Add image-generation task exclusion to leaderboard views

app/page.tsx

@@ -1,10 +1,11 @@
 import type { Metadata } from 'next'
-import { fetchLeaderboard, fetchBenchmarkVersions } from '@/lib/api'
-import { calculateRanks, transformLeaderboardEntry } from '@/lib/transforms'
+import { fetchLeaderboard, fetchBenchmarkVersions, fetchSubmission } from '@/lib/api'
+import { calculateRanks, transformLeaderboardEntry, EPSILON, estimateSuccessfulTasks } from '@/lib/transforms'
+import { isExcludedLeaderboardTask } from '@/lib/task-metadata'
 import { LeaderboardView } from '@/components/leaderboard-view'
 
 interface HomeProps {
-  searchParams: Promise<{ version?: string; view?: string; official?: string }>
+  searchParams: Promise<{ version?: string; view?: string; official?: string; excludeImageGen?: string }>
 }
 
 export async function generateMetadata({ searchParams }: HomeProps): Promise<Metadata> {
@@ -42,13 +43,94 @@ export async function generateMetadata({ searchParams }: HomeProps): Promise<Met
 }
 
 export default async function Home({ searchParams }: HomeProps) {
-  const { version, official } = await searchParams
+  const { version, official, excludeImageGen } = await searchParams
   const officialOnly = official !== 'false'
+  const excludeImageGenBool = excludeImageGen === 'true'
   const [response, versionsResponse] = await Promise.all([
     fetchLeaderboard(version, { officialOnly }),
     fetchBenchmarkVersions(),
   ])
-  const entries = calculateRanks(response.leaderboard.map(transformLeaderboardEntry))
+  let transformedEntries = response.leaderboard.map(transformLeaderboardEntry)
+
+  if (excludeImageGenBool) {
+    // Fetch best submission details and adjust scores to exclude image-generation tasks.
+    const bestSubmissions = []
+    const batchSize = 10
+
+    for (let i = 0; i < transformedEntries.length; i += batchSize) {
+      const batch = transformedEntries.slice(i, i + batchSize)
+      const results = await Promise.all(
+        batch.map(entry =>
+          fetchSubmission(entry.submission_id)
+            .then(res => res.submission)
+            .catch(err => {
+              console.error('Failed to fetch submission for', entry.submission_id, err)
+              return null
+            })
+        )
+      )
+      bestSubmissions.push(...results)
+    }
+
+    const adjustedEntries = []
+
+    for (let i = 0; i < transformedEntries.length; i++) {
+      const entry = transformedEntries[i]
+      const sub = bestSubmissions[i]
+
+      // Preserve the original entry if submission details are temporarily unavailable.
+      if (!sub) {
+        adjustedEntries.push(entry)
+        continue
+      }
+
+      const excludedTasks = sub.tasks.filter(task => isExcludedLeaderboardTask(task.task_id))
+      if (excludedTasks.length === 0) {
+        adjustedEntries.push(entry)
+        continue
+      }
+
+      const excludedScore = excludedTasks.reduce((sum, task) => sum + task.score, 0)
+      const excludedMax = excludedTasks.reduce((sum, task) => sum + task.max_score, 0)
+      const remainingTaskCount = sub.tasks.filter(task => !isExcludedLeaderboardTask(task.task_id)).length
+
+      const adjustedScore = sub.total_score - excludedScore
+      const adjustedMax = sub.max_score - excludedMax
+      if (adjustedMax <= 0 || remainingTaskCount <= 0) {
+        adjustedEntries.push(entry)
+        continue
+      }
+
+      const adjustedPercentage = adjustedScore / adjustedMax
+      if (!Number.isFinite(adjustedPercentage)) {
+        adjustedEntries.push(entry)
+        continue
+      }
+
+      const bestCost = entry.best_cost_usd
+      let value_score: number | null = null
+      if (bestCost != null && bestCost > EPSILON) {
+        value_score = (adjustedPercentage * 100) / bestCost
+      }
+
+      let cpst: number | null = null
+      const successfulTasks = estimateSuccessfulTasks(adjustedPercentage, remainingTaskCount)
+      if (bestCost != null && bestCost > EPSILON && successfulTasks != null && successfulTasks > 0) {
+        cpst = bestCost / successfulTasks
+      }
+
+      adjustedEntries.push({
+        ...entry,
+        percentage: adjustedPercentage * 100,
+        value_score,
+        cpst,
+      })
+    }
+
+    transformedEntries = adjustedEntries
+  }
+
+  const entries = calculateRanks(transformedEntries)
   const latestTimestamp = entries.reduce((latest, entry) => {
     const current = new Date(entry.timestamp).getTime()
     return Number.isNaN(current) ? latest : Math.max(latest, current)
@@ -71,6 +153,7 @@ export default async function Home({ searchParams }: HomeProps) {
       versions={versionsResponse.versions}
       currentVersion={version ?? null}
       officialOnly={officialOnly}
+      excludeImageGen={excludeImageGenBool}
     />
   )
 }

app/submission/[id]/page.tsx

@@ -13,16 +13,18 @@ import { PROVIDER_COLORS } from '@/lib/types'
 import { formatDistanceToNow } from 'date-fns'
 import { fetchSubmission } from '@/lib/api'
 import { transformSubmission } from '@/lib/transforms'
+import { isExcludedLeaderboardTask } from '@/lib/task-metadata'
 
 interface SubmissionPageProps {
   params: Promise<{ id: string }>
-  searchParams: Promise<{ official?: string }>
+  searchParams: Promise<{ official?: string; excludeImageGen?: string }>
 }
 
 export default async function SubmissionPage({ params, searchParams }: SubmissionPageProps) {
   const { id } = await params
-  const { official } = await searchParams
+  const { official, excludeImageGen } = await searchParams
   const officialOnly = official !== 'false'
+  const excludeImageGenBool = excludeImageGen === 'true'
   let submission
 
   try {
@@ -68,7 +70,23 @@ export default async function SubmissionPage({ params, searchParams }: Submissio
     notFound()
   }
 
-  const categoryStats = submission.task_results.reduce(
+  // Apply image generation exclusion if requested
+  let displayTasks = submission.task_results
+  let displayTotalScore = submission.total_score
+  let displayMaxScore = submission.max_score
+
+  if (excludeImageGenBool) {
+    const excludedTasks = submission.task_results.filter(task => isExcludedLeaderboardTask(task.task_id))
+    if (excludedTasks.length > 0) {
+      const excludedScore = excludedTasks.reduce((sum, task) => sum + task.score, 0)
+      const excludedMax = excludedTasks.reduce((sum, task) => sum + task.max_score, 0)
+      displayTotalScore = submission.total_score - excludedScore
+      displayMaxScore = submission.max_score - excludedMax
+      displayTasks = submission.task_results.filter(task => !isExcludedLeaderboardTask(task.task_id))
+    }
+  }
+
+  const categoryStats = displayTasks.reduce(
     (acc, task) => {
       if (!acc[task.category]) {
         acc[task.category] = { total: 0, max: 0, count: 0 }
@@ -191,8 +209,8 @@ export default async function SubmissionPage({ params, searchParams }: Submissio
         <div className="grid grid-cols-1 lg:grid-cols-3 gap-6 mb-8">
           <div className="lg:col-span-1">
             <ScoreGauge
-              score={submission.total_score}
-              maxScore={submission.max_score}
+              score={displayTotalScore}
+              maxScore={displayMaxScore}
             />
           </div>
 
@@ -237,10 +255,10 @@ export default async function SubmissionPage({ params, searchParams }: Submissio
               Task Breakdown
             </h2>
             <p className="text-sm text-muted-foreground">
-              {submission.task_results.length} tasks completed
+              {displayTasks.length} tasks completed
             </p>
           </div>
-          <TaskBreakdown tasks={submission.task_results} />
+          <TaskBreakdown tasks={displayTasks} />
 
           {/* Hardware Info */}
           {submission.metadata.system && (
@@ -276,8 +294,6 @@ export default async function SubmissionPage({ params, searchParams }: Submissio
           </div>
         </Card>
       </div>
-
-
     </div>
   )
 }

components/leaderboard-header.tsx

@@ -19,10 +19,12 @@ interface LeaderboardHeaderProps {
     scoreMode: ScoreMode
     officialOnly: boolean
     openWeightsOnly: boolean
+    excludeImageGen?: boolean
     onViewChange: (view: ViewMode) => void
     onScoreModeChange: (mode: ScoreMode) => void
     onOfficialOnlyChange: (officialOnly: boolean) => void
     onOpenWeightsOnlyChange: (openWeightsOnly: boolean) => void
+    onExcludeImageGenChange?: (value: boolean) => void
     onClearProviderFilter: () => void
 }
 
@@ -38,10 +40,12 @@ export function LeaderboardHeader({
     scoreMode,
     officialOnly,
     openWeightsOnly,
+    excludeImageGen,
     onViewChange,
     onScoreModeChange,
     onOfficialOnlyChange,
     onOpenWeightsOnlyChange,
+    onExcludeImageGenChange,
     onClearProviderFilter,
 }: LeaderboardHeaderProps) {
     return (
@@ -111,6 +115,15 @@ export function LeaderboardHeader({
                                 />
                                 Open-weight only
                             </label>
+                            <label className="flex items-center gap-2 text-xs text-muted-foreground/90 cursor-pointer hover:text-foreground transition-colors">
+                                <input
+                                    type="checkbox"
+                                    checked={excludeImageGen}
+                                    onChange={(e) => onExcludeImageGenChange?.(e.target.checked)}
+                                    className="h-3.5 w-3.5 rounded border border-border/70 bg-muted/30 text-muted-foreground checked:border-muted-foreground checked:bg-muted-foreground focus-visible:outline-none focus-visible:ring-1 focus-visible:ring-ring focus-visible:ring-offset-0"
+                                />
+                                Exclude image generation tasks
+                            </label>
                             <span className="text-xs text-muted-foreground/60">Updated {lastUpdated}</span>
                         </div>
                     </div>
@@ -227,6 +240,15 @@ export function LeaderboardHeader({
                         />
                         Open-weight only
                     </label>
+                    <label className="flex items-center gap-2 cursor-pointer hover:text-foreground transition-colors">
+                        <input
+                            type="checkbox"
+                            checked={excludeImageGen}
+                            onChange={(e) => onExcludeImageGenChange?.(e.target.checked)}
+                            className="h-3.5 w-3.5 rounded border border-border/70 bg-muted/30 text-muted-foreground checked:border-muted-foreground checked:bg-muted-foreground focus-visible:outline-none focus-visible:ring-1 focus-visible:ring-ring focus-visible:ring-offset-0"
+                        />
+                        Exclude image generation tasks
+                    </label>
                     <span className="text-muted-foreground/60">Updated {lastUpdated}</span>
                 </div>
             </div>

components/leaderboard-view.tsx

@@ -1,6 +1,6 @@
 'use client'
 
-import { useCallback, useMemo, useState } from 'react'
+import { useCallback, useMemo, useState, useEffect } from 'react'
 import { useSearchParams, useRouter, usePathname } from 'next/navigation'
 import type { LeaderboardEntry, BenchmarkVersion } from '@/lib/types'
 import { PROVIDER_COLORS } from '@/lib/types'
@@ -25,9 +25,10 @@ interface LeaderboardViewProps {
     versions: BenchmarkVersion[]
     currentVersion: string | null
     officialOnly: boolean
+    excludeImageGen?: boolean
 }
 
-export function LeaderboardView({ entries, lastUpdated, versions, currentVersion, officialOnly }: LeaderboardViewProps) {
+export function LeaderboardView({ entries, lastUpdated, versions, currentVersion, officialOnly, excludeImageGen = false }: LeaderboardViewProps) {
     const searchParams = useSearchParams()
     const router = useRouter()
     const pathname = usePathname()
@@ -51,6 +52,7 @@ export function LeaderboardView({ entries, lastUpdated, versions, currentVersion
     const [providerFilter, setProviderFilterState] = useState<string | null>(initialProvider)
     const [openWeightsOnly, setOpenWeightsOnlyState] = useState<boolean>(initialOpenWeights)
     const [graphSubTab, setGraphSubTabState] = useState<GraphSubTab>(initialGraphTab)
+    const [excludeImageGenLocal, setExcludeImageGenState] = useState<boolean>(excludeImageGen)
 
     // Helper to update URL params without full page reload
     const updateUrl = useCallback((updates: Record<string, string | null>) => {
@@ -94,6 +96,16 @@ export function LeaderboardView({ entries, lastUpdated, versions, currentVersion
         updateUrl({ graph: t === 'scatter' ? null : t })
     }, [updateUrl])
 
+    const setExcludeImageGen = useCallback((v: boolean) => {
+        setExcludeImageGenState(v)
+        updateUrl({ excludeImageGen: v ? 'true' : null })
+    }, [updateUrl])
+
+    // Sync prop to local state when URL changes (e.g., browser back/forward)
+    useEffect(() => {
+        setExcludeImageGenState(excludeImageGen)
+    }, [excludeImageGen])
+
     const setOfficialOnly = useCallback((v: boolean) => {
         setOfficialOnlyState(v)
         updateUrl({ official: v ? null : 'false' })
@@ -134,10 +146,12 @@ export function LeaderboardView({ entries, lastUpdated, versions, currentVersion
                 scoreMode={scoreMode}
                 officialOnly={officialOnlyState}
                 openWeightsOnly={openWeightsOnly}
+                excludeImageGen={excludeImageGenLocal}
                 onViewChange={setView}
                 onScoreModeChange={setScoreMode}
                 onOfficialOnlyChange={setOfficialOnly}
                 onOpenWeightsOnlyChange={setOpenWeightsOnly}
+                onExcludeImageGenChange={setExcludeImageGen}
                 onClearProviderFilter={() => setProviderFilter(null)}
             />
 
@@ -169,7 +183,7 @@ export function LeaderboardView({ entries, lastUpdated, versions, currentVersion
                             <ScatterGraphs entries={filteredEntries} scoreMode={scoreMode} />
                         )}
                         {graphSubTab === 'heatmap' && (
-                            <TaskHeatmap entries={filteredEntries} scoreMode={scoreMode} />
+                            <TaskHeatmap entries={filteredEntries} scoreMode={scoreMode} excludeImageGen={excludeImageGenLocal} />
                         )}
                         {graphSubTab === 'distribution' && (
                             <ScoreDistribution entries={filteredEntries} scoreMode={scoreMode} currentVersion={currentVersion} officialOnly={officialOnlyState} />

components/task-breakdown.tsx

@@ -8,6 +8,7 @@ import { ColoredProgress } from '@/components/ui/colored-progress'
 import { ChevronDown, ChevronRight } from 'lucide-react'
 import type { TaskResult } from '@/lib/types'
 import { CATEGORY_ICONS } from '@/lib/types'
+import { isExcludedLeaderboardTask } from '@/lib/task-metadata'
 import {
   Collapsible,
   CollapsibleContent,
@@ -16,6 +17,7 @@ import {
 
 interface TaskBreakdownProps {
   tasks: TaskResult[]
+  excludeImageGen?: boolean
 }
 
 const getScoreColor = (score: number, maxScore: number) => {
@@ -46,7 +48,7 @@ const getGradingBadgeVariant = (
   }
 }
 
-export function TaskBreakdown({ tasks }: TaskBreakdownProps) {
+export function TaskBreakdown({ tasks, excludeImageGen = false }: TaskBreakdownProps) {
   const [openTasks, setOpenTasks] = useState<Set<string>>(new Set())
 
   const toggleTask = (taskId: string) => {
@@ -61,7 +63,7 @@ export function TaskBreakdown({ tasks }: TaskBreakdownProps) {
 
   return (
     <div className="space-y-2">
-      {tasks.map((task) => {
+      {tasks.filter(task => !(excludeImageGen && isExcludedLeaderboardTask(task.task_id))).map((task) => {
         const percentage = (task.score / task.max_score) * 100
         const isOpen = openTasks.has(task.task_id)