Add filtering, sorting, and grouping to linkages page and implement comprehensive model enrichment pipeline

app/analytics/page.tsx

@@ -104,15 +104,15 @@ export default async function AnalyticsPage() {
   return (
     <div className="min-h-screen bg-background">
       <div className="container-content section-padding">
-        <div className="mb-8">
-          <h1 className="text-4xl font-semibold mb-4">Market Analytics</h1>
-          <p className="text-text-muted text-lg">
+        <div className="mb-4">
+          <h1 className="text-3xl font-semibold mb-1">Market Analytics</h1>
+          <p className="text-text-muted text-sm">
             Market structure, concentration, and trends
           </p>
         </div>
 
         {/* Key Stats */}
-        <div className="grid grid-cols-1 md:grid-cols-4 gap-6 mb-12">
+        <div className="grid grid-cols-1 md:grid-cols-4 gap-4 mb-8">
           <div className="stat-card">
             <div className="stat-value">{analytics.totalDeals}</div>
             <div className="stat-label">Total Deals</div>
@@ -132,8 +132,8 @@ export default async function AnalyticsPage() {
         </div>
 
         {/* Modality Breakdown */}
-        <div className="card mb-8">
-          <h2 className="text-2xl font-semibold mb-6">Deals by Modality</h2>
+        <div className="card mb-6">
+          <h2 className="text-xl font-semibold mb-4">Deals by Modality</h2>
           <div className="space-y-4">
             {Object.entries(analytics.modalityCounts)
               .sort(([, a], [, b]) => b - a)
@@ -152,9 +152,9 @@ export default async function AnalyticsPage() {
         </div>
 
         {/* Top Buyers */}
-        <div className="grid md:grid-cols-2 gap-8 mb-8">
+        <div className="grid md:grid-cols-2 gap-4 mb-6">
           <div className="card">
-            <h2 className="text-2xl font-semibold mb-6">Top Buyers by Spend</h2>
+            <h2 className="text-xl font-semibold mb-4">Top Buyers by Spend</h2>
             <div className="space-y-4">
               {analytics.topBuyers.map((buyer, idx) => (
                 <div key={buyer.name} className="flex items-center justify-between py-2">
@@ -173,7 +173,7 @@ export default async function AnalyticsPage() {
 
           {/* Top Providers */}
           <div className="card">
-            <h2 className="text-2xl font-semibold mb-6">Top Providers by Spend</h2>
+            <h2 className="text-xl font-semibold mb-4">Top Providers by Spend</h2>
             <div className="space-y-4">
               {analytics.topProviders.map((provider, idx) => (
                 <div key={provider.name} className="flex items-center justify-between py-2">

app/api/models/enrich-web/route.ts

@@ -0,0 +1,103 @@
+import { NextResponse } from 'next/server'
+import { exec } from 'child_process'
+import { promisify } from 'util'
+import path from 'path'
+import { existsSync } from 'fs'
+
+const execAsync = promisify(exec)
+
+export const dynamic = 'force-dynamic'
+export const maxDuration = 600 // 10 minutes for web enrichment
+
+/**
+ * POST /api/models/enrich-web - Enrich models using web search and LLM extraction
+ * 
+ * Optional query params:
+ * - limit: number of models to enrich (default: all)
+ * - no_web: disable web search (default: false)
+ * - no_llm: disable LLM extraction (default: false)
+ */
+export async function POST(request: Request) {
+  try {
+    const { searchParams } = new URL(request.url)
+    const limit = searchParams.get('limit')
+    const noWeb = searchParams.get('no_web') === 'true'
+    const noLlm = searchParams.get('no_llm') === 'true'
+
+    // Path to the enrichment script
+    const enrichScript = path.join(process.cwd(), 'registry', 'enrich_all_models.py')
+    const venvPython = path.join(process.cwd(), 'venv', 'bin', 'python3')
+    const pythonPath = process.env.PYTHON_PATH || (existsSync(venvPython) ? venvPython : 'python3')
+
+    console.log(`Starting web enrichment: limit=${limit || 'all'}, web=${!noWeb}, llm=${!noLlm}`)
+
+    // Build command
+    let command = `${pythonPath} ${enrichScript}`
+    if (limit) {
+      command += ` --limit ${limit}`
+    }
+    if (noWeb) {
+      command += ` --no-web`
+    }
+    if (noLlm) {
+      command += ` --no-llm`
+    }
+
+    const { stdout, stderr } = await execAsync(command, {
+      cwd: process.cwd(),
+      timeout: 540000, // 9 minutes timeout
+      env: {
+        ...process.env,
+        PYTHONUNBUFFERED: '1',
+      },
+    })
+
+    // Parse output to extract summary
+    const summary: {
+      success: boolean
+      limit?: number
+      no_web: boolean
+      no_llm: boolean
+      output: string
+      errors: string
+      timestamp: string
+      models_enriched?: number
+      errors_count?: number
+    } = {
+      success: true,
+      no_web: noWeb,
+      no_llm: noLlm,
+      output: stdout,
+      errors: stderr,
+      timestamp: new Date().toISOString(),
+    }
+
+    if (limit) {
+      summary.limit = parseInt(limit)
+    }
+
+    // Try to extract numbers from output
+    const enrichedMatch = stdout.match(/Successfully enriched: (\d+)\/(\d+)/)
+    const errorsMatch = stdout.match(/Errors: (\d+)/)
+
+    if (enrichedMatch) {
+      summary.models_enriched = parseInt(enrichedMatch[1])
+    }
+    if (errorsMatch) {
+      summary.errors_count = parseInt(errorsMatch[1])
+    }
+
+    return NextResponse.json(summary)
+  } catch (error: any) {
+    console.error('Web enrichment error:', error)
+    return NextResponse.json(
+      {
+        success: false,
+        error: error.message || 'Failed to enrich models with web search',
+        timestamp: new Date().toISOString(),
+      },
+      { status: 500 }
+    )
+  }
+}
+

app/components/Sidebar.tsx

@@ -10,37 +10,31 @@ export default function Sidebar() {
     {
       href: '/',
       label: 'Deals',
-      icon: '📊',
     },
     {
       href: '/timeline',
       label: 'Timeline',
-      icon: '📅',
     },
     {
       href: '/models',
       label: 'Models',
-      icon: '🤖',
     },
     {
       href: '/linkages',
       label: 'Linkages',
-      icon: '🔗',
     },
     {
       href: '/analytics',
       label: 'Analytics',
-      icon: '📈',
     },
   ]
 
   return (
     <aside className="w-64 bg-surface border-r border-border flex-shrink-0 min-h-screen sticky top-0">
-      <div className="p-6 border-b border-border">
-        <h2 className="text-lg font-semibold mb-1">AI Training Data</h2>
-        <p className="text-xs text-text-muted">Deals Dashboard</p>
+      <div className="p-4 border-b border-border">
+        <h2 className="text-lg font-semibold">AI Training Data</h2>
       </div>
-      <nav className="p-4">
+      <nav className="p-2">
         <ul className="space-y-1">
           {navItems.map((item) => {
             const isActive = pathname === item.href
@@ -48,14 +42,13 @@ export default function Sidebar() {
               <li key={item.href}>
                 <Link
                   href={item.href}
-                  className={`flex items-center gap-3 px-3 py-2 rounded-sm text-sm transition-colors ${
+                  className={`px-3 py-2 rounded-none text-sm transition-colors ${
                     isActive
                       ? 'bg-accent/10 text-accent font-medium'
                       : 'text-text-muted hover:bg-border-subtle hover:text-text'
                   }`}
                 >
-                  <span className="text-base">{item.icon}</span>
-                  <span>{item.label}</span>
+                  {item.label}
                 </Link>
               </li>
             )

app/components/Tooltip.tsx

@@ -33,7 +33,7 @@ export default function Tooltip({
       {children}
       {isVisible && (
         <div
-          className={`absolute z-50 ${positionClasses[position]} w-64 p-3 bg-surface border border-border rounded-sm shadow-lg text-xs text-text leading-relaxed pointer-events-none`}
+          className={`absolute z-50 ${positionClasses[position]} w-64 p-3 bg-surface border border-border rounded-none shadow-lg text-xs text-text leading-relaxed pointer-events-none`}
         >
           {content}
           {/* Arrow */}

app/components/deals/DealFeed.tsx

@@ -104,17 +104,17 @@ function FeedItemCard({ item }: { item: FeedItem }) {
   const isArticle = item.type === 'article' || (!isTwitter && item.url.startsWith('http'))
 
   return (
-    <div className="border border-border-subtle rounded-sm p-4 hover:border-border transition-colors">
+    <div className="border border-border-subtle rounded-none p-4 hover:border-border transition-colors">
       <div className="flex items-start justify-between gap-3">
         <div className="flex-1 min-w-0">
           <div className="flex items-center gap-2 mb-2">
             {isTwitter && (
-              <span className="text-[10px] px-1.5 py-0.5 bg-blue-500/20 text-blue-500 rounded-sm font-mono">
+              <span className="text-[10px] px-1.5 py-0.5 bg-blue-500/20 text-blue-500 rounded-none font-mono">
                 TWITTER
               </span>
             )}
             {isArticle && (
-              <span className="text-[10px] px-1.5 py-0.5 bg-accent/20 text-accent rounded-sm font-mono">
+              <span className="text-[10px] px-1.5 py-0.5 bg-accent/20 text-accent rounded-none font-mono">
                 ARTICLE
               </span>
             )}
@@ -192,7 +192,7 @@ function FeedItemCard({ item }: { item: FeedItem }) {
         <div className="mt-3 pt-3 border-t border-border-subtle">
           <iframe
             src={item.url}
-            className="w-full h-64 border border-border-subtle rounded-sm"
+            className="w-full h-64 border border-border-subtle rounded-none"
             title={item.title || 'Article preview'}
             sandbox="allow-same-origin allow-scripts"
           />

app/components/deals/DiscoveryButton.tsx

@@ -69,7 +69,7 @@ export default function DiscoveryButton() {
       </button>
       
       {status && (
-        <div className="absolute top-full right-0 mt-2 z-50 bg-surface border border-border rounded-sm shadow-lg p-3 min-w-[300px]">
+        <div className="absolute top-full right-0 mt-2 z-50 bg-surface border border-border rounded-none shadow-lg p-3 min-w-[300px]">
           <div className="text-xs font-semibold mb-2 text-text">{status}</div>
           {results && (
             <div className="text-xs text-text-muted space-y-1">

app/components/linkages/CreateLinkagesButton.tsx

@@ -54,7 +54,7 @@ export default function CreateLinkagesButton() {
       </button>
       
       {status && (
-        <div className="absolute top-full right-0 mt-2 z-50 bg-surface border border-border rounded-sm shadow-lg p-3 min-w-[300px]">
+        <div className="absolute top-full right-0 mt-2 z-50 bg-surface border border-border rounded-none shadow-lg p-3 min-w-[300px]">
           <div className="text-xs font-semibold mb-2 text-text">{status}</div>
           {results && (
             <div className="text-xs text-text-muted space-y-1">

app/components/models/IngestModelsButton.tsx

@@ -78,7 +78,7 @@ export default function IngestModelsButton() {
       </button>
       
       {status && (
-        <div className="absolute top-full right-0 mt-2 z-50 bg-surface border border-border rounded-sm shadow-lg p-3 min-w-[300px]">
+        <div className="absolute top-full right-0 mt-2 z-50 bg-surface border border-border rounded-none shadow-lg p-3 min-w-[300px]">
           <div className="text-xs font-semibold mb-2 text-text">{status}</div>
           {results && (
             <div className="text-xs text-text-muted space-y-1">

app/components/models/TokenCalculationCard.tsx

@@ -52,7 +52,7 @@ export default function TokenCalculationCard({
         
         <div className="pt-3 border-t border-border-subtle">
           <strong className="text-text">Step-by-Step Calculation:</strong>
-          <div className="mt-2 space-y-2 font-mono text-xs bg-[rgba(139,111,71,0.05)] p-3 rounded-sm">
+          <div className="mt-2 space-y-2 font-mono text-xs bg-[rgba(139,111,71,0.05)] p-3 rounded-none">
             <div>
               <span className="text-text-muted">Min estimate:</span> {paramsAbs.toLocaleString()} params × {ratioMin} = {tokensEstMin ? (tokensEstMin / 1e9).toFixed(1) : '—'}B tokens
             </div>

app/components/models/TokenEstimateTooltip.tsx

@@ -45,7 +45,7 @@ export default function TokenEstimateTooltip({
       
       {isOpen && (
         <div 
-          className="absolute z-50 left-0 top-full mt-2 w-80 bg-surface border border-border rounded-sm shadow-lg p-4 text-xs"
+          className="absolute z-50 left-0 top-full mt-2 w-80 bg-surface border border-border rounded-none shadow-lg p-4 text-xs"
           onClick={(e) => e.stopPropagation()}
         >
           <div className="font-semibold text-text mb-2">Token Estimate Calculation</div>

app/deals/DealModal.tsx

@@ -51,11 +51,11 @@ export default function DealModal({ deal, isOpen, onClose }: DealModalProps) {
       onClick={onClose}
     >
       <div
-        className="bg-surface rounded-sm shadow-xl max-w-4xl w-full max-h-[90vh] overflow-y-auto"
+        className="bg-surface rounded-none shadow-xl max-w-4xl w-full max-h-[90vh] overflow-y-auto"
         onClick={(e) => e.stopPropagation()}
       >
         {/* Header */}
-        <div className="sticky top-0 bg-surface border-b border-border px-6 py-4 flex items-start justify-between">
+        <div className="sticky top-0 bg-surface border-b border-border px-4 py-3 flex items-start justify-between">
           <div className="flex-1">
             <h2 className="text-2xl font-semibold mb-1">
               {deal.provider} → {deal.buyer}
@@ -249,7 +249,7 @@ export default function DealModal({ deal, isOpen, onClose }: DealModalProps) {
                   <h3 className="text-lg font-semibold mb-4">Pricing Normalizations</h3>
                   <div className="grid grid-cols-2 md:grid-cols-4 gap-4">
                     {normalizations.map((norm, idx) => (
-                      <div key={idx} className="border border-border-subtle rounded-sm p-3">
+                      <div key={idx} className="border border-border-subtle rounded-none p-3">
                         <div className="text-xs text-text-muted mb-1">Per {norm.unitType}</div>
                         <div className="font-medium">
                           {norm.normalizedCostPerUnit < 0.001

app/deals/DealsClient.tsx

@@ -391,7 +391,7 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
             <thead>
               <tr className="bg-border-subtle">
                 <th 
-                  className="cursor-pointer hover:bg-border select-none px-4 py-3"
+                  className="cursor-pointer hover:bg-border select-none"
                   onClick={() => handleSort('provider')}
                   title="Click to sort by provider"
                 >
@@ -508,7 +508,7 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                         onClick={() => toggleGroup(groupKey)}
                         className="cursor-pointer bg-border-subtle hover:bg-border transition-colors"
                       >
-                        <td colSpan={7} className="px-4 py-3">
+                        <td colSpan={7}>
                           <div className="flex items-center justify-between">
                             <div className="flex items-center gap-3">
                               <span className="text-xs text-text-muted">{isExpanded ? '▼' : '▶'}</span>
@@ -539,14 +539,14 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                           onClick={() => handleDealClick(deal)}
                           className="cursor-pointer transition-colors border-b border-border-subtle last:border-0 hover:bg-[rgba(232,225,217,0.3)]"
                         >
-                          <td className="px-4 py-3 pl-8">
+                          <td className="pl-6">
                             <div className="flex items-center gap-2">
                               <div className="font-medium text-accent hover:text-accent-hover">
                                 {deal.provider}
                               </div>
                               {deal.discoveredVia === 'exa' && (
                                 <span 
-                                  className="text-[10px] px-1.5 py-0.5 bg-accent/20 text-accent rounded-sm font-mono"
+                                  className="text-[10px] px-1.5 py-0.5 bg-accent/20 text-accent rounded-none font-mono"
                                   title={`Source: Exa${deal.exaQuery ? ` (${deal.exaQuery})` : ''}${deal.exaScore ? ` - Score: ${deal.exaScore.toFixed(2)}` : ''}`}
                                 >
                                   ARTICLE
@@ -554,7 +554,7 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                               )}
                               {deal.discoveredVia && deal.discoveredVia !== 'exa' && (
                                 <span 
-                                  className="text-[10px] px-1.5 py-0.5 bg-border-subtle text-text-muted rounded-sm"
+                                  className="text-[10px] px-1.5 py-0.5 bg-border-subtle text-text-muted rounded-none"
                                   title={`Discovered via ${deal.discoveredVia}`}
                                 >
                                   {deal.discoveredVia.toUpperCase()}
@@ -562,10 +562,10 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                               )}
                             </div>
                           </td>
-                          <td className="px-4 py-3">
+                          <td>
                             <div className="text-sm">{deal.buyer}</div>
                           </td>
-                          <td className="px-4 py-3">
+                          <td>
                             <span className="badge badge-secondary text-xs">{deal.modality}</span>
                           </td>
                           <td className="px-4 py-3 text-right">
@@ -589,7 +589,7 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                               <span className="text-text-muted/40 text-xs">—</span>
                             )}
                           </td>
-                          <td className="px-4 py-3">
+                          <td>
                             <div className="text-sm text-text-muted/80">{formatDate(deal.date)}</div>
                           </td>
                         </tr>
@@ -605,14 +605,14 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                     onClick={() => handleDealClick(deal)}
                     className="cursor-pointer transition-colors border-b border-border-subtle last:border-0 hover:bg-[rgba(232,225,217,0.3)]"
                   >
-                    <td className="px-4 py-3">
+                    <td>
                       <div className="flex items-center gap-2">
                         <div className="font-medium text-accent hover:text-accent-hover">
                           {deal.provider}
                         </div>
                         {deal.discoveredVia === 'exa' && (
                           <span 
-                            className="text-[10px] px-1.5 py-0.5 bg-accent/20 text-accent rounded-sm font-mono"
+                            className="text-[10px] px-1.5 py-0.5 bg-accent/20 text-accent rounded-none font-mono"
                             title={`Source: Exa${deal.exaQuery ? ` (${deal.exaQuery})` : ''}${deal.exaScore ? ` - Score: ${deal.exaScore.toFixed(2)}` : ''}`}
                           >
                             ARTICLE
@@ -620,7 +620,7 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                         )}
                         {deal.discoveredVia && deal.discoveredVia !== 'exa' && (
                           <span 
-                            className="text-[10px] px-1.5 py-0.5 bg-border-subtle text-text-muted rounded-sm"
+                            className="text-[10px] px-1.5 py-0.5 bg-border-subtle text-text-muted rounded-none"
                             title={`Discovered via ${deal.discoveredVia}`}
                           >
                             {deal.discoveredVia.toUpperCase()}
@@ -628,10 +628,10 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                         )}
                       </div>
                     </td>
-                    <td className="px-4 py-3">
+                    <td>
                       <div className="text-sm">{deal.buyer}</div>
                     </td>
-                    <td className="px-4 py-3">
+                    <td>
                       <span className="badge badge-secondary text-xs">{deal.modality}</span>
                     </td>
                     <td className="px-4 py-3 text-right">
@@ -655,7 +655,7 @@ export default function DealsClient({ initialDeals }: DealsClientProps) {
                         <span className="text-text-muted/40 text-xs">—</span>
                       )}
                     </td>
-                    <td className="px-4 py-3">
+                    <td>
                       <div className="text-sm text-text-muted/80">{formatDate(deal.date)}</div>
                     </td>
                   </tr>
@@ -712,7 +712,7 @@ function PriceCellWithTooltip({ deal }: { deal: Deal }) {
         </div>
       </div>
       {showTooltip && (
-        <div className="absolute right-0 top-full mt-2 z-50 bg-surface border border-border rounded-sm shadow-lg p-3 min-w-[280px]">
+        <div className="absolute right-0 top-full mt-2 z-50 bg-surface border border-border rounded-none shadow-lg p-3 min-w-[280px]">
           <div className="text-xs font-semibold mb-2 text-text">Normalized Pricing</div>
           <div className="space-y-2">
             {normalizations.map((norm, idx) => (

app/deals/[id]/page.tsx

@@ -62,13 +62,13 @@ export default async function DealDetailPage({
 
         <div className="max-w-4xl">
           {/* Header Card */}
-          <div className="card mb-8">
-            <div className="flex items-start justify-between mb-6">
+          <div className="card mb-6">
+            <div className="flex items-start justify-between mb-4">
               <div>
-                <h1 className="text-3xl font-semibold mb-2">
+                <h1 className="text-2xl font-semibold mb-1">
                   {deal.provider} → {deal.buyer}
                 </h1>
-                <p className="text-text-muted">{deal.dataType}</p>
+                <p className="text-text-muted text-sm">{deal.dataType}</p>
               </div>
               <div className="flex gap-2">
                 <span className="badge badge-secondary">{deal.modality}</span>
@@ -78,7 +78,7 @@ export default async function DealDetailPage({
               </div>
             </div>
 
-            <div className="grid grid-cols-2 md:grid-cols-4 gap-4 pt-6 border-t border-border">
+            <div className="grid grid-cols-2 md:grid-cols-4 gap-4 pt-4 border-t border-border">
               <div>
                 <div className="text-sm text-text-muted mb-1">Price</div>
                 <div className="text-xl font-semibold">{formatPrice(deal)}</div>
@@ -115,16 +115,16 @@ export default async function DealDetailPage({
 
           {/* Reported Terms */}
           {deal.reportedTerms && (
-            <div className="card mb-8">
-              <h2 className="text-xl font-semibold mb-4">Reported Terms</h2>
-              <p className="text-lg leading-relaxed">{deal.reportedTerms}</p>
+            <div className="card mb-6">
+              <h2 className="text-lg font-semibold mb-3">Reported Terms</h2>
+              <p className="text-sm leading-relaxed">{deal.reportedTerms}</p>
             </div>
           )}
 
           {/* Deal Details & Compensation */}
-          <div className="grid md:grid-cols-2 gap-6 mb-8">
+          <div className="grid md:grid-cols-2 gap-4 mb-6">
             <div className="card">
-              <h3 className="text-lg font-semibold mb-4">Deal Details</h3>
+              <h3 className="text-base font-semibold mb-3">Deal Details</h3>
               <div className="space-y-2">
                 <div className="flex justify-between">
                   <span className="text-text-muted">Deal Type</span>
@@ -158,7 +158,7 @@ export default async function DealDetailPage({
             </div>
 
             <div className="card">
-              <h3 className="text-lg font-semibold mb-4">Creator Compensation</h3>
+              <h3 className="text-base font-semibold mb-3">Creator Compensation</h3>
               <div className="space-y-2">
                 <div className="flex justify-between">
                   <span className="text-text-muted">Compensated</span>
@@ -190,15 +190,15 @@ export default async function DealDetailPage({
 
           {/* Notes */}
           {deal.notes && (
-            <div className="card mb-8">
-              <h3 className="text-lg font-semibold mb-4">Notes</h3>
-              <p className="leading-relaxed">{deal.notes}</p>
+            <div className="card mb-6">
+              <h3 className="text-base font-semibold mb-3">Notes</h3>
+              <p className="text-sm leading-relaxed">{deal.notes}</p>
             </div>
           )}
 
           {/* Sources - Hyperlinked */}
-          <div className="card mb-8">
-            <h3 className="text-lg font-semibold mb-4">Sources</h3>
+          <div className="card mb-6">
+            <h3 className="text-base font-semibold mb-3">Sources</h3>
             <div className="space-y-2">
               {deal.sourcePrimary && (
                 <div className="text-text-muted">

app/globals.css

@@ -120,21 +120,21 @@
 }
 
 @layer components {
-  /* Container - Max Width with Generous Padding */
+  /* Container - Max Width with Reduced Padding */
   .container-content {
-    @apply px-6;
+    @apply px-2;
     max-width: var(--max-width-content);
     margin: 0 auto;
   }
   
   .container-narrow {
-    @apply mx-auto px-6;
+    @apply mx-auto px-2;
     max-width: var(--max-width-narrow);
   }
   
-  /* Cards - Minimalist, Sharp */
+  /* Cards - Minimalist, Sharp, No Rounded Edges */
   .card {
-    @apply bg-surface border border-border rounded-sm p-6;
+    @apply bg-surface border border-border rounded-none p-4;
     box-shadow: 0 1px 3px 0 rgba(0, 0, 0, 0.05);
   }
   
@@ -147,9 +147,9 @@
     transform: translateY(-1px);
   }
   
-  /* Buttons - Minimalist, Sharp */
+  /* Buttons - Minimalist, Sharp, No Rounded Edges */
   .btn {
-    @apply px-4 py-2 rounded-sm font-medium transition-all duration-200;
+    @apply px-4 py-2 rounded-none font-medium transition-all duration-200;
     @apply focus:outline-none focus:ring-2 focus:ring-offset-2;
   }
   
@@ -177,11 +177,11 @@
   
   .table th {
     @apply text-left font-semibold text-xs text-text-muted uppercase tracking-wide;
-    @apply select-none;
+    @apply select-none px-3 py-2;
   }
   
   .table td {
-    @apply border-b border-border-subtle;
+    @apply border-b border-border-subtle px-3 py-2;
   }
   
   .table tbody tr:hover {
@@ -192,9 +192,9 @@
     @apply border-b-0;
   }
   
-  /* Badges/Tags - Subtle, Sharp */
+  /* Badges/Tags - Subtle, Sharp, No Rounded Edges */
   .badge {
-    @apply inline-flex items-center px-2.5 py-0.5 rounded-sm text-xs font-medium;
+    @apply inline-flex items-center px-2.5 py-0.5 rounded-none text-xs font-medium;
   }
   
   .badge-primary {
@@ -206,9 +206,9 @@
     @apply badge bg-border text-text-muted;
   }
   
-  /* Inputs - Clean, Sharp */
+  /* Inputs - Clean, Sharp, No Rounded Edges */
   .input {
-    @apply w-full px-4 py-2 border border-border rounded-sm;
+    @apply w-full px-4 py-2 border border-border rounded-none;
     @apply bg-surface text-text;
     @apply focus:outline-none focus:ring-2 focus:ring-accent focus:border-transparent;
     @apply transition-all duration-200;
@@ -234,9 +234,9 @@
     text-wrap: balance;
   }
   
-  /* Spacing Utilities */
+  /* Spacing Utilities - Reduced Padding */
   .section-padding {
-    @apply py-12 md:py-16 lg:py-20;
+    @apply py-4 md:py-6;
   }
   
   /* Modal utilities */
@@ -260,7 +260,7 @@
   
   .overflow-x-auto::-webkit-scrollbar-thumb {
     background-color: var(--color-border);
-    border-radius: 3px;
+    border-radius: 0;
   }
   
   .overflow-x-auto::-webkit-scrollbar-thumb:hover {
@@ -271,7 +271,7 @@
   .timeline-container {
     max-width: 1200px;
     margin: 0 auto;
-    padding: 50px;
+    padding: 16px 8px;
   }
 
   .timeline-header {
@@ -377,11 +377,11 @@
 
   /* Progress bar - container and fill using CSS custom properties */
   .progress-bar-container {
-    @apply w-full bg-border-subtle rounded-sm h-1.5 relative overflow-hidden;
+    @apply w-full bg-border-subtle rounded-none h-1.5 relative overflow-hidden;
   }
   
   .progress-bar-fill {
-    @apply bg-accent h-1.5 rounded-sm absolute top-0 left-0 transition-all;
+    @apply bg-accent h-1.5 rounded-none absolute top-0 left-0 transition-all;
     width: calc(var(--progress-percentage, 0) * 1%);
   }
   

app/linkages/LinkagesClient.tsx

@@ -0,0 +1,622 @@
+'use client'
+
+import { useState, useEffect } from 'react'
+import React from 'react'
+import Link from 'next/link'
+import { formatDate } from '@/lib/utils'
+import Tooltip from '@/app/components/Tooltip'
+
+interface Linkage {
+  id: string
+  linkageType: string
+  linkageStrength: string
+  impactInference: string | null
+  deal: {
+    id: string
+    provider: string
+    buyer: string
+    modality: string
+    priceUsd: number | null
+    date: string | null
+  }
+  model: {
+    id: string
+    modelId: string
+    provider: string
+    family: string | null
+    tokensEstMid: number | null
+  }
+}
+
+interface LinkagesClientProps {
+  initialLinkages: Linkage[]
+}
+
+function formatTokens(value: number | null | undefined): string {
+  if (!value) return '—'
+  if (value >= 1e15) return `${(value / 1e15).toFixed(1)}P`
+  if (value >= 1e12) return `${(value / 1e12).toFixed(1)}T`
+  if (value >= 1e9) return `${(value / 1e9).toFixed(1)}B`
+  return `${(value / 1e6).toFixed(0)}M`
+}
+
+export default function LinkagesClient({ initialLinkages }: LinkagesClientProps) {
+  const [linkages] = useState<Linkage[]>(initialLinkages)
+  const [filters, setFilters] = useState({
+    linkageType: '',
+    linkageStrength: '',
+    provider: '',
+    buyer: '',
+    modelProvider: '',
+    modality: '',
+  })
+  const [searchQuery, setSearchQuery] = useState('')
+  const [sortBy, setSortBy] = useState<{ column: string; direction: 'asc' | 'desc' }>({
+    column: 'linkageStrength',
+    direction: 'desc',
+  })
+  const [groupBy, setGroupBy] = useState<string>('')
+  const [expandedGroups, setExpandedGroups] = useState<Set<string>>(new Set())
+
+  // Filter linkages
+  let filteredLinkages = linkages.filter(linkage => {
+    if (!linkage || !linkage.deal || !linkage.model) return false
+    
+    if (filters.linkageType && linkage.linkageType !== filters.linkageType) return false
+    if (filters.linkageStrength && linkage.linkageStrength !== filters.linkageStrength) return false
+    if (filters.provider && !linkage.deal.provider.toLowerCase().includes(filters.provider.toLowerCase())) return false
+    if (filters.buyer && !linkage.deal.buyer.toLowerCase().includes(filters.buyer.toLowerCase())) return false
+    if (filters.modelProvider && !linkage.model.provider.toLowerCase().includes(filters.modelProvider.toLowerCase())) return false
+    if (filters.modality && linkage.deal.modality !== filters.modality) return false
+    
+    if (searchQuery) {
+      const query = searchQuery.toLowerCase()
+      if (
+        !linkage.deal.provider.toLowerCase().includes(query) &&
+        !linkage.deal.buyer.toLowerCase().includes(query) &&
+        !linkage.model.modelId.toLowerCase().includes(query) &&
+        !linkage.model.provider.toLowerCase().includes(query) &&
+        !linkage.deal.modality.toLowerCase().includes(query)
+      ) {
+        return false
+      }
+    }
+    return true
+  })
+
+  // Sort linkages
+  const sortedLinkages = [...filteredLinkages].sort((a, b) => {
+    const { column, direction } = sortBy
+    let comparison = 0
+
+    switch (column) {
+      case 'deal':
+        comparison = a.deal.provider.localeCompare(b.deal.provider)
+        break
+      case 'model':
+        comparison = a.model.modelId.localeCompare(b.model.modelId)
+        break
+      case 'linkageType':
+        comparison = a.linkageType.localeCompare(b.linkageType)
+        break
+      case 'linkageStrength':
+        const strengthOrder = { 'high': 3, 'medium': 2, 'low': 1 }
+        comparison = (strengthOrder[a.linkageStrength as keyof typeof strengthOrder] || 0) - 
+                     (strengthOrder[b.linkageStrength as keyof typeof strengthOrder] || 0)
+        break
+      case 'date':
+        const dateA = a.deal.date || ''
+        const dateB = b.deal.date || ''
+        comparison = dateA.localeCompare(dateB)
+        break
+      default:
+        comparison = 0
+    }
+
+    return direction === 'asc' ? comparison : -comparison
+  })
+
+  // Extract unique values for filters
+  const linkageTypes = Array.from(new Set(linkages.map(l => l.linkageType))).filter(Boolean).sort()
+  const linkageStrengths = Array.from(new Set(linkages.map(l => l.linkageStrength))).filter(Boolean).sort()
+  const providers = Array.from(new Set(linkages.map(l => l.deal.provider))).filter(Boolean).sort()
+  const buyers = Array.from(new Set(linkages.flatMap(l => l.deal.buyer.split(',').map(b => b.trim())))).filter(Boolean).sort()
+  const modelProviders = Array.from(new Set(linkages.map(l => l.model.provider))).filter(Boolean).sort()
+  const modalities = Array.from(new Set(linkages.map(l => l.deal.modality))).filter(Boolean).sort()
+
+  // Group linkages
+  function groupLinkages(linkages: Linkage[], groupByField: string): Record<string, Linkage[]> {
+    if (!groupByField) {
+      return { 'All': linkages }
+    }
+
+    const groups: Record<string, Linkage[]> = {}
+
+    linkages.forEach(linkage => {
+      let groupKey = 'Unknown'
+      
+      switch (groupByField) {
+        case 'linkageType':
+          groupKey = linkage.linkageType === 'temporal_overlap' ? 'Same Time Period' : 
+                     linkage.linkageType === 'inferred' ? 'Same Company' : 
+                     linkage.linkageType || 'Unknown'
+          break
+        case 'linkageStrength':
+          groupKey = linkage.linkageStrength || 'Unknown'
+          break
+        case 'provider':
+          groupKey = linkage.deal.provider || 'Unknown'
+          break
+        case 'buyer':
+          groupKey = linkage.deal.buyer || 'Unknown'
+          break
+        case 'modelProvider':
+          groupKey = linkage.model.provider || 'Unknown'
+          break
+        case 'modality':
+          groupKey = linkage.deal.modality || 'Unknown'
+          break
+        default:
+          groupKey = 'All'
+      }
+
+      if (!groups[groupKey]) {
+        groups[groupKey] = []
+      }
+      groups[groupKey].push(linkage)
+    })
+
+    return groups
+  }
+
+  const groupedLinkages = groupLinkages(sortedLinkages, groupBy)
+  const groupKeys = Object.keys(groupedLinkages).sort((a, b) => {
+    // Sort by strength order if grouping by strength
+    if (groupBy === 'linkageStrength') {
+      const strengthOrder = { 'high': 3, 'medium': 2, 'low': 1 }
+      return (strengthOrder[b.toLowerCase() as keyof typeof strengthOrder] || 0) - 
+             (strengthOrder[a.toLowerCase() as keyof typeof strengthOrder] || 0)
+    }
+    return a.localeCompare(b)
+  })
+
+  const toggleGroup = (groupKey: string) => {
+    setExpandedGroups(prev => {
+      const next = new Set(prev)
+      if (next.has(groupKey)) {
+        next.delete(groupKey)
+      } else {
+        next.add(groupKey)
+      }
+      return next
+    })
+  }
+
+  // Expand all groups by default
+  useEffect(() => {
+    if (groupBy && expandedGroups.size === 0) {
+      setExpandedGroups(new Set(groupKeys))
+    }
+  }, [groupBy, groupKeys])
+
+  const handleSort = (column: string) => {
+    setSortBy(prev => ({
+      column,
+      direction: prev.column === column && prev.direction === 'asc' ? 'desc' : 'asc',
+    }))
+  }
+
+  const getSortIndicator = (column: string) => {
+    if (sortBy.column !== column) return null
+    return sortBy.direction === 'asc' ? '↑' : '↓'
+  }
+
+  return (
+    <>
+      {/* Search and Filters */}
+      <div className="card mb-6 p-4">
+        <div className="mb-4">
+          <input
+            type="text"
+            placeholder="Search linkages..."
+            value={searchQuery}
+            onChange={(e) => setSearchQuery(e.target.value)}
+            className="input w-full text-sm"
+          />
+        </div>
+        
+        <div className="grid grid-cols-2 md:grid-cols-6 gap-3">
+          <div>
+            <label className="block text-xs font-medium text-text-muted mb-1.5 uppercase tracking-wide">Connection Type</label>
+            <select
+              value={filters.linkageType}
+              onChange={(e) => setFilters({ ...filters, linkageType: e.target.value })}
+              className="input text-sm py-1.5"
+            >
+              <option value="">All</option>
+              {linkageTypes.map(type => (
+                <option key={type} value={type}>
+                  {type === 'temporal_overlap' ? 'Same Time Period' : 
+                   type === 'inferred' ? 'Same Company' : type}
+                </option>
+              ))}
+            </select>
+          </div>
+          
+          <div>
+            <label className="block text-xs font-medium text-text-muted mb-1.5 uppercase tracking-wide">Confidence</label>
+            <select
+              value={filters.linkageStrength}
+              onChange={(e) => setFilters({ ...filters, linkageStrength: e.target.value })}
+              className="input text-sm py-1.5"
+            >
+              <option value="">All</option>
+              {linkageStrengths.map(strength => (
+                <option key={strength} value={strength}>{strength.charAt(0).toUpperCase() + strength.slice(1)}</option>
+              ))}
+            </select>
+          </div>
+          
+          <div>
+            <label className="block text-xs font-medium text-text-muted mb-1.5 uppercase tracking-wide">Data Provider</label>
+            <select
+              value={filters.provider}
+              onChange={(e) => setFilters({ ...filters, provider: e.target.value })}
+              className="input text-sm py-1.5"
+            >
+              <option value="">All</option>
+              {providers.map(p => (
+                <option key={p} value={p}>{p}</option>
+              ))}
+            </select>
+          </div>
+          
+          <div>
+            <label className="block text-xs font-medium text-text-muted mb-1.5 uppercase tracking-wide">Buyer</label>
+            <select
+              value={filters.buyer}
+              onChange={(e) => setFilters({ ...filters, buyer: e.target.value })}
+              className="input text-sm py-1.5"
+            >
+              <option value="">All</option>
+              {buyers.map(b => (
+                <option key={b} value={b}>{b}</option>
+              ))}
+            </select>
+          </div>
+          
+          <div>
+            <label className="block text-xs font-medium text-text-muted mb-1.5 uppercase tracking-wide">Model Provider</label>
+            <select
+              value={filters.modelProvider}
+              onChange={(e) => setFilters({ ...filters, modelProvider: e.target.value })}
+              className="input text-sm py-1.5"
+            >
+              <option value="">All</option>
+              {modelProviders.map(p => (
+                <option key={p} value={p}>{p}</option>
+              ))}
+            </select>
+          </div>
+          
+          <div>
+            <label className="block text-xs font-medium text-text-muted mb-1.5 uppercase tracking-wide">Modality</label>
+            <select
+              value={filters.modality}
+              onChange={(e) => setFilters({ ...filters, modality: e.target.value })}
+              className="input text-sm py-1.5"
+            >
+              <option value="">All</option>
+              {modalities.map(m => (
+                <option key={m} value={m}>{m}</option>
+              ))}
+            </select>
+          </div>
+        </div>
+      </div>
+
+      {/* Grouping and Results Count */}
+      <div className="mb-3 flex items-center justify-between flex-wrap gap-3">
+        <div className="flex items-center gap-3">
+          <div className="flex items-center gap-2">
+            <label className="text-xs font-medium text-text-muted uppercase tracking-wide">Group by:</label>
+            <select
+              value={groupBy}
+              onChange={(e) => {
+                setGroupBy(e.target.value)
+                setExpandedGroups(new Set())
+              }}
+              className="input text-sm py-1.5"
+            >
+              <option value="">None</option>
+              <option value="linkageType">Connection Type</option>
+              <option value="linkageStrength">Confidence</option>
+              <option value="provider">Data Provider</option>
+              <option value="buyer">Buyer</option>
+              <option value="modelProvider">Model Provider</option>
+              <option value="modality">Modality</option>
+            </select>
+          </div>
+          {groupBy && (
+            <button
+              onClick={() => {
+                setExpandedGroups(new Set(groupKeys))
+              }}
+              className="text-xs text-accent hover:text-accent-hover"
+            >
+              Expand All
+            </button>
+          )}
+          {groupBy && (
+            <button
+              onClick={() => {
+                setExpandedGroups(new Set())
+              }}
+              className="text-xs text-accent hover:text-accent-hover"
+            >
+              Collapse All
+            </button>
+          )}
+        </div>
+        <div className="text-sm text-text-muted">
+          Showing <span className="font-medium text-text">{sortedLinkages.length}</span> of <span className="font-medium text-text">{linkages.length}</span> linkages
+        </div>
+      </div>
+
+      {/* Results */}
+      <div className="card overflow-hidden p-0">
+        <div className="overflow-x-auto">
+          <table className="table text-sm">
+            <thead>
+              <tr className="bg-border-subtle">
+                <th 
+                  className="cursor-pointer hover:bg-border select-none"
+                  onClick={() => handleSort('deal')}
+                  title="Click to sort by deal"
+                >
+                  <div className="flex items-center gap-2">
+                    <Tooltip content="The training data deal, showing the data provider (who owns the data) and the buyer (the AI company licensing it).">
+                      <span className="underline decoration-dotted cursor-help">Deal</span>
+                    </Tooltip>
+                    {getSortIndicator('deal') && (
+                      <span className="text-text-muted text-xs">{getSortIndicator('deal')}</span>
+                    )}
+                  </div>
+                  <div className="text-xs font-normal text-text-muted mt-0.5">Data provider → Buyer</div>
+                </th>
+                <th 
+                  className="cursor-pointer hover:bg-border select-none"
+                  onClick={() => handleSort('model')}
+                  title="Click to sort by model"
+                >
+                  <div className="flex items-center gap-2">
+                    <Tooltip content="The AI model that may have been trained using data from this deal. Linkages are inferred based on company matches and timing.">
+                      <span className="underline decoration-dotted cursor-help">Model</span>
+                    </Tooltip>
+                    {getSortIndicator('model') && (
+                      <span className="text-text-muted text-xs">{getSortIndicator('model')}</span>
+                    )}
+                  </div>
+                  <div className="text-xs font-normal text-text-muted mt-0.5">AI model that may have used this data</div>
+                </th>
+                <th 
+                  className="cursor-pointer hover:bg-border select-none"
+                  onClick={() => handleSort('linkageType')}
+                  title="Click to sort by connection type"
+                >
+                  <div className="flex items-center gap-2">
+                    <Tooltip content="The type of connection: 'Same Time Period' (deal and model within 1 year), 'Same Company' (buyer matches model provider), or 'Explicit' (directly stated).">
+                      <span className="underline decoration-dotted cursor-help">Connection Type</span>
+                    </Tooltip>
+                    {getSortIndicator('linkageType') && (
+                      <span className="text-text-muted text-xs">{getSortIndicator('linkageType')}</span>
+                    )}
+                  </div>
+                  <div className="text-xs font-normal text-text-muted mt-0.5">How the link was determined</div>
+                </th>
+                <th 
+                  className="cursor-pointer hover:bg-border select-none"
+                  onClick={() => handleSort('linkageStrength')}
+                  title="Click to sort by confidence"
+                >
+                  <div className="flex items-center gap-2">
+                    <Tooltip content="The confidence level in the linkage: High (strong evidence like temporal overlap), Medium (moderate evidence), or Low (weak evidence).">
+                      <span className="underline decoration-dotted cursor-help">Confidence</span>
+                    </Tooltip>
+                    {getSortIndicator('linkageStrength') && (
+                      <span className="text-text-muted text-xs">{getSortIndicator('linkageStrength')}</span>
+                    )}
+                  </div>
+                  <div className="text-xs font-normal text-text-muted mt-0.5">How certain we are</div>
+                </th>
+                <th className="text-left font-semibold">
+                  <Tooltip content="An interpretation of what this linkage means - how the deal's data may have impacted the model's training.">
+                    <span className="underline decoration-dotted cursor-help">What This Means</span>
+                  </Tooltip>
+                  <div className="text-xs font-normal text-text-muted mt-0.5">Interpretation of the connection</div>
+                </th>
+              </tr>
+            </thead>
+            <tbody>
+              {sortedLinkages.length === 0 ? (
+                <tr>
+                  <td colSpan={5} className="text-center py-12 text-text-muted">
+                    No linkages found matching your filters
+                  </td>
+                </tr>
+              ) : groupBy ? (
+                // Grouped view
+                groupKeys.map((groupKey) => {
+                  const groupLinkages = groupedLinkages[groupKey]
+                  const isExpanded = expandedGroups.has(groupKey)
+                  
+                  return (
+                    <React.Fragment key={groupKey}>
+                      <tr
+                        onClick={() => toggleGroup(groupKey)}
+                        className="cursor-pointer bg-border-subtle hover:bg-border transition-colors"
+                      >
+                        <td colSpan={5}>
+                          <div className="flex items-center justify-between">
+                            <div className="flex items-center gap-3">
+                              <span className="text-xs text-text-muted">{isExpanded ? '▼' : '▶'}</span>
+                              <span className="font-semibold text-sm">{groupKey}</span>
+                              <span className="text-xs text-text-muted">
+                                ({groupLinkages.length} {groupLinkages.length === 1 ? 'linkage' : 'linkages'})
+                              </span>
+                            </div>
+                          </div>
+                        </td>
+                      </tr>
+                      {isExpanded && groupLinkages.map((linkage) => (
+                        <tr
+                          key={linkage.id}
+                          className="transition-colors border-b border-border-subtle last:border-0 hover:bg-[rgba(232,225,217,0.3)]"
+                        >
+                          <td className="pl-6">
+                            <Link
+                              href={`/deals/${linkage.deal.id}`}
+                              className="font-medium text-accent hover:text-accent-hover"
+                            >
+                              {linkage.deal.provider} → {linkage.deal.buyer}
+                            </Link>
+                            <div className="text-xs text-text-muted mt-0.5">
+                              {linkage.deal.modality} • {linkage.deal.date ? formatDate(linkage.deal.date) : '—'}
+                            </div>
+                          </td>
+                          <td>
+                            <Link
+                              href={`/models/${linkage.model.id}`}
+                              className="font-medium text-accent hover:text-accent-hover"
+                            >
+                              {linkage.model.modelId}
+                            </Link>
+                            {linkage.model.family && (
+                              <div className="text-xs text-text-muted mt-0.5">
+                                {linkage.model.family} • {linkage.model.provider}
+                              </div>
+                            )}
+                            {linkage.model.tokensEstMid && (
+                              <div className="text-xs text-text-muted mt-0.5">
+                                {formatTokens(linkage.model.tokensEstMid)} tokens
+                              </div>
+                            )}
+                          </td>
+                          <td>
+                            <div className="flex flex-col gap-1">
+                              <span className="badge badge-secondary text-xs">
+                                {linkage.linkageType === 'temporal_overlap' ? 'Same Time Period' : 
+                                 linkage.linkageType === 'inferred' ? 'Same Company' : 
+                                 linkage.linkageType || '—'}
+                              </span>
+                              <div className="text-xs text-text-muted/70">
+                                {linkage.linkageType === 'temporal_overlap' 
+                                  ? 'Deal & model within 1 year'
+                                  : linkage.linkageType === 'inferred'
+                                  ? 'Buyer matches model provider'
+                                  : ''}
+                              </div>
+                            </div>
+                          </td>
+                          <td>
+                            <span className={`badge ${
+                              linkage.linkageStrength === 'high' 
+                                ? 'badge-primary' 
+                                : 'badge-secondary'
+                            } text-xs`}>
+                              {linkage.linkageStrength === 'high' ? 'High' : 
+                               linkage.linkageStrength === 'medium' ? 'Medium' : 
+                               linkage.linkageStrength === 'low' ? 'Low' : 
+                               '—'}
+                            </span>
+                          </td>
+                          <td>
+                            <div className="text-sm text-text leading-relaxed">
+                              {linkage.impactInference || '—'}
+                            </div>
+                          </td>
+                        </tr>
+                      ))}
+                    </React.Fragment>
+                  )
+                })
+              ) : (
+                // Ungrouped view
+                sortedLinkages.map((linkage) => (
+                  <tr
+                    key={linkage.id}
+                    className="transition-colors border-b border-border-subtle last:border-0 hover:bg-[rgba(232,225,217,0.3)]"
+                  >
+                    <td>
+                      <Link
+                        href={`/deals/${linkage.deal.id}`}
+                        className="font-medium text-accent hover:text-accent-hover"
+                      >
+                        {linkage.deal.provider} → {linkage.deal.buyer}
+                      </Link>
+                      <div className="text-xs text-text-muted mt-0.5">
+                        {linkage.deal.modality} • {linkage.deal.date ? formatDate(linkage.deal.date) : '—'}
+                      </div>
+                    </td>
+                    <td>
+                      <Link
+                        href={`/models/${linkage.model.id}`}
+                        className="font-medium text-accent hover:text-accent-hover"
+                      >
+                        {linkage.model.modelId}
+                      </Link>
+                      {linkage.model.family && (
+                        <div className="text-xs text-text-muted mt-0.5">
+                          {linkage.model.family} • {linkage.model.provider}
+                        </div>
+                      )}
+                      {linkage.model.tokensEstMid && (
+                        <div className="text-xs text-text-muted mt-0.5">
+                          {formatTokens(linkage.model.tokensEstMid)} tokens
+                        </div>
+                      )}
+                    </td>
+                    <td>
+                      <div className="flex flex-col gap-1">
+                        <span className="badge badge-secondary text-xs">
+                          {linkage.linkageType === 'temporal_overlap' ? 'Same Time Period' : 
+                           linkage.linkageType === 'inferred' ? 'Same Company' : 
+                           linkage.linkageType || '—'}
+                        </span>
+                        <div className="text-xs text-text-muted/70">
+                          {linkage.linkageType === 'temporal_overlap' 
+                            ? 'Deal & model within 1 year'
+                            : linkage.linkageType === 'inferred'
+                            ? 'Buyer matches model provider'
+                            : ''}
+                        </div>
+                      </div>
+                    </td>
+                    <td>
+                      <span className={`badge ${
+                        linkage.linkageStrength === 'high' 
+                          ? 'badge-primary' 
+                          : 'badge-secondary'
+                      } text-xs`}>
+                        {linkage.linkageStrength === 'high' ? 'High' : 
+                         linkage.linkageStrength === 'medium' ? 'Medium' : 
+                         linkage.linkageStrength === 'low' ? 'Low' : 
+                         '—'}
+                      </span>
+                    </td>
+                    <td>
+                      <div className="text-sm text-text leading-relaxed">
+                        {linkage.impactInference || '—'}
+                      </div>
+                    </td>
+                  </tr>
+                ))
+              )}
+            </tbody>
+          </table>
+        </div>
+      </div>
+    </>
+  )
+}
+

app/linkages/page.tsx

@@ -1,69 +1,45 @@
 import { prisma } from '@/lib/prisma'
-import Link from 'next/link'
-import { formatDate } from '@/lib/utils'
 import AutoCreate from '@/app/components/linkages/AutoCreate'
-import Tooltip from '@/app/components/Tooltip'
+import LinkagesClient from './LinkagesClient'
 
 async function getLinkages() {
-  try {
-    // Fetch linkages with relations
-    const linkages = await prisma.dealModelLinkage.findMany({
-      include: {
-        deal: {
-          select: {
-            id: true,
-            provider: true,
-            buyer: true,
-            modality: true,
-            priceUsd: true,
-            date: true,
-          },
+  const linkages = await prisma.dealModelLinkage.findMany({
+    include: {
+      deal: {
+        select: {
+          id: true,
+          provider: true,
+          buyer: true,
+          modality: true,
+          priceUsd: true,
+          date: true,
         },
-        model: {
-          select: {
-            id: true,
-            modelId: true,
-            provider: true,
-            family: true,
-            tokensEstMid: true,
-          },
+      },
+      model: {
+        select: {
+          id: true,
+          modelId: true,
+          provider: true,
+          family: true,
+          tokensEstMid: true,
         },
       },
-      orderBy: [
-        { linkageStrength: 'desc' },
-        { analysisTimestamp: 'desc' },
-      ],
-    })
-    return linkages
-  } catch (error: any) {
-    console.error('Error fetching linkages:', error)
-    // Return empty array on error to prevent page crash
-    return []
-  }
+    },
+    orderBy: [
+      { linkageStrength: 'desc' },
+      { analysisTimestamp: 'desc' },
+    ],
+  })
+  return linkages
 }
 
-function formatTokens(value: number | null | undefined): string {
-  if (!value) return '—'
-  if (value >= 1e15) return `${(value / 1e15).toFixed(1)}P`
-  if (value >= 1e12) return `${(value / 1e12).toFixed(1)}T`
-  if (value >= 1e9) return `${(value / 1e9).toFixed(1)}B`
-  return `${(value / 1e6).toFixed(0)}M`
-}
 
 async function getDealCount() {
-  try {
-    return await prisma.deal.count()
-  } catch {
-    return 0
-  }
+  return await prisma.deal.count()
 }
 
 async function getModelCount() {
-  try {
-    return await prisma.modelRegistry.count()
-  } catch {
-    return 0
-  }
+  return await prisma.modelRegistry.count()
 }
 
 export default async function LinkagesPage() {
@@ -82,196 +58,68 @@ export default async function LinkagesPage() {
         />
         
         {/* Header */}
-        <div className="mb-6">
-          <div className="mb-6">
-            <h1 className="text-4xl font-semibold mb-2">Deal-Model Linkages</h1>
-            <p className="text-text-muted text-lg mb-4">
-              Connections between training data deals and AI models
-            </p>
-            
-            {/* Simple Explanation */}
-            <div className="card bg-[rgba(139,111,71,0.05)] border border-accent/20">
-              <div className="p-4">
-                <h3 className="font-semibold text-text mb-2">What are linkages?</h3>
-                <p className="text-sm text-text-muted leading-relaxed">
-                  Linkages connect training data deals to AI models. For example: if OpenAI signed a deal with News Corp in 2023, 
-                  and GPT-4 was released in 2023, there's a linkage suggesting the News Corp data may have been used to train GPT-4.
-                </p>
-                <p className="text-sm text-text-muted leading-relaxed mt-2">
-                  The system automatically creates linkages when: (1) the deal buyer matches the model provider (e.g., OpenAI deal → OpenAI model), 
-                  and (2) optionally, when the deal date and model release date are close in time (within 1 year).
-                </p>
-              </div>
-            </div>
-          </div>
+        <div className="mb-4">
+          <h1 className="text-3xl font-semibold mb-1">Deal-Model Linkages</h1>
+          <p className="text-text-muted text-sm">
+            Connections between training data deals and AI models
+          </p>
         </div>
 
-
-        {/* Linkages Table */}
-        <div className="card overflow-hidden p-0">
-          <div className="overflow-x-auto">
-            <table className="table text-sm">
-                      <thead>
-                        <tr className="border-b border-border-subtle">
-                          <th className="px-4 py-3 text-left font-semibold">
-                            <Tooltip content="The training data deal, showing the data provider (who owns the data) and the buyer (the AI company licensing it).">
-                              <div className="underline decoration-dotted cursor-help">Deal</div>
-                            </Tooltip>
-                            <div className="text-xs font-normal text-text-muted mt-0.5">Data provider → Buyer</div>
-                          </th>
-                          <th className="px-4 py-3 text-left font-semibold">
-                            <Tooltip content="The AI model that may have been trained using data from this deal. Linkages are inferred based on company matches and timing.">
-                              <div className="underline decoration-dotted cursor-help">Model</div>
-                            </Tooltip>
-                            <div className="text-xs font-normal text-text-muted mt-0.5">AI model that may have used this data</div>
-                          </th>
-                          <th className="px-4 py-3 text-left font-semibold">
-                            <Tooltip content="The type of connection: 'Same Time Period' (deal and model within 1 year), 'Same Company' (buyer matches model provider), or 'Explicit' (directly stated).">
-                              <div className="underline decoration-dotted cursor-help">Connection Type</div>
-                            </Tooltip>
-                            <div className="text-xs font-normal text-text-muted mt-0.5">How the link was determined</div>
-                          </th>
-                          <th className="px-4 py-3 text-left font-semibold">
-                            <Tooltip content="The confidence level in the linkage: High (strong evidence like temporal overlap), Medium (moderate evidence), or Low (weak evidence).">
-                              <div className="underline decoration-dotted cursor-help">Confidence</div>
-                            </Tooltip>
-                            <div className="text-xs font-normal text-text-muted mt-0.5">How certain we are</div>
-                          </th>
-                          <th className="px-4 py-3 text-left font-semibold">
-                            <Tooltip content="An interpretation of what this linkage means - how the deal's data may have impacted the model's training.">
-                              <div className="underline decoration-dotted cursor-help">What This Means</div>
-                            </Tooltip>
-                            <div className="text-xs font-normal text-text-muted mt-0.5">Interpretation of the connection</div>
-                          </th>
-                        </tr>
-                      </thead>
-              <tbody>
-                {linkages.length === 0 ? (
-                  <tr>
-                    <td colSpan={5} className="text-center py-12 text-text-muted">
-                      {dealCount === 0 || modelCount === 0 
-                        ? `No ${dealCount === 0 ? 'deals' : 'models'} found. Please seed the database first.`
-                        : 'No linkages found. Linkage creation will start automatically.'}
-                    </td>
-                  </tr>
-                ) : (
-                  linkages
-                    .filter(linkage => linkage && linkage.deal && linkage.model)
-                    .map((linkage) => (
-                      <tr
-                        key={linkage.id}
-                        className="transition-colors border-b border-border-subtle last:border-0 hover:bg-[rgba(232,225,217,0.3)]"
-                      >
-                        <td className="px-4 py-3">
-                          <Link
-                            href={`/deals/${linkage.deal.id}`}
-                            className="font-medium text-accent hover:text-accent-hover"
-                          >
-                            {linkage.deal.provider} → {linkage.deal.buyer}
-                          </Link>
-                          <div className="text-xs text-text-muted mt-0.5">
-                            {linkage.deal.modality} • {linkage.deal.date ? formatDate(linkage.deal.date) : '—'}
-                          </div>
-                        </td>
-                        <td className="px-4 py-3">
-                          <Link
-                            href={`/models/${linkage.model.id}`}
-                            className="font-medium text-accent hover:text-accent-hover"
-                          >
-                            {linkage.model.modelId}
-                          </Link>
-                          {linkage.model.family && (
-                            <div className="text-xs text-text-muted mt-0.5">
-                              {linkage.model.family} • {linkage.model.provider}
-                            </div>
-                          )}
-                          {linkage.model.tokensEstMid && (
-                            <div className="text-xs text-text-muted mt-0.5">
-                              {formatTokens(linkage.model.tokensEstMid)} tokens
-                            </div>
-                          )}
-                        </td>
-                        <td className="px-4 py-3">
-                          <div className="flex flex-col gap-1">
-                            <span className="badge badge-secondary text-xs">
-                              {linkage.linkageType === 'temporal_overlap' ? 'Same Time Period' : 
-                               linkage.linkageType === 'inferred' ? 'Same Company' : 
-                               linkage.linkageType || '—'}
-                            </span>
-                            <div className="text-xs text-text-muted/70">
-                              {linkage.linkageType === 'temporal_overlap' 
-                                ? 'Deal & model within 1 year'
-                                : linkage.linkageType === 'inferred'
-                                ? 'Buyer matches model provider'
-                                : ''}
-                            </div>
-                          </div>
-                        </td>
-                        <td className="px-4 py-3">
-                          <span className={`badge ${
-                            linkage.linkageStrength === 'high' 
-                              ? 'badge-primary' 
-                              : 'badge-secondary'
-                          } text-xs`}>
-                            {linkage.linkageStrength === 'high' ? 'High' : 
-                             linkage.linkageStrength === 'medium' ? 'Medium' : 
-                             linkage.linkageStrength === 'low' ? 'Low' : 
-                             '—'}
-                          </span>
-                        </td>
-                        <td className="px-4 py-3">
-                          <div className="text-sm text-text leading-relaxed">
-                            {linkage.impactInference || '—'}
-                          </div>
-                        </td>
-                      </tr>
-                    ))
-                )}
-              </tbody>
-            </table>
-          </div>
-        </div>
-
-        {/* Detailed Explanation */}
+        {/* Detailed Explanation - Above table, collapsible */}
         {linkages.length > 0 && (
-          <div className="card mt-8">
-            <h2 className="text-xl font-semibold mb-4">How Linkages Work</h2>
-            <div className="space-y-4 text-sm text-text-muted">
-              <div>
-                <h3 className="font-semibold text-text mb-2">Example:</h3>
-                <p className="leading-relaxed">
-                  If you see a linkage: <strong className="text-text">News Corp → OpenAI</strong> connected to <strong className="text-text">GPT-4</strong>, 
-                  it means OpenAI signed a deal with News Corp, and because GPT-4 is an OpenAI model, there's a potential connection. 
-                  If the deal happened in 2023 and GPT-4 was released in 2023, that's a stronger connection (temporal overlap).
-                </p>
-              </div>
-              
-              <div className="pt-3 border-t border-border-subtle">
-                <h3 className="font-semibold text-text mb-2">Connection Types:</h3>
-                <ul className="space-y-2">
-                  <li>
-                    <strong className="text-text">Same Time Period:</strong> Deal and model release are within 1 year. 
-                    Suggests the deal's data may have been used in training.
-                  </li>
-                  <li>
-                    <strong className="text-text">Same Company:</strong> Deal buyer matches model provider, but different time periods. 
-                    Shows organizational relationship but less direct connection.
-                  </li>
-                </ul>
-              </div>
-              
-              <div className="pt-3 border-t border-border-subtle">
-                <h3 className="font-semibold text-text mb-2">Confidence Levels:</h3>
-                <p className="leading-relaxed">
-                  Currently all linkages are marked as <strong className="text-text">High</strong> confidence because they require 
-                  a clear match between the deal buyer and model provider. The system automatically creates these connections 
-                  when it finds matching company names (e.g., "OpenAI" in both the deal and model).
-                </p>
+          <div className="card mb-6">
+            <details className="group">
+              <summary className="cursor-pointer list-none flex items-center justify-between">
+                <h2 className="text-lg font-semibold">How Linkages Work</h2>
+                <span className="text-text-muted text-sm group-open:hidden">Click to expand</span>
+                <span className="text-text-muted text-sm hidden group-open:inline">Click to collapse</span>
+              </summary>
+              <div className="mt-3 pt-3 border-t border-border-subtle space-y-3 text-sm text-text-muted">
+                <div>
+                  <h3 className="font-semibold text-text mb-2">Example:</h3>
+                  <p className="leading-relaxed">
+                    If you see a linkage: <strong className="text-text">News Corp → OpenAI</strong> connected to <strong className="text-text">GPT-4</strong>, 
+                    it means OpenAI signed a deal with News Corp, and because GPT-4 is an OpenAI model, there's a potential connection. 
+                    If the deal happened in 2023 and GPT-4 was released in 2023, that's a stronger connection (temporal overlap).
+                  </p>
+                </div>
+                
+                <div className="pt-2 border-t border-border-subtle">
+                  <h3 className="font-semibold text-text mb-2">Connection Types:</h3>
+                  <ul className="space-y-2">
+                    <li>
+                      <strong className="text-text">Same Time Period:</strong> Deal and model release are within 1 year. 
+                      Suggests the deal's data may have been used in training.
+                    </li>
+                    <li>
+                      <strong className="text-text">Same Company:</strong> Deal buyer matches model provider, but different time periods. 
+                      Shows organizational relationship but less direct connection.
+                    </li>
+                  </ul>
+                </div>
+                
+                <div className="pt-2 border-t border-border-subtle">
+                  <h3 className="font-semibold text-text mb-2">Confidence Levels:</h3>
+                  <p className="leading-relaxed">
+                    Currently all linkages are marked as <strong className="text-text">High</strong> confidence because they require 
+                    a clear match between the deal buyer and model provider. The system automatically creates these connections 
+                    when it finds matching company names (e.g., "OpenAI" in both the deal and model).
+                  </p>
+                </div>
               </div>
+            </details>
+          </div>
+        )}
+
+        {/* Linkages Table with Filtering, Sorting, and Grouping */}
+        {linkages.length === 0 ? (
+          <div className="card">
+            <div className="text-center py-12 text-text-muted">
+              {dealCount === 0 || modelCount === 0 
+                ? `No ${dealCount === 0 ? 'deals' : 'models'} found. Please seed the database first.`
+                : 'No linkages found. Linkage creation will start automatically.'}
             </div>
           </div>
+        ) : (
+          <LinkagesClient initialLinkages={linkages.filter(l => l && l.deal && l.model)} />
         )}
-      </div>
-    </main>
-  )
-}

app/models/[id]/page.tsx

@@ -91,15 +91,15 @@ export default async function ModelDetailPage({
 
         <div className="max-w-4xl">
           {/* Header Card */}
-          <div className="card mb-8">
-            <div className="mb-4">
-              <h1 className="text-3xl font-semibold mb-2">{model.modelId}</h1>
+          <div className="card mb-6">
+            <div className="mb-3">
+              <h1 className="text-2xl font-semibold mb-1">{model.modelId}</h1>
               {model.family && (
-                <p className="text-text-muted text-lg">{model.family}</p>
+                <p className="text-text-muted text-sm">{model.family}</p>
               )}
             </div>
 
-            <div className="grid grid-cols-2 md:grid-cols-4 gap-4 pt-6 border-t border-border">
+            <div className="grid grid-cols-2 md:grid-cols-4 gap-4 pt-4 border-t border-border">
               <div>
                 <div className="text-sm text-text-muted mb-1">Provider</div>
                 <div className="font-medium">{model.provider}</div>
@@ -137,9 +137,9 @@ export default async function ModelDetailPage({
 
           {/* Token Estimates */}
           {(model.tokensEstMin || model.tokensEstMax || model.tokensEstMid) && (
-            <div className="mb-8">
-              <div className="card mb-6">
-                <h2 className="text-xl font-semibold mb-4">Training Token Estimates</h2>
+            <div className="mb-6">
+              <div className="card mb-4">
+                <h2 className="text-lg font-semibold mb-3">Training Token Estimates</h2>
                 <div className="grid grid-cols-1 md:grid-cols-3 gap-4">
                   <div>
                     <div className="text-sm text-text-muted mb-1">Minimum</div>
@@ -180,8 +180,8 @@ export default async function ModelDetailPage({
 
           {/* Evidence Profile */}
           {(evidenceTypes.length > 0 || model.evidenceStrength || uncertaintySources.length > 0) && (
-            <div className="card mb-8">
-              <h2 className="text-xl font-semibold mb-4">Evidence Profile</h2>
+            <div className="card mb-6">
+              <h2 className="text-lg font-semibold mb-3">Evidence Profile</h2>
               <div className="space-y-4">
                 {model.evidenceStrength && (
                   <div>
@@ -221,8 +221,8 @@ export default async function ModelDetailPage({
 
           {/* Training Compute */}
           {(model.flopsReported || model.flopsEstimated) && (
-            <div className="card mb-8">
-              <h2 className="text-xl font-semibold mb-4">Training Compute</h2>
+            <div className="card mb-6">
+              <h2 className="text-lg font-semibold mb-3">Training Compute</h2>
               <div className="grid grid-cols-1 md:grid-cols-2 gap-4">
                 {model.flopsReported && (
                   <div>
@@ -242,13 +242,13 @@ export default async function ModelDetailPage({
 
           {/* Linked Deals */}
           {model.modelLinkages && model.modelLinkages.length > 0 && (
-            <div className="card mb-8">
-              <h2 className="text-xl font-semibold mb-4">Linked Training Data Deals</h2>
+            <div className="card mb-6">
+              <h2 className="text-lg font-semibold mb-3">Linked Training Data Deals</h2>
               <div className="space-y-3">
                 {model.modelLinkages.map((linkage) => (
                   <div
                     key={linkage.id}
-                    className="p-4 border border-border-subtle rounded-sm hover:bg-border-subtle/30 transition-colors"
+                    className="p-4 border border-border-subtle rounded-none hover:bg-border-subtle/30 transition-colors"
                   >
                     <div className="flex items-start justify-between mb-2">
                       <div>
@@ -285,7 +285,7 @@ export default async function ModelDetailPage({
           {/* Sources */}
           {sources.length > 0 && (
             <div className="card">
-              <h2 className="text-xl font-semibold mb-4">Sources</h2>
+              <h2 className="text-lg font-semibold mb-3">Sources</h2>
               <div className="space-y-2">
                 {sources.map((source, idx) => (
                   <div key={idx}>

app/models/page.tsx

@@ -59,14 +59,12 @@ export default async function ModelsPage() {
         />
         
         {/* Header */}
-        <div className="mb-6">
-                  <div className="mb-4">
-                    <h1 className="text-4xl font-semibold mb-2">Model Registry</h1>
-                    <p className="text-text-muted text-lg">
-                      Training data scale estimates for major AI models
-                    </p>
-                  </div>
-                </div>
+        <div className="mb-4">
+          <h1 className="text-3xl font-semibold mb-1">Model Registry</h1>
+          <p className="text-text-muted text-sm">
+            Training data scale estimates for major AI models
+          </p>
+        </div>
 
         {/* Stats */}
         <div className="grid grid-cols-2 md:grid-cols-3 gap-3 mb-8">
@@ -94,9 +92,9 @@ export default async function ModelsPage() {
             <table className="table text-sm">
               <thead>
                 <tr className="border-b border-border-subtle">
-                  <th className="px-4 py-3 text-left font-semibold">Model</th>
-                  <th className="px-4 py-3 text-left font-semibold">Provider</th>
-                  <th className="px-4 py-3 text-left font-semibold">
+                  <th className="text-left font-semibold">Model</th>
+                  <th className="text-left font-semibold">Provider</th>
+                  <th className="text-left font-semibold">
                     <Tooltip content="The number of trainable parameters in the model, typically measured in billions (B) or trillions (T). More parameters generally mean more capacity to learn complex patterns.">
                       <span className="underline decoration-dotted cursor-help">Params</span>
                     </Tooltip>
@@ -132,7 +130,7 @@ export default async function ModelsPage() {
                       key={model.id}
                       className="cursor-pointer transition-colors border-b border-border-subtle last:border-0 hover:bg-[rgba(232,225,217,0.3)]"
                     >
-                      <td className="px-4 py-3">
+                      <td>
                         <Link
                           href={`/models/${model.id}`}
                           className="font-medium text-accent hover:text-accent-hover"
@@ -143,10 +141,10 @@ export default async function ModelsPage() {
                           <div className="text-xs text-text-muted mt-0.5">{model.family}</div>
                         )}
                       </td>
-                      <td className="px-4 py-3">
+                      <td>
                         <div className="text-sm">{model.provider}</div>
                       </td>
-                      <td className="px-4 py-3">
+                      <td>
                         <div className="text-sm">{formatParams(model.params)}</div>
                         {model.isMoe && (
                           <Tooltip content="Mixture of Experts (MoE): A model architecture that uses multiple specialized sub-networks (experts) but only activates a subset for each input. This allows for larger models with lower computational costs.">
@@ -174,7 +172,7 @@ export default async function ModelsPage() {
                           <span className="text-text-muted/60">—</span>
                         )}
                       </td>
-                      <td className="px-4 py-3">
+                      <td>
                         <div className="flex items-center gap-2">
                           {model.architectureType && (
                             <span className="badge badge-secondary text-xs">
@@ -188,14 +186,14 @@ export default async function ModelsPage() {
                           )}
                         </div>
                       </td>
-                      <td className="px-4 py-3">
+                      <td>
                         {model.evidenceStrength && (
                           <span className="badge badge-secondary text-xs">
                             {model.evidenceStrength.replace('S-', '')}
                           </span>
                         )}
                       </td>
-                      <td className="px-4 py-3">
+                      <td>
                         <div className="text-sm text-text-muted/80">
                           {model.releaseDate ? formatDate(model.releaseDate instanceof Date ? model.releaseDate.toISOString() : String(model.releaseDate)) : '—'}
                         </div>

app/normalization/page.tsx

@@ -141,16 +141,16 @@ export default async function NormalizationPage() {
   return (
     <main className="min-h-screen bg-background">
       <div className="container-content section-padding">
-        <div className="mb-8">
-          <h1 className="text-4xl font-semibold mb-4">Pricing Normalization Tool</h1>
-          <p className="text-text-muted text-lg">
+        <div className="mb-4">
+          <h1 className="text-3xl font-semibold mb-1">Pricing Normalization Tool</h1>
+          <p className="text-text-muted text-sm">
             Compare deals on an apples-to-apples basis by normalizing to per-unit pricing
           </p>
         </div>
 
-        <div className="card mb-8">
-          <h2 className="text-2xl font-semibold mb-4">How It Works</h2>
-          <p className="text-text-muted leading-relaxed mb-4">
+        <div className="card mb-6">
+          <h2 className="text-xl font-semibold mb-3">How It Works</h2>
+          <p className="text-text-muted text-sm leading-relaxed mb-3">
             Different deals use different pricing models (per-book, per-track, aggregate licensing, etc.).
             This tool normalizes prices to common units (tokens, records, images, minutes) to enable
             direct comparison.

app/page.tsx

@@ -176,11 +176,11 @@ export default async function Home() {
         {/* Auto-enrich notification */}
         <AutoEnrich dealCount={deals.length} dealsWithAllFields={dealsWithAllFields} />
         {/* Header */}
-        <div className="mb-6">
-          <div className="flex items-center justify-between mb-4">
+        <div className="mb-4">
+          <div className="flex items-center justify-between">
             <div>
-              <h1 className="text-4xl font-semibold mb-2">Deals Explorer</h1>
-              <p className="text-text-muted text-lg">
+              <h1 className="text-3xl font-semibold mb-1">Deals Explorer</h1>
+              <p className="text-text-muted text-sm">
                 Global licensing, acquisition, and commissioning deals (2020–2025)
               </p>
             </div>
@@ -224,7 +224,7 @@ export default async function Home() {
           </div>
 
           {/* Scrollable Analytics Cards */}
-          <div className="overflow-x-auto pb-2 -mx-6 px-6">
+          <div className="overflow-x-auto pb-2 -mx-2 px-2">
             <div className="flex gap-4 min-w-max">
               {/* Modality Breakdown */}
               <div className="card min-w-[280px] flex-shrink-0">

app/timeline/page.tsx

@@ -208,8 +208,8 @@ export default async function TimelinePage() {
     <div className="min-h-screen bg-background">
       <div className="container-content section-padding">
         {/* Header */}
-        <div className="mb-9">
-          <h1 className="text-4xl font-semibold mb-2 tracking-tight">Major AI Training Data Deals (2020–2025)</h1>
+        <div className="mb-4">
+          <h1 className="text-3xl font-semibold mb-1 tracking-tight">Major AI Training Data Deals (2020–2025)</h1>
           <p className="text-text-muted text-sm mb-1">Tracking the emergence of data markets</p>
           <p className="text-text-muted text-xs font-medium">Source: Open Data Labs (opendatalabs.xyz)</p>
         </div>

registry/enrich_all_models.py

@@ -0,0 +1,207 @@
+"""
+Batch enrichment script for all existing models
+Enriches all models in the database with comprehensive metadata
+"""
+
+import asyncio
+import os
+import sys
+from pathlib import Path
+from typing import List, Optional
+from datetime import datetime
+import json
+
+# Add project root to path
+project_root = Path(__file__).parent.parent
+sys.path.insert(0, str(project_root))
+
+from registry.enrichment.comprehensive_enrichment import ComprehensiveModelEnricher
+from dotenv import load_dotenv
+
+# Prisma imports
+try:
+    from prisma import Prisma
+    PRISMA_AVAILABLE = True
+except ImportError:
+    PRISMA_AVAILABLE = False
+    print("Warning: Prisma not available. Install with: npm run db:generate")
+
+load_dotenv()
+
+
+async def enrich_all_models(
+    limit: Optional[int] = None,
+    use_web_search: bool = True,
+    use_llm_extraction: bool = True
+):
+    """
+    Enrich all models in the database
+    
+    Args:
+        limit: Limit number of models to enrich (for testing)
+        use_web_search: Enable web search enrichment
+        use_llm_extraction: Enable LLM extraction
+    """
+    if not PRISMA_AVAILABLE:
+        raise RuntimeError("Prisma not available. Run: npm run db:generate")
+    
+    # Connect to database
+    prisma = Prisma()
+    await prisma.connect()
+    
+    try:
+        # Get all models
+        models = await prisma.modelregistry.find_many(
+            take=limit
+        )
+        
+        print(f"🚀 Starting enrichment of {len(models)} models")
+        print(f"   Started at: {datetime.now().isoformat()}\n")
+        
+        # Initialize enricher
+        enricher = ComprehensiveModelEnricher(
+            use_web_search=use_web_search,
+            use_llm_extraction=use_llm_extraction
+        )
+        
+        success_count = 0
+        error_count = 0
+        
+        for i, model in enumerate(models, 1):
+            print(f"\n[{i}/{len(models)}] Enriching: {model.modelId} ({model.provider})")
+            
+            try:
+                # Get existing data
+                existing_data = {
+                    "params": model.params,
+                    "releaseDate": model.releaseDate.isoformat() if model.releaseDate else None,
+                    "architectureType": model.architectureType,
+                    "isMoe": model.isMoe,
+                    "multimodal": model.multimodal,
+                    "tokensEstMid": model.tokensEstMid,
+                }
+                
+                # Enrich model
+                enriched = await enricher.enrich_model(
+                    model_id=model.modelId,
+                    provider=model.provider,
+                    family=model.family,
+                    existing_data=existing_data
+                )
+                
+                # Prepare update data
+                update_data = {}
+                
+                # Release date
+                if enriched.get("releaseDate"):
+                    if isinstance(enriched["releaseDate"], str):
+                        update_data["releaseDate"] = datetime.fromisoformat(enriched["releaseDate"])
+                    else:
+                        update_data["releaseDate"] = enriched["releaseDate"]
+                
+                # Architecture
+                if enriched.get("architectureType"):
+                    update_data["architectureType"] = enriched["architectureType"]
+                if enriched.get("isMoe") is not None:
+                    update_data["isMoe"] = enriched["isMoe"]
+                if enriched.get("numExperts"):
+                    update_data["numExperts"] = enriched["numExperts"]
+                if enriched.get("multimodal") is not None:
+                    update_data["multimodal"] = enriched["multimodal"]
+                
+                # Parameters (only if not already set)
+                if not model.params and enriched.get("params"):
+                    update_data["params"] = enriched["params"]
+                
+                # Token estimates
+                if enriched.get("tokensEstMin"):
+                    update_data["tokensEstMin"] = enriched["tokensEstMin"]
+                if enriched.get("tokensEstMax"):
+                    update_data["tokensEstMax"] = enriched["tokensEstMax"]
+                if enriched.get("tokensEstMid"):
+                    update_data["tokensEstMid"] = enriched["tokensEstMid"]
+                if enriched.get("tokensRangeGeneratedAt"):
+                    update_data["tokensRangeGeneratedAt"] = enriched["tokensRangeGeneratedAt"]
+                
+                # Evidence profile
+                if enriched.get("evidenceTypes"):
+                    update_data["evidenceTypes"] = enriched["evidenceTypes"]
+                if enriched.get("evidenceStrength"):
+                    update_data["evidenceStrength"] = enriched["evidenceStrength"]
+                if enriched.get("uncertaintySources"):
+                    update_data["uncertaintySources"] = enriched["uncertaintySources"]
+                if enriched.get("evidenceProfileGeneratedAt"):
+                    update_data["evidenceProfileGeneratedAt"] = enriched["evidenceProfileGeneratedAt"]
+                
+                # Sources
+                if enriched.get("sources"):
+                    update_data["sources"] = enriched["sources"]
+                
+                # Raw evidence snippets
+                if enriched.get("rawEvidenceSnippets"):
+                    update_data["rawEvidenceSnippets"] = enriched["rawEvidenceSnippets"]
+                
+                # Composition estimates
+                if enriched.get("compositionEstimates"):
+                    update_data["compositionEstimates"] = enriched["compositionEstimates"]
+                
+                # Training period
+                if enriched.get("trainingPeriodStart"):
+                    if isinstance(enriched["trainingPeriodStart"], str):
+                        update_data["trainingPeriodStart"] = datetime.fromisoformat(enriched["trainingPeriodStart"])
+                    else:
+                        update_data["trainingPeriodStart"] = enriched["trainingPeriodStart"]
+                if enriched.get("trainingPeriodEnd"):
+                    if isinstance(enriched["trainingPeriodEnd"], str):
+                        update_data["trainingPeriodEnd"] = datetime.fromisoformat(enriched["trainingPeriodEnd"])
+                    else:
+                        update_data["trainingPeriodEnd"] = enriched["trainingPeriodEnd"]
+                
+                # Update model
+                if update_data:
+                    update_data["updatedAt"] = datetime.now()
+                    await prisma.modelregistry.update(
+                        where={"id": model.id},
+                        data=update_data
+                    )
+                    print(f"  ✅ Updated {len(update_data)} fields")
+                    success_count += 1
+                else:
+                    print(f"  ⚠️  No new data to update")
+                    success_count += 1
+                    
+            except Exception as e:
+                print(f"  ❌ Error: {e}")
+                error_count += 1
+                continue
+        
+        print(f"\n✅ Enrichment complete!")
+        print(f"   Successfully enriched: {success_count}/{len(models)}")
+        print(f"   Errors: {error_count}")
+        print(f"   Completed at: {datetime.now().isoformat()}")
+        
+    finally:
+        await prisma.disconnect()
+
+
+async def main():
+    """Main entry point"""
+    import argparse
+    
+    parser = argparse.ArgumentParser(description="Enrich all models in database")
+    parser.add_argument("--limit", type=int, help="Limit number of models to enrich")
+    parser.add_argument("--no-web", action="store_true", help="Disable web search enrichment")
+    parser.add_argument("--no-llm", action="store_true", help="Disable LLM extraction")
+    
+    args = parser.parse_args()
+    
+    await enrich_all_models(
+        limit=args.limit,
+        use_web_search=not args.no_web,
+        use_llm_extraction=not args.no_llm
+    )
+
+
+if __name__ == "__main__":
+    asyncio.run(main())
+

registry/enrichment/__init__.py

@@ -0,0 +1,4 @@
+"""
+Model enrichment module for web-based data collection
+"""
+

registry/enrichment/comprehensive_enrichment.py

@@ -0,0 +1,370 @@
+"""
+Comprehensive Model Enrichment Pipeline
+Orchestrates all enrichment sources and merges data
+"""
+
+import sys
+import asyncio
+from pathlib import Path
+from typing import Dict, List, Optional, Any
+from datetime import datetime
+import json
+
+# Add project root to path
+project_root = Path(__file__).parent.parent.parent
+sys.path.insert(0, str(project_root))
+
+from registry.collectors.epoch_collector import EpochCollector
+from registry.collectors.hf_collector import HuggingFaceCollector
+from registry.enrichment.web_enrichment import WebModelEnricher
+from registry.evidence_profile import EvidenceProfileManager
+from registry.inference.reconciliation import TokenInferenceReconciler
+
+
+class ComprehensiveModelEnricher:
+    """Orchestrates all enrichment sources for comprehensive model metadata"""
+    
+    def __init__(
+        self,
+        use_web_search: bool = True,
+        use_llm_extraction: bool = True,
+        exa_api_key: Optional[str] = None,
+        llm_provider: str = "openai",
+        llm_api_key: Optional[str] = None
+    ):
+        """
+        Initialize comprehensive enricher
+        
+        Args:
+            use_web_search: Enable web search enrichment
+            use_llm_extraction: Enable LLM extraction
+            exa_api_key: Exa API key
+            llm_provider: LLM provider ("openai" or "anthropic")
+            llm_api_key: LLM API key
+        """
+        self.epoch_collector = EpochCollector()
+        self.hf_collector = HuggingFaceCollector()
+        self.inference_reconciler = TokenInferenceReconciler()
+        self.evidence_manager = EvidenceProfileManager()
+        
+        if use_web_search:
+            try:
+                self.web_enricher = WebModelEnricher(
+                    exa_api_key=exa_api_key,
+                    llm_provider=llm_provider,
+                    llm_api_key=llm_api_key
+                )
+            except Exception as e:
+                print(f"Warning: Web enricher initialization failed: {e}")
+                self.web_enricher = None
+        else:
+            self.web_enricher = None
+    
+    async def enrich_model(
+        self,
+        model_id: str,
+        provider: str,
+        family: Optional[str] = None,
+        existing_data: Optional[Dict[str, Any]] = None
+    ) -> Dict[str, Any]:
+        """
+        Enrich model with data from all sources
+        
+        Args:
+            model_id: Model identifier
+            provider: Model provider
+            family: Model family
+            existing_data: Existing model data
+        
+        Returns:
+            Comprehensive enriched metadata
+        """
+        # Start with existing data or empty dict
+        enriched = existing_data.copy() if existing_data else {}
+        enriched["modelId"] = model_id
+        enriched["provider"] = provider
+        enriched["family"] = family or provider
+        
+        # Source 1: Epoch AI
+        epoch_data = None
+        try:
+            # Epoch collector may be async or sync, handle both
+            if hasattr(self.epoch_collector, 'fetch_notable_models'):
+                epoch_models = self.epoch_collector.fetch_notable_models()
+                if asyncio.iscoroutine(epoch_models):
+                    epoch_models = await epoch_models
+                for model in epoch_models:
+                    if self._matches_model(model.get("model_name", ""), model_id, provider):
+                        epoch_data = model
+                        break
+        except Exception as e:
+            print(f"Epoch collection error: {e}")
+        
+        # Source 2: HuggingFace
+        hf_data = None
+        try:
+            # HF collector may be async or sync, handle both
+            if hasattr(self.hf_collector, 'search_models'):
+                hf_result = self.hf_collector.search_models(model_id)
+                if asyncio.iscoroutine(hf_result):
+                    hf_result = await hf_result
+                if hf_result:
+                    hf_data = hf_result[0] if isinstance(hf_result, list) else hf_result
+        except Exception as e:
+            print(f"HF collection error: {e}")
+        
+        # Source 3: Web search (if enabled)
+        web_data = None
+        if self.web_enricher:
+            try:
+                web_data = self.web_enricher.enrich_model(
+                    model_id=model_id,
+                    provider=provider,
+                    existing_data=enriched
+                )
+            except Exception as e:
+                print(f"Web enrichment error: {e}")
+        
+        # Merge all sources with priority
+        merged = self._merge_sources(
+            enriched,
+            epoch_data,
+            hf_data,
+            web_data
+        )
+        
+        # Run token inference if we have params
+        if merged.get("params"):
+            try:
+                inference_input = {
+                    "params": merged.get("params"),
+                    "flops": merged.get("flopsReported"),
+                    "architecture": {
+                        "is_moe": merged.get("isMoe", False),
+                        "num_experts": merged.get("numExperts"),
+                    },
+                    "provider": provider,
+                    "model_id": model_id,
+                }
+                inference_result = self.inference_reconciler.reconcile(inference_input)
+                
+                # Add token estimates
+                merged["tokensEstMin"] = inference_result.get("min")
+                merged["tokensEstMax"] = inference_result.get("max")
+                merged["tokensEstMid"] = inference_result.get("mid")
+                merged["tokensRangeGeneratedAt"] = datetime.now()
+            except Exception as e:
+                print(f"Token inference error: {e}")
+        
+        # Generate evidence profile
+        evidence_profile = self._generate_evidence_profile(
+            epoch_data,
+            hf_data,
+            web_data,
+            merged
+        )
+        
+        # Add evidence profile fields
+        merged["evidenceTypes"] = json.dumps(evidence_profile.get("evidence_types", []))
+        merged["evidenceStrength"] = evidence_profile.get("strength")
+        merged["uncertaintySources"] = json.dumps(evidence_profile.get("uncertainty", []))
+        merged["evidenceProfileGeneratedAt"] = datetime.now()
+        
+        # Combine sources
+        sources = []
+        if epoch_data and epoch_data.get("source_url"):
+            sources.append({
+                "type": "epoch",
+                "url": epoch_data.get("source_url"),
+                "retrieved_at": datetime.now().isoformat(),
+            })
+        if hf_data and hf_data.get("url"):
+            sources.append({
+                "type": "huggingface",
+                "url": hf_data.get("url"),
+                "retrieved_at": datetime.now().isoformat(),
+            })
+        if web_data:
+            sources.extend(web_data.get("sources", []))
+        
+        if sources:
+            merged["sources"] = json.dumps(sources)
+        
+        # Add raw evidence snippets
+        raw_snippets = []
+        if web_data:
+            raw_snippets.extend(web_data.get("raw_evidence_snippets", []))
+        
+        if raw_snippets:
+            merged["rawEvidenceSnippets"] = json.dumps(raw_snippets)
+        
+        # Add composition estimates if training data info found
+        if web_data and web_data.get("training_data_composition"):
+            composition = {
+                "description": web_data.get("training_data_composition"),
+                "sources": web_data.get("training_data_sources", []),
+            }
+            merged["compositionEstimates"] = json.dumps(composition)
+        
+        return merged
+    
+    def _matches_model(self, name: str, model_id: str, provider: str) -> bool:
+        """Check if model name matches"""
+        name_lower = name.lower()
+        model_lower = model_id.lower()
+        provider_lower = provider.lower()
+        
+        # Exact match
+        if model_lower in name_lower or name_lower in model_lower:
+            return True
+        
+        # Provider match + partial model match
+        if provider_lower in name_lower:
+            # Check for common model patterns
+            if any(pattern in model_lower for pattern in ["gpt", "claude", "gemini", "llama", "mistral"]):
+                if any(pattern in name_lower for pattern in ["gpt", "claude", "gemini", "llama", "mistral"]):
+                    return True
+        
+        return False
+    
+    def _merge_sources(
+        self,
+        base: Dict[str, Any],
+        epoch_data: Optional[Dict[str, Any]],
+        hf_data: Optional[Dict[str, Any]],
+        web_data: Optional[Dict[str, Any]]
+    ) -> Dict[str, Any]:
+        """Merge data from all sources with priority"""
+        merged = base.copy()
+        
+        # Priority: Direct disclosures > Third-party analysis > Inferred
+        
+        # From Epoch (high priority - curated dataset)
+        if epoch_data:
+            if not merged.get("params") and epoch_data.get("parameter_count"):
+                merged["params"] = epoch_data.get("parameter_count") / 1e9
+            if not merged.get("releaseDate") and epoch_data.get("release_date"):
+                merged["releaseDate"] = self._parse_date(epoch_data.get("release_date"))
+            if not merged.get("architectureType") and epoch_data.get("architecture_type"):
+                merged["architectureType"] = epoch_data.get("architecture_type")
+            if epoch_data.get("architecture_type", "").lower() == "moe":
+                merged["isMoe"] = True
+            if not merged.get("multimodal"):
+                modality = epoch_data.get("modality", "").lower()
+                merged["multimodal"] = modality in ["multimodal", "vision", "image"]
+            if not merged.get("flopsReported") and epoch_data.get("compute_PF_days"):
+                merged["flopsReported"] = epoch_data.get("compute_PF_days")
+        
+        # From HuggingFace (medium priority)
+        if hf_data:
+            if not merged.get("params") and hf_data.get("params"):
+                merged["params"] = hf_data.get("params") / 1e9
+            if not merged.get("releaseDate") and hf_data.get("created_at"):
+                merged["releaseDate"] = self._parse_date(hf_data.get("created_at"))
+        
+        # From Web search (lower priority but comprehensive)
+        if web_data:
+            if not merged.get("releaseDate") and web_data.get("release_date"):
+                merged["releaseDate"] = self._parse_date(web_data.get("release_date"))
+            if not merged.get("architectureType") and web_data.get("architecture_type"):
+                merged["architectureType"] = web_data.get("architecture_type")
+            if web_data.get("is_moe") is not None:
+                merged["isMoe"] = web_data.get("is_moe")
+            if web_data.get("num_experts"):
+                merged["numExperts"] = web_data.get("num_experts")
+            if web_data.get("multimodal") is not None:
+                merged["multimodal"] = web_data.get("multimodal")
+            if web_data.get("training_period_start"):
+                merged["trainingPeriodStart"] = self._parse_date(web_data.get("training_period_start"))
+            if web_data.get("training_period_end"):
+                merged["trainingPeriodEnd"] = self._parse_date(web_data.get("training_period_end"))
+        
+        return merged
+    
+    def _generate_evidence_profile(
+        self,
+        epoch_data: Optional[Dict[str, Any]],
+        hf_data: Optional[Dict[str, Any]],
+        web_data: Optional[Dict[str, Any]],
+        merged: Dict[str, Any]
+    ) -> Dict[str, Any]:
+        """Generate evidence profile from all sources"""
+        evidence_types = set()
+        sources_count = 0
+        
+        # Epoch data - usually E4 (third-party analysis)
+        if epoch_data:
+            sources_count += 1
+            evidence_types.add("E4")
+            if epoch_data.get("parameter_count"):
+                evidence_types.add("E3")  # Architecture evidence
+        
+        # HuggingFace - E4 (third-party)
+        if hf_data:
+            sources_count += 1
+            evidence_types.add("E4")
+        
+        # Web data - can be E1-E5 depending on source
+        if web_data:
+            sources_count += len(web_data.get("sources", []))
+            web_evidence = web_data.get("evidence_types", [])
+            evidence_types.update(web_evidence)
+        
+        # Direct disclosure if we have official sources
+        if web_data:
+            for source in web_data.get("sources", []):
+                url = source.get("url", "").lower()
+                if any(domain in url for domain in ["openai.com", "anthropic.com", "google.com", "meta.com"]):
+                    evidence_types.add("E1")  # Direct disclosure
+        
+        # Compute evidence
+        if merged.get("flopsReported"):
+            evidence_types.add("E2")
+        
+        # Architecture evidence
+        if merged.get("params") or merged.get("architectureType"):
+            evidence_types.add("E3")
+        
+        # Calculate strength
+        if sources_count >= 3 and "E1" in evidence_types:
+            strength = "S-High"
+        elif sources_count >= 2 or "E1" in evidence_types:
+            strength = "S-Medium"
+        else:
+            strength = "S-Low"
+        
+        # Identify uncertainties
+        uncertainty = []
+        if not merged.get("releaseDate"):
+            uncertainty.append("U5")
+        if not merged.get("architectureType"):
+            uncertainty.append("U3")
+        if not merged.get("trainingPeriodStart"):
+            uncertainty.append("U2")
+        
+        return {
+            "evidence_types": list(evidence_types),
+            "strength": strength,
+            "uncertainty": uncertainty,
+        }
+    
+    def _parse_date(self, date_value: Any) -> Optional[datetime]:
+        """Parse date from various formats"""
+        if not date_value:
+            return None
+        
+        if isinstance(date_value, datetime):
+            return date_value
+        
+        if isinstance(date_value, str):
+            try:
+                return datetime.fromisoformat(date_value.replace("Z", "+00:00"))
+            except (ValueError, AttributeError):
+                if len(date_value) == 4 and date_value.isdigit():
+                    return datetime(int(date_value), 1, 1)
+        
+        if isinstance(date_value, int):
+            return datetime(date_value, 1, 1)
+        
+        return None
+

registry/enrichment/llm_extractor.py

@@ -0,0 +1,227 @@
+"""
+LLM Extraction Service
+Uses OpenAI or Anthropic API to extract structured data from web content
+"""
+
+import os
+import json
+from typing import Dict, List, Optional, Any
+from datetime import datetime
+import re
+
+
+class LLMExtractor:
+    """Extracts structured model metadata from web content using LLM"""
+    
+    def __init__(self, provider: str = "openai", api_key: Optional[str] = None):
+        """
+        Initialize LLM extractor
+        
+        Args:
+            provider: "openai" or "anthropic"
+            api_key: API key (if None, reads from env)
+        """
+        self.provider = provider.lower()
+        self.api_key = api_key or os.getenv(
+            "OPENAI_API_KEY" if self.provider == "openai" else "ANTHROPIC_API_KEY"
+        )
+        
+        if not self.api_key:
+            raise ValueError(
+                f"{provider.upper()}_API_KEY environment variable required"
+            )
+        
+        if self.provider == "openai":
+            try:
+                import openai
+                self.client = openai.OpenAI(api_key=self.api_key)
+            except ImportError:
+                raise ImportError("openai package required. Install with: pip install openai")
+        elif self.provider == "anthropic":
+            try:
+                import anthropic
+                self.client = anthropic.Anthropic(api_key=self.api_key)
+            except ImportError:
+                raise ImportError("anthropic package required. Install with: pip install anthropic")
+        else:
+            raise ValueError(f"Unsupported provider: {provider}")
+    
+    def extract_model_metadata(
+        self,
+        model_id: str,
+        provider: str,
+        web_content: str,
+        context: Optional[Dict[str, Any]] = None
+    ) -> Dict[str, Any]:
+        """
+        Extract structured metadata from web content
+        
+        Args:
+            model_id: Model identifier (e.g., "GPT-4")
+            provider: Model provider (e.g., "OpenAI")
+            web_content: Text content from web sources
+            context: Additional context (existing model data)
+        
+        Returns:
+            Dict with extracted metadata
+        """
+        # Build prompt
+        prompt = self._build_extraction_prompt(model_id, provider, web_content, context)
+        
+        try:
+            if self.provider == "openai":
+                response = self.client.chat.completions.create(
+                    model="gpt-4o-mini",  # Use cheaper model for extraction
+                    messages=[
+                        {"role": "system", "content": "You are a data extraction assistant. Extract structured information about AI models from text. Always return valid JSON."},
+                        {"role": "user", "content": prompt}
+                    ],
+                    response_format={"type": "json_object"},
+                    temperature=0.1,  # Low temperature for consistent extraction
+                )
+                result_text = response.choices[0].message.content
+            else:  # anthropic
+                response = self.client.messages.create(
+                    model="claude-3-5-sonnet-20241022",
+                    max_tokens=2000,
+                    system="You are a data extraction assistant. Extract structured information about AI models from text. Always return valid JSON.",
+                    messages=[
+                        {"role": "user", "content": prompt}
+                    ],
+                )
+                result_text = response.content[0].text
+            
+            # Parse JSON response
+            extracted = json.loads(result_text)
+            
+            # Validate and normalize
+            return self._validate_extraction(extracted, model_id, provider)
+            
+        except json.JSONDecodeError as e:
+            print(f"JSON decode error: {e}")
+            return self._empty_extraction()
+        except Exception as e:
+            print(f"LLM extraction error: {e}")
+            return self._empty_extraction()
+    
+    def _build_extraction_prompt(
+        self,
+        model_id: str,
+        provider: str,
+        web_content: str,
+        context: Optional[Dict[str, Any]] = None
+    ) -> str:
+        """Build extraction prompt"""
+        context_str = ""
+        if context:
+            context_str = f"\n\nExisting known information:\n{json.dumps(context, indent=2)}"
+        
+        prompt = f"""Extract structured information about the AI model "{model_id}" by {provider} from the following text.
+
+Text content:
+{web_content[:8000]}  # Limit content size
+{context_str}
+
+Extract the following information and return as JSON:
+{{
+  "release_date": "YYYY-MM-DD or null if not found",
+  "architecture_type": "Transformer, MoE, or null",
+  "is_moe": true/false/null,
+  "num_experts": number or null,
+  "multimodal": true/false/null,
+  "training_data_sources": ["source1", "source2"] or [],
+  "training_data_composition": "description or null",
+  "training_period_start": "YYYY-MM-DD or null",
+  "training_period_end": "YYYY-MM-DD or null",
+  "evidence_types": ["E1", "E2", "E3", "E4", "E5"] based on disclosure level,
+  "confidence": "high/medium/low",
+  "raw_snippets": ["relevant quote 1", "relevant quote 2"]
+}}
+
+Evidence type mapping:
+- E1: Direct disclosure (company blog, paper, official announcement)
+- E2: Compute evidence (FLOPs, hardware mentioned)
+- E3: Architecture evidence (parameters, MoE details)
+- E4: Third-party analysis (research paper, news article)
+- E5: Qualitative hints (vague mentions, speculation)
+
+Return only valid JSON, no additional text."""
+        
+        return prompt
+    
+    def _validate_extraction(
+        self,
+        extracted: Dict[str, Any],
+        model_id: str,
+        provider: str
+    ) -> Dict[str, Any]:
+        """Validate and normalize extracted data"""
+        validated = {
+            "release_date": self._parse_date(extracted.get("release_date")),
+            "architecture_type": extracted.get("architecture_type"),
+            "is_moe": extracted.get("is_moe"),
+            "num_experts": extracted.get("num_experts"),
+            "multimodal": extracted.get("multimodal"),
+            "training_data_sources": extracted.get("training_data_sources", []),
+            "training_data_composition": extracted.get("training_data_composition"),
+            "training_period_start": self._parse_date(extracted.get("training_period_start")),
+            "training_period_end": self._parse_date(extracted.get("training_period_end")),
+            "evidence_types": extracted.get("evidence_types", []),
+            "confidence": extracted.get("confidence", "medium"),
+            "raw_snippets": extracted.get("raw_snippets", []),
+        }
+        
+        # Normalize architecture type
+        if validated["architecture_type"]:
+            arch = validated["architecture_type"].lower()
+            if "moe" in arch or "mixture" in arch:
+                validated["architecture_type"] = "MoE"
+                validated["is_moe"] = True
+            elif "transformer" in arch:
+                validated["architecture_type"] = "Transformer"
+            else:
+                validated["architecture_type"] = "Transformer"  # Default
+        
+        return validated
+    
+    def _parse_date(self, date_str: Optional[str]) -> Optional[str]:
+        """Parse date string to ISO format"""
+        if not date_str or date_str.lower() in ["null", "none", ""]:
+            return None
+        
+        # Try to extract date from various formats
+        # YYYY-MM-DD
+        if re.match(r'^\d{4}-\d{2}-\d{2}$', date_str):
+            return date_str
+        
+        # Try to parse common formats
+        try:
+            dt = datetime.fromisoformat(date_str.replace('Z', '+00:00'))
+            return dt.strftime("%Y-%m-%d")
+        except (ValueError, AttributeError):
+            pass
+        
+        # Try year only
+        year_match = re.search(r'\b(20\d{2})\b', date_str)
+        if year_match:
+            return f"{year_match.group(1)}-01-01"
+        
+        return None
+    
+    def _empty_extraction(self) -> Dict[str, Any]:
+        """Return empty extraction result"""
+        return {
+            "release_date": None,
+            "architecture_type": None,
+            "is_moe": None,
+            "num_experts": None,
+            "multimodal": None,
+            "training_data_sources": [],
+            "training_data_composition": None,
+            "training_period_start": None,
+            "training_period_end": None,
+            "evidence_types": [],
+            "confidence": "low",
+            "raw_snippets": [],
+        }
+

registry/enrichment/web_enrichment.py

@@ -0,0 +1,220 @@
+"""
+Web Search Enrichment Service
+Uses Exa API to search for model information and extract metadata
+"""
+
+import os
+import sys
+from pathlib import Path
+from typing import Dict, List, Optional, Any
+from datetime import datetime, timedelta
+import json
+
+# Add project root to path
+project_root = Path(__file__).parent.parent.parent
+sys.path.insert(0, str(project_root))
+
+try:
+    from ingestion.exa_client import ExaClient
+except ImportError:
+    # Fallback if exa_client not available
+    ExaClient = None
+
+from .llm_extractor import LLMExtractor
+
+
+class WebModelEnricher:
+    """Enriches model metadata using web search via Exa API"""
+    
+    def __init__(
+        self,
+        exa_api_key: Optional[str] = None,
+        llm_provider: str = "openai",
+        llm_api_key: Optional[str] = None
+    ):
+        """
+        Initialize web enricher
+        
+        Args:
+            exa_api_key: Exa API key (if None, reads from env)
+            llm_provider: LLM provider ("openai" or "anthropic")
+            llm_api_key: LLM API key (if None, reads from env)
+        """
+        # Initialize Exa client
+        if ExaClient:
+            try:
+                self.exa_client = ExaClient(api_key=exa_api_key)
+            except Exception as e:
+                print(f"Warning: Exa client initialization failed: {e}")
+                self.exa_client = None
+        else:
+            self.exa_client = None
+        
+        # Initialize LLM extractor
+        try:
+            self.llm_extractor = LLMExtractor(provider=llm_provider, api_key=llm_api_key)
+        except Exception as e:
+            print(f"Warning: LLM extractor initialization failed: {e}")
+            self.llm_extractor = None
+    
+    def enrich_model(
+        self,
+        model_id: str,
+        provider: str,
+        existing_data: Optional[Dict[str, Any]] = None,
+        num_search_results: int = 5
+    ) -> Dict[str, Any]:
+        """
+        Enrich model metadata using web search
+        
+        Args:
+            model_id: Model identifier
+            provider: Model provider
+            existing_data: Existing model data (for context)
+            num_search_results: Number of search results to fetch
+        
+        Returns:
+            Dict with enriched metadata
+        """
+        if not self.exa_client:
+            return self._empty_enrichment()
+        
+        # Build search queries
+        queries = self._build_search_queries(model_id, provider)
+        
+        # Search for information
+        all_results = []
+        for query in queries:
+            try:
+                results = self.exa_client.search(
+                    query=query,
+                    num_results=num_search_results
+                )
+                all_results.extend(results)
+            except Exception as e:
+                print(f"Exa search error for query '{query}': {e}")
+                continue
+        
+        if not all_results:
+            return self._empty_enrichment()
+        
+        # Deduplicate by URL
+        seen_urls = set()
+        unique_results = []
+        for result in all_results:
+            if result.url not in seen_urls:
+                seen_urls.add(result.url)
+                unique_results.append(result)
+        
+        # Sort by score
+        unique_results.sort(key=lambda x: x.score, reverse=True)
+        
+        # Fetch content from top results
+        top_urls = [r.url for r in unique_results[:num_search_results]]
+        contents = {}
+        if self.exa_client:
+            try:
+                contents = self.exa_client.get_contents(top_urls)
+            except Exception as e:
+                print(f"Error fetching contents: {e}")
+        
+        # Combine all content
+        combined_content = self._combine_content(unique_results, contents)
+        
+        # Extract structured data using LLM
+        extracted = {}
+        if self.llm_extractor and combined_content:
+            try:
+                extracted = self.llm_extractor.extract_model_metadata(
+                    model_id=model_id,
+                    provider=provider,
+                    web_content=combined_content,
+                    context=existing_data
+                )
+            except Exception as e:
+                print(f"LLM extraction error: {e}")
+        
+        # Build enrichment result
+        enrichment = {
+            "release_date": extracted.get("release_date"),
+            "architecture_type": extracted.get("architecture_type"),
+            "is_moe": extracted.get("is_moe"),
+            "num_experts": extracted.get("num_experts"),
+            "multimodal": extracted.get("multimodal"),
+            "training_data_sources": extracted.get("training_data_sources", []),
+            "training_data_composition": extracted.get("training_data_composition"),
+            "training_period_start": extracted.get("training_period_start"),
+            "training_period_end": extracted.get("training_period_end"),
+            "evidence_types": extracted.get("evidence_types", []),
+            "confidence": extracted.get("confidence", "medium"),
+            "sources": [
+                {
+                    "type": "web_search",
+                    "url": result.url,
+                    "title": result.title,
+                    "score": result.score,
+                    "published_date": result.published_date,
+                }
+                for result in unique_results[:num_search_results]
+            ],
+            "raw_evidence_snippets": [
+                {
+                    "text": snippet,
+                    "source_url": unique_results[0].url if unique_results else None,
+                }
+                for snippet in extracted.get("raw_snippets", [])
+            ],
+        }
+        
+        return enrichment
+    
+    def _build_search_queries(self, model_id: str, provider: str) -> List[str]:
+        """Build search queries for model information"""
+        queries = [
+            f"{model_id} {provider} release date architecture training data",
+            f"{model_id} {provider} system card technical details",
+            f"{model_id} {provider} training dataset sources",
+            f"{model_id} {provider} model card paper",
+        ]
+        
+        # Add provider-specific queries
+        if provider.lower() in ["openai", "anthropic", "google", "meta"]:
+            queries.append(f"{provider} {model_id} official announcement blog")
+        
+        return queries
+    
+    def _combine_content(
+        self,
+        results: List[Any],
+        contents: Dict[str, str]
+    ) -> str:
+        """Combine content from search results"""
+        combined = []
+        
+        for result in results:
+            # Prefer full content, fallback to summary
+            if result.url in contents:
+                combined.append(f"--- Content from {result.url} ---\n{contents[result.url]}")
+            elif result.summary:
+                combined.append(f"--- Summary from {result.url} ---\n{result.title}\n{result.summary}")
+        
+        return "\n\n".join(combined)
+    
+    def _empty_enrichment(self) -> Dict[str, Any]:
+        """Return empty enrichment result"""
+        return {
+            "release_date": None,
+            "architecture_type": None,
+            "is_moe": None,
+            "num_experts": None,
+            "multimodal": None,
+            "training_data_sources": [],
+            "training_data_composition": None,
+            "training_period_start": None,
+            "training_period_end": None,
+            "evidence_types": [],
+            "confidence": "low",
+            "sources": [],
+            "raw_evidence_snippets": [],
+        }
+

registry/evidence_profile.py

@@ -143,4 +143,68 @@ class EvidenceProfileManager:
             explanation += f". Uncertainties: {unc_desc}"
         
         return explanation
+    
+    @staticmethod
+    def generate_from_web_data(
+        web_data: Dict[str, Any],
+        existing_evidence: Optional[Dict[str, Any]] = None
+    ) -> Dict[str, Any]:
+        """
+        Generate evidence profile from web-extracted data
+        
+        Args:
+            web_data: Extracted data from web search/LLM
+            existing_evidence: Existing evidence profile to merge with
+        
+        Returns:
+            Evidence profile dict
+        """
+        evidence_types = set(web_data.get("evidence_types", []))
+        confidence = web_data.get("confidence", "medium")
+        sources_count = len(web_data.get("sources", []))
+        raw_snippets = web_data.get("raw_evidence_snippets", [])
+        
+        # Determine evidence strength based on sources and confidence
+        if sources_count >= 3 and confidence == "high":
+            strength = "S-High"
+        elif sources_count >= 2 or confidence == "high":
+            strength = "S-Medium"
+        else:
+            strength = "S-Low"
+        
+        # Identify uncertainty sources for missing information
+        uncertainty_sources = []
+        if not web_data.get("release_date"):
+            uncertainty_sources.append("U5")  # Intentional opacity or missing
+        if not web_data.get("architecture_type"):
+            uncertainty_sources.append("U3")  # Architecture unclear
+        if not web_data.get("training_data_sources"):
+            uncertainty_sources.append("U2")  # Data composition unknown
+        
+        # If we have existing evidence, merge it
+        if existing_evidence:
+            existing_types = set(existing_evidence.get("evidence_types", []))
+            evidence_types.update(existing_types)
+            
+            # Use higher strength if available
+            existing_strength = existing_evidence.get("strength")
+            if existing_strength:
+                strength_order = {"S-High": 3, "S-Medium": 2, "S-Low": 1}
+                if strength_order.get(existing_strength, 0) > strength_order.get(strength, 0):
+                    strength = existing_strength
+            
+            # Merge uncertainties
+            existing_unc = set(existing_evidence.get("uncertainty", []))
+            uncertainty_sources = list(set(uncertainty_sources) | existing_unc)
+        
+        # Create evidence profile
+        profile = {
+            "evidence_types": list(evidence_types),
+            "strength": strength,
+            "uncertainty": uncertainty_sources,
+            "generated_at": datetime.now().isoformat(),
+            "evidence_version": "1.0",
+        }
+        
+        return profile
 

registry/ingest_priority_models.py

@@ -25,6 +25,7 @@ from registry.collectors.epoch_collector import EpochCollector
 from registry.collectors.hf_collector import HuggingFaceCollector
 from registry.inference.reconciliation import TokenInferenceReconciler
 from registry.linkage import create_deal_model_linkages
+from registry.enrichment.comprehensive_enrichment import ComprehensiveModelEnricher
 from dotenv import load_dotenv
 
 # Prisma imports
@@ -41,12 +42,25 @@ load_dotenv()
 class PriorityModelIngester:
     """Programmatic ingester for priority models"""
     
-    def __init__(self):
+    def __init__(self, use_web_enrichment: bool = True):
         self.epoch_collector = EpochCollector()
         self.hf_collector = HuggingFaceCollector()
         self.inference_reconciler = TokenInferenceReconciler()
         self.prisma = None
         
+        # Initialize comprehensive enricher if web enrichment enabled
+        if use_web_enrichment:
+            try:
+                self.comprehensive_enricher = ComprehensiveModelEnricher(
+                    use_web_search=True,
+                    use_llm_extraction=True
+                )
+            except Exception as e:
+                print(f"Warning: Comprehensive enricher initialization failed: {e}")
+                self.comprehensive_enricher = None
+        else:
+            self.comprehensive_enricher = None
+        
     async def connect_db(self):
         """Connect to Prisma database"""
         if not PRISMA_AVAILABLE:
@@ -226,20 +240,63 @@ class PriorityModelIngester:
             where={"modelId": model_id}
         )
         
+        # Prepare data for Prisma (convert field names)
+        prisma_data = {}
+        
+        # Map field names from model_data to Prisma schema
+        field_mapping = {
+            "modelId": "modelId",
+            "provider": "provider",
+            "family": "family",
+            "params": "params",
+            "releaseDate": "releaseDate",
+            "architectureType": "architectureType",
+            "isMoe": "isMoe",
+            "numExperts": "numExperts",
+            "multimodal": "multimodal",
+            "tokensEstMin": "tokensEstMin",
+            "tokensEstMax": "tokensEstMax",
+            "tokensEstMid": "tokensEstMid",
+            "tokensRangeGeneratedAt": "tokensRangeGeneratedAt",
+            "evidenceTypes": "evidenceTypes",
+            "evidenceStrength": "evidenceStrength",
+            "uncertaintySources": "uncertaintySources",
+            "evidenceProfileGeneratedAt": "evidenceProfileGeneratedAt",
+            "sources": "sources",
+            "rawEvidenceSnippets": "rawEvidenceSnippets",
+            "compositionEstimates": "compositionEstimates",
+            "trainingPeriodStart": "trainingPeriodStart",
+            "trainingPeriodEnd": "trainingPeriodEnd",
+        }
+        
+        for key, value in model_data.items():
+            if key in field_mapping and value is not None:
+                prisma_key = field_mapping[key]
+                # Convert date strings to datetime if needed
+                if prisma_key in ["releaseDate", "tokensRangeGeneratedAt", "evidenceProfileGeneratedAt", "trainingPeriodStart", "trainingPeriodEnd"]:
+                    if isinstance(value, str):
+                        try:
+                            prisma_data[prisma_key] = datetime.fromisoformat(value.replace("Z", "+00:00"))
+                        except:
+                            pass
+                    elif isinstance(value, datetime):
+                        prisma_data[prisma_key] = value
+                else:
+                    prisma_data[prisma_key] = value
+        
         if existing:
             # Update
+            prisma_data["updatedAt"] = datetime.now()
             updated = await self.prisma.modelregistry.update(
                 where={"id": existing.id},
-                data={
-                    **{k: v for k, v in model_data.items() if k != "modelId"},
-                    "updatedAt": datetime.now(),
-                }
+                data=prisma_data
             )
             return updated.id
         else:
             # Create
+            prisma_data["modelId"] = model_id
             created = await self.prisma.modelregistry.create(
-                data=model_data
+                data=prisma_data
             )
             return created.id
     
@@ -317,24 +374,40 @@ class PriorityModelIngester:
         
         print(f"\n📦 Processing: {model_id} ({provider})")
         
-        # Step 1: Fetch metadata
-        print(f"  🔍 Fetching metadata...")
+        # Step 1: Fetch metadata from Epoch and HF
+        print(f"  🔍 Fetching metadata from Epoch/HF...")
         epoch_data = await self.fetch_epoch_data(model_id, provider)
         hf_data = await self.fetch_hf_data(model_id, provider)
         
-        # Step 2: Merge metadata
+        # Step 2: Merge metadata from Epoch/HF
         model_data = self.merge_metadata(priority_model, epoch_data, hf_data)
         
-        # Step 3: Run token inference
+        # Step 3: Web enrichment (if enabled)
+        if self.comprehensive_enricher:
+            print(f"  🌐 Running web enrichment...")
+            try:
+                # Use comprehensive enricher which includes web search
+                enriched = await self.comprehensive_enricher.enrich_model(
+                    model_id=model_id,
+                    provider=provider,
+                    family=priority_model.get("family"),
+                    existing_data=model_data
+                )
+                # Merge web enrichment results
+                model_data.update(enriched)
+            except Exception as e:
+                print(f"    Warning: Web enrichment error: {e}")
+        
+        # Step 4: Run token inference
         print(f"  🧠 Running token inference...")
         inference_results = await self.run_token_inference(model_data)
         model_data.update(inference_results)
         
-        # Step 4: Store in database
+        # Step 5: Store in database
         print(f"  💾 Storing in database...")
         await self.upsert_model(model_data)
         
-        # Step 5: Create linkages
+        # Step 6: Create linkages
         print(f"  🔗 Creating deal linkages...")
         await self.create_linkages(model_id)
         

registry/requirements.txt

@@ -4,3 +4,6 @@ huggingface-hub>=0.16.0
 pandas>=2.0.0
 python-dotenv>=1.0.0
 prisma>=0.11.0
+openai>=1.0.0
+anthropic>=0.18.0
+requests>=2.31.0