingDifficultyHistoryRuns([
+      run,
+      ...runs.filter((entry) => entry.id !== run.id)
+    ]);
+    await mkdir(dirname(path), { recursive: true });
+    await writeFile(path, JSON.stringify({
+      runs: nextRuns
+    }, null, 2));
+  }
+});
 var loadRAGEvaluationHistory = async ({
   store,
   suite,
@@ -2044,6 +2137,29 @@ var loadRAGAnswerGroundingEvaluationHistory = async ({
     suiteLabel: suite.label ?? suite.id
   };
 };
+var loadRAGAnswerGroundingCaseDifficultyHistory = async ({
+  store,
+  suite,
+  limit = DEFAULT_HISTORY_LIMIT
+}) => {
+  const runs = normalizeGroundingDifficultyHistoryRuns(await Promise.resolve(store.listRuns({
+    limit,
+    suiteId: suite.id
+  })));
+  const latestRun = runs[0];
+  const previousRun = runs[1];
+  return {
+    diff: latestRun && previousRun ? buildRAGAnswerGroundingCaseDifficultyRunDiff({
+      current: latestRun,
+      previous: previousRun
+    }) : undefined,
+    latestRun,
+    previousRun,
+    runs,
+    suiteId: suite.id,
+    suiteLabel: suite.label ?? suite.id
+  };
+};
 var persistRAGEvaluationSuiteRun = async ({
   store,
   run
@@ -2058,6 +2174,13 @@ var persistRAGAnswerGroundingEvaluationRun = async ({
   await Promise.resolve(store.saveRun(run));
   return run;
 };
+var persistRAGAnswerGroundingCaseDifficultyRun = async ({
+  store,
+  run
+}) => {
+  await Promise.resolve(store.saveRun(run));
+  return run;
+};
 var buildRAGEvaluationResponse = (cases) => {
   const totalCases = cases.length;
   const passedCases = cases.filter((entry) => entry.status === "pass").length;
@@ -3074,5 +3197,5 @@ export {
   AIStreamKey
 };
 
-//# debugId=EB5A5D06CF24564764756E2164756E21
+//# debugId=C3E8EFD0B7AD1FB864756E2164756E21
 //# sourceMappingURL=index.js.map