Fix AI assignment returning nothing: cap tokens, optimize prompt, show errors

- Cap maxTokens at 12000 (was unlimited dynamic calc that could exceed model limits)
- Replace massive EXISTING array with compact CURRENT_JUROR_LOAD counts and
  ALREADY_ASSIGNED per-project map (keeps prompt small across batches)
- Add coverage gap-filler: algorithmically fills projects below required reviews
- Show error state inline on page when AI fails (red banner with message)
- Add server-side logging for debugging assignment flow
- Reduce batch size to 10 projects

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/app/(admin)/admin/rounds/[roundId]/page.tsx

@@ -186,7 +186,8 @@ export default function RoundDetailPage() {
       })
     },
     onError: (err) => {
-      toast.error(`AI generation failed: ${err.message}`)
+      toast.error(`AI generation failed: ${err.message}`, { duration: 15000 })
+      console.error('[AI Assignment]', err)
     },
   })
   const [exportOpen, setExportOpen] = useState(false)
@@ -1725,6 +1726,19 @@ export default function RoundDetailPage() {
                     </div>
                   </div>
                 )}
+                {aiAssignmentMutation.error && !aiAssignmentMutation.isPending && (
+                  <div className="flex items-center gap-3 p-3 rounded-lg bg-red-50 border border-red-200 dark:bg-red-950/20 dark:border-red-800">
+                    <AlertTriangle className="h-5 w-5 text-red-600 shrink-0" />
+                    <div className="flex-1">
+                      <p className="text-sm font-medium text-red-800 dark:text-red-200">
+                        AI generation failed
+                      </p>
+                      <p className="text-xs text-red-600 dark:text-red-400">
+                        {aiAssignmentMutation.error.message}
+                      </p>
+                    </div>
+                  </div>
+                )}
                 {aiAssignmentMutation.data && !aiAssignmentMutation.isPending && (
                   <div className="flex items-center gap-3 p-3 rounded-lg bg-emerald-50 border border-emerald-200 dark:bg-emerald-950/20 dark:border-emerald-800">
                     <CheckCircle2 className="h-5 w-5 text-emerald-600 shrink-0" />

src/server/routers/roundAssignment.ts

@@ -152,6 +152,7 @@ export const roundAssignmentRouter = router({
       }
 
       // Call AI service
+      console.log(`[AI Assignment Router] Starting for ${projects.length} projects, ${jurors.length} jurors, ${input.requiredReviews} reviews/project, max ${maxPerJuror}/juror`)
       const result = await generateAIAssignments(
         jurors,
         projects,
@@ -159,6 +160,7 @@ export const roundAssignmentRouter = router({
         ctx.user.id,
         input.roundId,
       )
+      console.log(`[AI Assignment Router] Got ${result.suggestions.length} suggestions, success=${result.success}, fallback=${result.fallbackUsed}`)
 
       // Filter out COI pairs and already-assigned pairs
       const existingPairSet = new Set(existingAssignments.map((a) => `${a.userId}:${a.projectId}`))

src/server/services/ai-assignment.ts

@@ -173,9 +173,9 @@ async function processAssignmentBatch(
 
   try {
     // Calculate maxTokens based on expected assignments
-    // ~150 tokens per assignment JSON object
+    // ~150 tokens per assignment JSON object, capped at 12000
     const expectedAssignments = batchProjects.length * constraints.requiredReviewsPerProject
-    const estimatedTokens = Math.max(4000, expectedAssignments * 200 + 500)
+    const estimatedTokens = Math.min(12000, Math.max(4000, expectedAssignments * 200 + 500))
 
     const params = buildCompletionParams(model, {
       messages: [
@@ -359,13 +359,33 @@ function buildBatchPrompt(
 
   const expectedTotal = projects.length * constraints.requiredReviewsPerProject
 
+  // Instead of full existing assignment list, send per-juror current load counts
+  // This keeps the prompt shorter as batches accumulate
+  const jurorCurrentLoad: Record<string, number> = {}
+  for (const a of constraints.existingAssignments) {
+    const anonId = jurorIdMap.get(a.jurorId)
+    if (anonId) jurorCurrentLoad[anonId] = (jurorCurrentLoad[anonId] || 0) + 1
+  }
+
+  // Also track which projects in this batch already have assignments
+  const projectExistingReviewers: Record<string, string[]> = {}
+  for (const a of constraints.existingAssignments) {
+    const anonProjectId = projectIdMap.get(a.projectId)
+    const anonJurorId = jurorIdMap.get(a.jurorId)
+    if (anonProjectId && anonJurorId) {
+      if (!projectExistingReviewers[anonProjectId]) projectExistingReviewers[anonProjectId] = []
+      projectExistingReviewers[anonProjectId].push(anonJurorId)
+    }
+  }
+
   return `JURORS: ${JSON.stringify(jurors)}
 PROJECTS: ${JSON.stringify(projects)}
 REVIEWS_PER_PROJECT: ${constraints.requiredReviewsPerProject} (each project MUST get exactly ${constraints.requiredReviewsPerProject} different jurors)
 MAX_PER_JUROR: ${constraints.maxAssignmentsPerJuror || 'unlimited'} (HARD LIMIT — never exceed)${jurorLimitsStr}${targetStr}
-EXISTING: ${JSON.stringify(anonymousExisting)}
-EXPECTED_OUTPUT_SIZE: approximately ${expectedTotal} assignment objects (${projects.length} projects × ${constraints.requiredReviewsPerProject} reviewers each)
-IMPORTANT: Every project must appear ${constraints.requiredReviewsPerProject} times with ${constraints.requiredReviewsPerProject} DIFFERENT juror_ids. Pick the least-loaded jurors first. Never exceed a juror's max.
+CURRENT_JUROR_LOAD: ${JSON.stringify(jurorCurrentLoad)} (add these to currentAssignmentCount to get true total)
+ALREADY_ASSIGNED: ${JSON.stringify(projectExistingReviewers)} (do NOT assign these juror-project pairs again)
+EXPECTED_OUTPUT: ${expectedTotal} assignment objects (${projects.length} projects × ${constraints.requiredReviewsPerProject} reviewers)
+IMPORTANT: Every project must appear ${constraints.requiredReviewsPerProject} times with ${constraints.requiredReviewsPerProject} DIFFERENT juror_ids. Pick the least-loaded jurors first.
 Return JSON: {"assignments": [...]}`
 }