{emptyResultsText(job, resultsError)}
; + } + return ( + <> ++ Router matched or beat your current model +
++ {pct(results.overall_shadow_win_rate_pct + results.overall_tie_rate_pct)} +
+of {(job.judged_count ?? 0).toLocaleString()} judged responses
++ Shadowing {job.shadow_percentage}% via {job.router_name} +
++ {(job.judged_count ?? 0).toLocaleString()} of {job.max_turns.toLocaleString()} turns judged ·{" "} + {(job.error_count ?? 0).toLocaleString()} errored · {usd(job.judge_spend ?? 0)} judge spend + {active && remaining ? ` · ${remaining}` : ""} +
++ Last failure: {job.last_error} +
+ )} ++ Duplicates a sampled slice of the key's traffic through the auto-router and has an LLM judge compare both + answers blind. The router's answers are never served to users; judge calls bill to the shadowed key. +
+Enter a value from 0.1 to 100
+ )} +Enter a value from 1 to 2000
+ )} ++ Would the auto-router have answered as well as the models you use today? Find out on your real traffic, before + switching anything. +
+Existing evaluations could not be loaded. Refresh the page to retry.
+ )} + + {isPending && error == null &&Loading evaluations...
} + + {showcased.map((job) => ( +{children},
+}));
+
+vi.mock("react-syntax-highlighter/dist/esm/styles/prism", () => ({ coy: {} }));
+
+vi.mock("@/contexts/ChatShellContext", () => ({
+ useChatShell: () => {
+ const history = useChatHistory(null, "metrics-test-user");
+ return {
+ accessToken: "sk-test",
+ userId: "metrics-test-user",
+ userEmail: "tester@example.com",
+ userRole: "Admin",
+ premiumUser: false,
+ selectedMCPServers: [],
+ setSelectedMCPServers: vi.fn(),
+ conversations: history.conversations,
+ activeConversation: history.activeConversation,
+ activeConversationId: history.currentActiveId,
+ storageUnavailable: false,
+ staleId: false,
+ createConversation: history.createConversation,
+ appendMessage: history.appendMessage,
+ updateLastAssistantMessage: history.updateLastAssistantMessage,
+ truncateFromMessage: history.truncateFromMessage,
+ deleteConversation: vi.fn(),
+ renameConversation: vi.fn(),
+ };
+ },
+}));
+
+const ONE_TURN_ARG_COUNT = 25;
+const ON_TIMING_DATA_INDEX = 7;
+const ON_USAGE_DATA_INDEX = 8;
+const ON_TOTAL_LATENCY_INDEX = 24;
+
+async function sendOneMessage(): Promise