@goodandready/dsh-moa 0.2.25 → 0.2.27

package/lib/moa-report.js

@@ -0,0 +1,124 @@
+/**
+ * Automated Benchmark & Post-Mortem PR Reports
+ * Feature 9 (Automated Benchmark & Post-Mortem PR Reports)
+ */
+
+export function generateMoABenchmarkReport({
+  runId = '',
+  timestamp = new Date().toISOString(),
+  preset = 'default',
+  prompt = '',
+  durationMs = 0,
+  usage = {},
+  costUsd = 0,
+  candidates = [],
+  winningIndex = 1,
+  winningLabel = '',
+  consensus = null,
+  testGate = null,
+  isComposite = false,
+}) {
+  const durationSec = (durationMs / 1000).toFixed(2)
+  const totalCost = typeof costUsd === 'number' ? costUsd.toFixed(4) : '0.0000'
+  const inTokens = usage?.inputTokens || 0
+  const outTokens = usage?.outputTokens || 0
+  const totalTokens = usage?.totalTokens || (inTokens + outTokens)
+
+  const candidateRows = (candidates || []).map((c) => {
+    const idx = c.index || '?'
+    const label = c.label || `${c.slot?.provider}:${c.slot?.model}`
+    const role = c.role_persona || 'general'
+    const status = c.ok ? '✅ OK' : '❌ Failed'
+    const candTokens = c.usage?.totalTokens || (c.usage?.inputTokens || 0) + (c.usage?.outputTokens || 0)
+    const candCost = c.costUsd ? `$${c.costUsd.toFixed(4)}` : '$0.0000'
+    const score = consensus?.averageScores?.[c.index] !== undefined
+      ? `${consensus.averageScores[c.index]}/10`
+      : 'N/A'
+    const isWin = c.index === winningIndex ? ' 🏆' : ''
+    return `| ${idx}${isWin} | \`${label}\` | ${role} | ${status} | ${candTokens} | ${candCost} | ${score} |`
+  }).join('\n')
+
+  let testGateSection = '### 🧪 Automated Test Gate\n- **Status**: Not Configured / Skipped'
+  if (testGate && testGate.enabled) {
+    const tgStatus = testGate.passed ? '✅ PASSED' : '❌ FAILED'
+    testGateSection = `### 🧪 Automated Test Gate
+- **Status**: ${tgStatus}
+- **Test Command**: \`${testGate.command || 'N/A'}\`
+- **Exit Code**: \`${testGate.exitCode ?? 0}\`
+- **Output Snippet**:
+\`\`\`
+${(testGate.output || 'No output recorded').slice(0, 500)}
+\`\`\``
+  }
+
+  let consensusSection = '### ⚖️ Multi-Judge Consensus\n- **Status**: Standard single-aggregator synthesis'
+  if (consensus && consensus.consensus) {
+    consensusSection = `### ⚖️ Multi-Judge Consensus
+- **Winning Candidate**: Candidate ${winningIndex} (${winningLabel})
+- **Strategy**: \`${consensus.strategy || 'majority'}\`
+- **Unanimous**: ${consensus.isUnanimous ? 'Yes ✅' : 'No (Majority)'}
+- **Summary**: ${consensus.consensusReport || 'Consensus reached'}
+`
+  }
+
+  const markdown = `# 🏆 Mixture of Agents (MoA) Benchmark & Post-Mortem Report
+
+> **Run ID**: \`${runId}\`  
+> **Timestamp**: \`${timestamp}\`  
+> **Preset**: \`${preset}\`  
+> **Total Latency**: \`${durationSec}s\` | **Total Cost**: \`$${totalCost}\` | **Tokens**: \`${totalTokens}\` (In: ${inTokens}, Out: ${outTokens})
+
+---
+
+## 🎯 Task Prompt
+\`\`\`text
+${prompt.slice(0, 500)}${prompt.length > 500 ? '...' : ''}
+\`\`\`
+
+---
+
+## 📊 Candidate Models Benchmark Matrix
+| # | Model / Provider | Role | Status | Tokens | Cost | Judge Score |
+|---|------------------|------|--------|--------|------|-------------|
+${candidateRows || '| - | No candidate data | - | - | - | - | - |'}
+
+---
+
+${consensusSection}
+
+---
+
+${testGateSection}
+
+---
+
+### 🧬 Synthesis Mode
+- **Mode**: ${isComposite ? 'Composite Hybrid AST/Block Merge (Multi-Candidate Synthesis)' : `Candidate Selection (Winning model: ${winningLabel || 'Aggregator'})`}
+`
+
+  const json = {
+    runId,
+    timestamp,
+    preset,
+    prompt,
+    durationMs,
+    usage: { inputTokens: inTokens, outputTokens: outTokens, totalTokens },
+    costUsd,
+    candidates: (candidates || []).map((c) => ({
+      index: c.index,
+      label: c.label,
+      role: c.role_persona,
+      ok: c.ok,
+      costUsd: c.costUsd || 0,
+      usage: c.usage,
+      score: consensus?.averageScores?.[c.index] ?? null,
+    })),
+    winningIndex,
+    winningLabel,
+    consensus,
+    testGate,
+    isComposite,
+  }
+
+  return { markdown, json }
+}

package/lib/moa-router.js

@@ -0,0 +1,102 @@
+/**
+ * Smart Preset Router for Mixture of Agents (MoA).
+ * Supports heuristic keyword classification and optional JEV/LLM zero-shot routing.
+ */
+
+const INTENT_RULES = [
+  { preset: 'security-audit', regex: /\b(security|vulnerability|exploit|cve|xss|injection|sanitize|threat|auth|permission|secret)\b/i },
+  { preset: 'bug-hunter', regex: /\b(bug|fix|error|crash|exception|regression|reproduce|broken|fail|issue)\b/i },
+  { preset: 'refactor-cleanup', regex: /\b(refactor|clean|simplify|ponytail|yagni|prune|dedup|modular|dead\s*code)\b/i },
+  { preset: 'code-review', regex: /\b(review|critique|audit|pr|pull\s*request|diff|inspect|check\s*code)\b/i },
+  { preset: 'frontend-ui', regex: /\b(frontend|ui|css|html|react|vue|tailwind|styling|styles?|stylesheet|component|button|layout|page|landing|modal)\b/i },
+  { preset: 'deep-architect', regex: /\b(architect|distributed|system|microservice|database|scalab|pipeline|infra|schema)\b/i },
+  { preset: 'math-logic', regex: /\b(math|algorithm|proof|matrix|calc|complexity|combinatorics|graph|tree|dynamic\s*programming)\b/i },
+  { preset: 'creative-brainstorm', regex: /\b(brainstorm|idea|concept|creative|alternative|options|feature\s*idea)\b/i },
+  { preset: 'fast-audit', regex: /\b(quick|fast|brief|summary|one-line|short|tldr)\b/i },
+]
+
+/**
+ * Fast keyword-based intent classification.
+ */
+export function classifyPromptIntent(prompt = '') {
+  if (!prompt || typeof prompt !== 'string') return null
+  const text = prompt.trim()
+  for (const rule of INTENT_RULES) {
+    if (rule.regex.test(text)) {
+      return rule.preset
+    }
+  }
+  return null
+}
+
+/**
+ * Resolves the optimal preset for a given user prompt.
+ */
+export async function resolvePresetForPrompt({
+  prompt = '',
+  presets = [],
+  defaultPreset = 'default',
+  routingModel = null,
+  callLlm = null,
+  enabled = false,
+  timeoutMs = 2000,
+}) {
+  const availableNames = (presets || []).map((p) => p.name).filter(Boolean)
+  const safeDefault = availableNames.includes(defaultPreset) ? defaultPreset : (availableNames[0] || 'default')
+
+  if (!enabled || !prompt || typeof prompt !== 'string') {
+    return { presetName: safeDefault, reason: 'default', isAutoRouted: false }
+  }
+
+  // 1. If LLM routing model (e.g. JEV) is configured and callable, run fast zero-shot classifier
+  if (routingModel?.provider && routingModel?.model && typeof callLlm === 'function') {
+    try {
+      const systemPrompt = `You are the MoA Preset Router. Given user prompt, classify which preset is best suited.
+Available presets: ${availableNames.join(', ')}
+Respond with ONLY the chosen preset name and nothing else.`
+
+      const abortCtrl = new AbortController()
+      const timer = setTimeout(() => abortCtrl.abort(), timeoutMs)
+      timer.unref?.()
+
+      const res = await callLlm({
+        provider: routingModel.provider,
+        model: routingModel.model,
+        messages: [
+          { role: 'system', content: systemPrompt },
+          { role: 'user', content: prompt.slice(0, 500) },
+        ],
+        temperature: 0.1,
+        maxTokens: 30,
+        signal: abortCtrl.signal,
+      })
+      clearTimeout(timer)
+
+      const rawChoice = typeof res === 'string' ? res : (res?.content || res?.text || '')
+      const match = availableNames.find((name) => new RegExp(`\\b${name}\\b`, 'i').test(rawChoice))
+      if (match) {
+        return {
+          presetName: match,
+          reason: 'llm_classifier',
+          routerModel: `${routingModel.provider}:${routingModel.model}`,
+          isAutoRouted: true,
+        }
+      }
+    } catch {
+      // Fallback silently to heuristic rules on timeout or network error
+    }
+  }
+
+  // 2. Keyword heuristic classifier
+  const heuristic = classifyPromptIntent(prompt)
+  if (heuristic && availableNames.includes(heuristic)) {
+    return {
+      presetName: heuristic,
+      reason: 'keyword_heuristic',
+      isAutoRouted: true,
+    }
+  }
+
+  // 3. Fallback to default
+  return { presetName: safeDefault, reason: 'default_fallback', isAutoRouted: false }
+}

package/lib/moa-runner.js

@@ -1,28 +1,26 @@
-import { bestEffort } from './best-effort.js'
-/**
- * DeepSeek Harness Mixture of Agents (MoA) — Runner Engine
- * Parallel fan-out, Consilium Round 2 peer critique, aggregator synthesis & streaming.
- */
-
 import path from 'node:path'
 import crypto from 'node:crypto'
+import { bestEffort } from './best-effort.js'
 import { extractFileBlocks, collectProjectContext, formatProjectContext, isRefinementTask, writeCandidateWorkspace, promoteCandidateWorkspace, cleanMoaWorkspaces, verifyFileSyntax } from './file-workspace.js'
 import { estimateTokenCost, summarizeMoAUsage } from './pricing.js'
-import { recordMoaRun, recordMoaRunAsync, candidatesForHistory } from './history.js'
+import { recordMoaRun, candidatesForHistory } from './history.js'
 import { createPromotedPreview } from './live-canvas.js'
 import { slotLabel, cleanAdvisoryMessages, isBroadPromptRequiringQuestions, buildQuestionSynthesisPrompt, buildCuratorSynthesisPrompt, buildSynthesisPrompt, buildPeerCritiquePrompt, ROLE_PERSONA_PROMPTS, SYSTEM_ROLE_PROPOSER, ANTIPATTERNS_RUBRIC } from './moa-prompts.js'
 import { parseWinnerIndex, parseRecommendedAssembler, parseMoACommand, stripOrSummarizeCode, formatMoAResponse } from './moa-parser.js'
+import { REFERENCE_SYSTEM_PROMPT, callWithTransientRetry, runReferencesParallel } from './moa-candidates.js'
+import { executeTestGateForCandidates, runCandidateTestGate } from './moa-test-gate.js'
+import { executeMultiJudgePanel, buildCompositeBlockDirectives } from './moa-multi-judge.js'
+import { applyBudgetGuardrails } from './moa-budget.js'
+import { extractPriorTurnBaseline, pruneMultiTurnMessages } from './moa-context.js'
+import { generateMoABenchmarkReport } from './moa-report.js'
 
 // Re-exports for consumers & backward compatibility
 export { slotLabel, cleanAdvisoryMessages, isBroadPromptRequiringQuestions, buildQuestionSynthesisPrompt, buildCuratorSynthesisPrompt, buildSynthesisPrompt, ANTIPATTERNS_RUBRIC } from './moa-prompts.js'
 export { parseWinnerIndex, parseRecommendedAssembler, parseMoACommand, stripOrSummarizeCode, formatMoAResponse } from './moa-parser.js'
 export { estimateTokenCost, summarizeMoAUsage } from './pricing.js'
-
-/**
- * Invokes LLM call with transient retry for recoverable network/rate-limit errors.
- */
-import { REFERENCE_SYSTEM_PROMPT, callWithTransientRetry, runReferencesParallel } from './moa-candidates.js'
-export { REFERENCE_SYSTEM_PROMPT, callWithTransientRetry, runReferencesParallel }
+export { REFERENCE_SYSTEM_PROMPT, callWithTransientRetry, runReferencesParallel } from './moa-candidates.js'
+export { executeTestGateForCandidates, runCandidateTestGate } from './moa-test-gate.js'
+export { streamMoATurn } from './moa-stream.js'
 
 /**
  * Executes the full Mixture of Agents pipeline.
@@ -39,18 +37,18 @@ export async function runMoAPipeline({
   liveCanvas = null,
   onStreamDelta = null,
   signal = null,
+  testGateEnabled = null,
+  testCommand = null,
+  execFn = null,
 }) {
   if (signal?.aborted) return { error: new Error('Turn aborted') }
   const startTime = Date.now()
 
   // 1. Resolve configurations
   const referenceModels = Array.isArray(preset?.reference_models) && preset.reference_models.length > 0
-    ? preset.reference_models
-    : [{ provider: 'opencode-go', model: 'deepseek-v4-flash' }]
-
+    ? preset.reference_models : [{ provider: 'opencode-go', model: 'deepseek-v4-flash' }]
   const primaryJudge = preset?.aggregator?.provider && preset?.aggregator?.model
-    ? preset.aggregator
-    : { provider: 'codex', model: 'gpt-5.6-sol' }
+    ? preset.aggregator : { provider: 'codex', model: 'gpt-5.6-sol' }
 
   const fallbackJudges = Array.isArray(preset?.aggregator_fallbacks) ? preset.aggregator_fallbacks : []
   const judgesChain = [primaryJudge, ...fallbackJudges]
@@ -60,17 +58,44 @@ export async function runMoAPipeline({
   const maxTokens = typeof preset?.max_tokens === 'number' ? preset.max_tokens : 4096
   const judgeCriteria = preset?.judge_criteria || ''
   const isFastMode = referenceModels.length === 1 && !preset?.curator_synthesis
-  const isCuratorSynthesis = Boolean(preset?.curator_synthesis)
-  const isStreamAggregator = preset?.stream_aggregator !== false
-  const isQuorumEnabled = Boolean(preset?.quorum_enabled)
-  const gracePeriodSec = typeof preset?.grace_period_sec === 'number' ? preset.grace_period_sec : 10
-  const candidateRetries = 1
-  const refTimeoutSec = typeof preset?.reference_timeout_sec === 'number' ? preset.reference_timeout_sec : 60
+  const isCuratorSynthesis = Boolean(preset?.curator_synthesis), isStreamAggregator = preset?.stream_aggregator !== false
+  const isQuorumEnabled = Boolean(preset?.quorum_enabled), gracePeriodSec = typeof preset?.grace_period_sec === 'number' ? preset.grace_period_sec : 10
+  const candidateRetries = 1, refTimeoutSec = typeof preset?.reference_timeout_sec === 'number' ? preset.reference_timeout_sec : 60
   const aggTimeoutSec = typeof preset?.aggregator_timeout_sec === 'number' ? preset.aggregator_timeout_sec : 180
-  const isBlindEvaluation = Boolean(preset?.blind_evaluation)
-  const isPeerCritiqueEnabled = Boolean(preset?.peer_critique_enabled)
-  const allowCandidateOverride = Boolean(preset?.allow_candidate_override)
-  const runId = crypto.randomUUID()
+  const isBlindEvaluation = Boolean(preset?.blind_evaluation), isPeerCritiqueEnabled = Boolean(preset?.peer_critique_enabled)
+  const allowCandidateOverride = Boolean(preset?.allow_candidate_override), runId = crypto.randomUUID()
+
+  // Feature 6: Budget Guardrail check
+  const budgetCheck = applyBudgetGuardrails({
+    references: referenceModels,
+    enabled: preset?.budget_guard_enabled,
+    maxBudgetUsd: preset?.max_budget_usd,
+    action: preset?.budget_action || 'trim',
+    prices,
+    promptLength: userPrompt?.length || 1000,
+  })
+  if (!budgetCheck.allowed) {
+    return {
+      kind: 'failure',
+      content: `🛑 Budget Guardrail Abort: ${budgetCheck.reason}`,
+      aggregator: slotLabel(primaryJudge),
+      references: [],
+      presetName: preset?.name || 'default',
+      isRefinement: false,
+      isFastMode,
+      winningIndex: 0,
+      winnerModel: 'none',
+      promotedFiles: [],
+      usage: { totalTokens: 0, totalCostUsd: 0, candidates: [], aggregator: { totalTokens: 0, costUsd: 0 } },
+      skippedFiles: 0,
+      skippedList: [],
+      durationMs: Date.now() - startTime,
+    }
+  }
+  const effectiveReferences = budgetCheck.references
+  if (budgetCheck.action === 'trim' && typeof onProgress === 'function') {
+    onProgress(`✂️ *${budgetCheck.reason}*\n`)
+  }
 
   // 2. Collect project context for refinement tasks
   const collectedCtx = await collectProjectContext(cwd, 16000)
@@ -91,23 +116,28 @@ export async function runMoAPipeline({
     })
   }
 
-  // 3. Build prompts & evaluate broad questionnaire needs
+  // 3. Build prompts & Feature 8 multi-turn context
+  const priorTurnBaseline = preset?.multi_turn_enabled !== false ? extractPriorTurnBaseline(messages) : null
+  const prunedHistory = preset?.multi_turn_enabled !== false ? pruneMultiTurnMessages(messages) : messages
+
   const candidateSystemPrompt = projectContext
     ? `${REFERENCE_SYSTEM_PROMPT}\n\n### Current Project Files & Context:\n${projectContext}`
     : REFERENCE_SYSTEM_PROMPT
 
   const askClarifyingQuestions = preset?.ask_clarifying_questions !== false
   const needsQuestions = askClarifyingQuestions && isBroadPromptRequiringQuestions(userPrompt, messages)
+  const enrichedMessages = [...prunedHistory, { role: 'user', content: userPrompt }]
 
-  const enrichedMessages = [...messages, { role: 'user', content: userPrompt }]
-
-  // 4. Parallel fan-out to candidate models with quorum & transient retry
+  // 4. Parallel fan-out to candidate models with temperature gradient & local fallback
   const referenceOutputs = await runReferencesParallel(
-    referenceModels,
+    effectiveReferences,
     enrichedMessages,
     {
       systemPrompt: candidateSystemPrompt,
       temperature: refTemp,
+      temperature_gradient_enabled: Boolean(preset?.temperature_gradient_enabled),
+      local_fallback_enabled: Boolean(preset?.local_fallback_enabled),
+      local_fallback_models: preset?.local_fallback_models || [],
       maxTokens,
       prices,
       timeoutSec: refTimeoutSec,
@@ -159,10 +189,20 @@ export async function runMoAPipeline({
     const files = extractFileBlocks(ref.text)
     ref.files = files
     if (files.length > 0) {
+      ref.syntaxWarning = (verifyFileSyntax(files) || []).map((w) => `${w.file}: ${w.error}`).join('; ')
       await writeCandidateWorkspace(cwd, i + 1, files)
     }
   }
 
+  // 5b. Test Execution Gate
+  await executeTestGateForCandidates({
+    cwd,
+    referenceOutputs,
+    preset,
+    options: { testGateEnabled, testCommand, execFn },
+    onProgress,
+  })
+
   // 6. Questionnaire synthesis branch
   if (needsQuestions) {
     if (typeof onProgress === 'function') {
@@ -188,11 +228,7 @@ export async function runMoAPipeline({
       qUsage = estimateTokenCost(primaryJudge, qFallbackUsage, prices)
     } catch (err) {
       console.warn('[dsh-moa] Questionnaire synthesis failed, proceeding with fallback questions:', err)
-      questionsContent = `### Clarification of Requirements: "${userPrompt}"\n\n` +
-        '1. **Architecture & Scope**: Single-file deliverable or multi-module project structure?\n' +
-        '2. **Design & Style**: Minimalist, dark mode, or clean neutral theme?\n' +
-        '3. **Functional Priorities**: Core MVP or comprehensive extended implementation?\n\n' +
-        '*Reply with your preferences (e.g. "1, 2") or proceed with defaults.*'
+      questionsContent = `### Clarification of Requirements: "${userPrompt}"\n\n1. **Architecture & Scope**: Single-file deliverable or multi-module project structure?\n2. **Design & Style**: Minimalist, dark mode, or clean neutral theme?\n3. **Functional Priorities**: Core MVP or comprehensive extended implementation?\n\n*Reply with your preferences (e.g. "1, 2") or proceed with defaults.*`
     }
 
     await cleanMoaWorkspaces(cwd)
@@ -344,12 +380,41 @@ export async function runMoAPipeline({
       }
     })
     await Promise.allSettled(r2Promises)
+    await executeTestGateForCandidates({
+      cwd,
+      referenceOutputs: successfulRefs,
+      preset,
+      options: { testGateEnabled, testCommand, execFn },
+      onProgress,
+    })
   }
 
+  // 7c. Feature 2: Multi-Judge Panel & Consensus Voting
+  let multiJudgeResult = null
+  if (preset?.multi_judge_enabled && successfulRefs.length > 1) {
+    multiJudgeResult = await executeMultiJudgePanel({
+      judges: (preset.judge_models && preset.judge_models.length > 0) ? preset.judge_models : [primaryJudge],
+      candidates: successfulRefs,
+      userPrompt,
+      callLlm,
+      strategy: preset.judge_voting_strategy || 'majority',
+      signal,
+      onProgress,
+    })
+  }
+
+  // 7d. Feature 3: Composite Block Merge Directive
+  const compositeMergeDirective = (preset?.composite_merge_enabled && successfulRefs.length > 1)
+    ? buildCompositeBlockDirectives(successfulRefs)
+    : null
+
   // 8. Synthesis phase via primary judge or fallback chain
   const synthesisPrompt = buildSynthesisPrompt(userPrompt, referenceOutputs, judgeCriteria, {
     curatorSynthesis: isCuratorSynthesis,
     blindEvaluation: isBlindEvaluation,
+    priorTurnBaseline,
+    compositeMergeDirective,
+    consensusReport: multiJudgeResult?.consensusReport,
   })
   let synthesizedText = ''
   let aggUsage = { inputTokens: 0, outputTokens: 0, totalTokens: 0, costUsd: 0 }
@@ -411,7 +476,10 @@ export async function runMoAPipeline({
   }
 
   // 9. Evaluate winner & promote files
-  const winningIndex = parseWinnerIndex(synthesizedText, 1, referenceOutputs.length)
+  const winningIndex = (multiJudgeResult?.consensus && multiJudgeResult?.winningCandidateIndex)
+    ? multiJudgeResult.winningCandidateIndex
+    : parseWinnerIndex(synthesizedText, 1, referenceOutputs.length)
+
   const recommendedAssembler = isCuratorSynthesis
     ? parseRecommendedAssembler(synthesizedText, winningIndex, referenceOutputs.length)
     : null
@@ -434,13 +502,32 @@ export async function runMoAPipeline({
   }
 
   const livePreview = await createPromotedPreview(liveCanvas, cwd, promotedFiles, referenceOutputs)
-
   const durationMs = Date.now() - startTime
   const usageSummary = summarizeMoAUsage(referenceOutputs, aggUsage)
   const winningRef = referenceOutputs[winningIndex - 1]
   const winnerModel = winningRef?.label || slotLabel(referenceModels[0])
   const finalAggLabel = slotLabel(chosenJudge)
 
+  // Feature 9: Benchmark & Post-Mortem PR report
+  let benchmarkReport = null
+  if (preset?.report_generation_enabled) {
+    benchmarkReport = generateMoABenchmarkReport({
+      runId,
+      timestamp: new Date().toISOString(),
+      preset: preset?.name || 'default',
+      prompt: userPrompt,
+      durationMs,
+      usage: usageSummary,
+      costUsd: usageSummary.totalCostUsd,
+      candidates: referenceOutputs,
+      winningIndex,
+      winningLabel: winnerModel,
+      consensus: multiJudgeResult,
+      testGate: referenceOutputs.find((r) => r.testResult)?.testResult || null,
+      isComposite: Boolean(preset?.composite_merge_enabled),
+    })
+  }
+
   try {
     recordMoaRun({
       id: runId,
@@ -452,6 +539,8 @@ export async function runMoAPipeline({
       winnerIndex: winningIndex,
       winnerModel,
       promotedFiles,
+      benchmarkReport,
+      consensus: multiJudgeResult,
       totalTokens: usageSummary.totalTokens,
       totalCostUsd: usageSummary.totalCostUsd,
       durationMs,
@@ -475,6 +564,8 @@ export async function runMoAPipeline({
     promotedFiles,
     runId,
     allowCandidateOverride,
+    benchmarkReport,
+    consensus: multiJudgeResult,
     ...(livePreview ? { liveCanvas: livePreview } : {}),
     usage: usageSummary,
     skippedFiles: collectedCtx?.skippedFiles || 0,
@@ -482,115 +573,3 @@ export async function runMoAPipeline({
     durationMs,
   }
 }
-
-/**
- * Streams a full MoA turn into chat with live aggregator tokens and progress feedback.
- */
-export async function* streamMoATurn({ targetPreset, userPrompt, messages, callLlm, cwd, prices, historyFilePath, liveCanvas }, options = {}) {
-  const signal = options?.signal
-  if (signal?.aborted) return
-
-  yield { type: 'block-start', index: 0, blockType: 'text' }
-  yield { type: 'text-delta', index: 0, text: '🧠 *Mixture of Agents started...*\n\n' }
-
-  // Async push-queue for zero-latency live delta streaming
-  const queue = []
-  let notify = null
-
-  const pushUpdate = (text) => {
-    queue.push({ type: 'text-delta', index: 0, text })
-    if (notify) {
-      notify()
-      notify = null
-    }
-  }
-
-  let done = false
-  const pipelinePromise = runMoAPipeline({
-    userPrompt,
-    messages,
-    preset: targetPreset,
-    callLlm,
-    cwd,
-    onProgress: pushUpdate,
-    onStreamDelta: (delta) => {
-      pushUpdate(delta)
-    },
-    prices,
-    historyFilePath,
-    liveCanvas,
-    signal,
-  })
-    .catch((err) => ({ error: err }))
-    .finally(() => {
-      done = true
-      if (notify) {
-        notify()
-        notify = null
-      }
-    })
-
-  const startTime = Date.now()
-  let lastYieldTime = Date.now()
-
-  try {
-    while (!done || queue.length > 0) {
-      if (signal?.aborted) {
-        await cleanMoaWorkspaces(cwd)
-        return
-      }
-
-      while (queue.length > 0) {
-        const item = queue.shift()
-        yield item
-        lastYieldTime = Date.now()
-      }
-
-      if (done) break
-
-      // Wait for next push item or max 2.5s heartbeat
-      await Promise.race([
-        new Promise((resolve) => { notify = resolve }),
-        new Promise((resolve) => { const t = setTimeout(resolve, 2500); t.unref?.(); }),
-      ])
-
-      const elapsedSec = Math.floor((Date.now() - startTime) / 1000)
-      if (!done && Date.now() - lastYieldTime >= 3000) {
-        yield { type: 'text-delta', index: 0, text: `⏳ *[${elapsedSec}s] Still processing...*\n` }
-        lastYieldTime = Date.now()
-      }
-    }
-
-    const result = await pipelinePromise
-    if (signal?.aborted) {
-      await cleanMoaWorkspaces(cwd)
-      return
-    }
-
-    if (result?.error) {
-      const errText = '\n\n⚠️ **Mixture of Agents error**: ' + (result.error?.message || String(result.error))
-      yield { type: 'text-delta', index: 0, text: errText }
-      yield { type: 'block-end', index: 0, block: { type: 'text', text: errText } }
-      yield { type: 'finish', reason: { kind: 'stop' } }
-      return
-    }
-
-    const formatted = formatMoAResponse({
-      moaResult: result,
-      presetName: targetPreset?.name || 'default',
-    })
-
-    yield { type: 'text-delta', index: 0, text: '\n---\n\n' + formatted }
-    yield { type: 'block-end', index: 0, block: { type: 'text', text: formatted } }
-    yield {
-      type: 'usage',
-      usage: {
-        inputTokens: result?.usage?.totalTokens || 0,
-        outputTokens: Math.round((formatted.length || 0) / 4),
-      },
-    }
-    yield { type: 'finish', reason: { kind: 'stop' } }
-  } finally {
-    // cleanup
-  }
-}