import { Helmet } from "react-helmet-async";
import { Target, Clock, Database, FileCheck, Microscope } from "lucide-react";
import TrustPage from "../TrustPage";

export default function PerformanceBenchmarks() {
  return (
    <>
      <Helmet>
        <title>Verified Performance Benchmarks - Grep AI</title>
        <meta name="description" content="Transparent, verified performance benchmarks for Grep AI. 94% research accuracy, 15-minute turnaround, 50+ sources per report. Updated quarterly." />
        <meta property="og:title" content="Verified Performance Benchmarks - Grep AI" />
        <meta property="og:description" content="Transparent, verified performance benchmarks for Grep AI. 94% research accuracy, 15-minute turnaround, 50+ sources per report. Updated quarterly." />
        <meta property="og:image" content="https://storage.googleapis.com/parcha-ai-public-assets/grep-images/og/benchmarks.jpg" />
        <meta property="og:url" content="https://grep.ai/trust/performance-benchmarks" />
        <meta property="og:type" content="website" />
        <meta name="twitter:card" content="summary_large_image" />
        <meta name="twitter:title" content="Verified Performance Benchmarks - Grep AI" />
        <meta name="twitter:description" content="Transparent, verified performance benchmarks for Grep AI. 94% research accuracy, 15-minute turnaround, 50+ sources per report. Updated quarterly." />
        <meta name="twitter:image" content="https://storage.googleapis.com/parcha-ai-public-assets/grep-images/og/benchmarks.jpg" />
      </Helmet>
    <TrustPage
      title="Verified Performance Benchmarks"
      subtitle="Performance & Accuracy"
      description="We believe in transparency about our capabilities. These benchmarks are derived from systematic testing across thousands of research tasks, verified against ground-truth datasets, and updated quarterly."
      icon={Target}
      stats={[
        { value: "94%", label: "Research Accuracy", context: "Verified against ground-truth datasets" },
        { value: "15 min", label: "Average Completion", context: "Deep research report turnaround" },
        { value: "50+", label: "Sources per Report", context: "Average databases checked per research" },
        { value: "10,000+", label: "Reports Generated", context: "Across all expert modes" },
      ]}
      methodology={{
        title: "How We Measure Performance",
        description: "Our benchmarking process is designed to be rigorous, reproducible, and transparent.",
        steps: [
          { step: 1, title: "Ground-Truth Dataset", description: "We maintain a curated dataset of entities with known risk profiles, verified by human experts. This serves as the baseline for measuring accuracy." },
          { step: 2, title: "Blind Testing", description: "Grep research is run against the ground-truth dataset without any prior knowledge of expected results. Outputs are compared against known findings." },
          { step: 3, title: "Multi-Dimensional Scoring", description: "We measure accuracy across multiple dimensions: entity identification, risk factor detection, source citation accuracy, and completeness of coverage." },
          { step: 4, title: "Quarterly Review", description: "Benchmarks are re-run quarterly with updated datasets and methodology. Results are published transparently, including any declines." },
          { step: 5, title: "Continuous Improvement", description: "Benchmark results directly inform engineering priorities. When we identify gaps, we address them in subsequent releases." },
        ],
      }}
      dataPoints={[
        {
          category: "Accuracy Metrics",
          icon: Target,
          items: [
            { name: "Entity Identification", detail: "97% accuracy in correctly identifying and disambiguating target entities" },
            { name: "Risk Factor Detection", detail: "94% detection rate for known risk factors (sanctions, litigation, adverse media)" },
            { name: "Citation Accuracy", detail: "99%+ of cited sources link to valid, verifiable primary documents" },
            { name: "False Positive Rate", detail: "Less than 5% false positive rate on sanctions and adverse media screening" },
          ],
        },
        {
          category: "Speed Metrics",
          icon: Clock,
          items: [
            { name: "Ultra Fast Mode", detail: "Average completion in 2-3 minutes with core risk screening" },
            { name: "Standard Mode", detail: "Average completion in 8-12 minutes with comprehensive coverage" },
            { name: "Deep Research Mode", detail: "Average completion in 12-18 minutes with exhaustive analysis" },
            { name: "API Response Time", detail: "Sub-second API acknowledgment with webhook delivery on completion" },
          ],
        },
        {
          category: "Coverage Metrics",
          icon: Database,
          items: [
            { name: "Jurisdictional Coverage", detail: "Corporate registry data available for 200+ countries and territories" },
            { name: "Sanctions List Coverage", detail: "50+ sanctions, watchlist, and PEP databases screened per query" },
            { name: "Media Coverage", detail: "Global news archive spanning 10+ years with multi-language support" },
            { name: "Court Record Coverage", detail: "Federal and major state court records across the United States" },
          ],
        },
        {
          category: "Reliability Metrics",
          icon: FileCheck,
          items: [
            { name: "System Uptime", detail: "99.9% uptime over the trailing 12 months" },
            { name: "Data Freshness", detail: "Sanctions data updated within hours of official publication" },
            { name: "Report Consistency", detail: "Same query produces consistent results — no methodology variance" },
            { name: "Source Availability", detail: "98%+ source availability with automatic failover for maintenance windows" },
          ],
        },
      ]}
      faqs={[
        {
          question: "How do you define 'accuracy'?",
          answer: "We measure accuracy as the percentage of known risk factors correctly identified in our ground-truth test dataset. This includes sanctions matches, litigation findings, adverse media, and corporate registry data. Our 94% accuracy rate reflects the overall detection rate across all risk categories.",
        },
        {
          question: "Why not 100% accuracy?",
          answer: "No research system — human or AI — achieves 100% accuracy. The remaining 6% typically involves edge cases: entities with extremely common names requiring disambiguation, findings only available in non-digital records, or very recent events not yet indexed by our sources. We're transparent about our limitations.",
        },
        {
          question: "How do these benchmarks compare to manual research?",
          answer: "Independent comparisons show that a typical manual research process using 5-10 databases achieves approximately 60-75% coverage of the findings Grep produces from 50+ databases. Human researchers may catch nuances that AI misses, but miss far more through limited source coverage.",
        },
        {
          question: "Are these benchmarks independently verified?",
          answer: "Our benchmarking methodology is documented and reproducible. We welcome third-party verification and have shared our methodology with enterprise customers for independent validation.",
        },
        {
          question: "How often are benchmarks updated?",
          answer: "We re-run our complete benchmark suite quarterly with updated ground-truth datasets. Results are published on this page within two weeks of testing completion. We report all results, including any performance declines.",
        },
      ]}
      cta={{
        title: "Verify Our Claims Yourself",
        description: "Don't take our benchmarks at face value. Run a research report on an entity you already know well and compare Grep's findings against your existing intelligence.",
      }}
    />
    </>
  );
}