@goodandready/dsh-moa 0.2.61 → 0.2.63

package/lib/moa-runner.js

@@ -1,19 +1,3 @@
-import { logger } from './logger.js'
-import crypto from 'node:crypto'
-import { extractFileBlocks, collectProjectContext, formatProjectContext, isRefinementTask, writeCandidateWorkspace, promoteCandidateWorkspace, cleanMoaWorkspaces, verifyFileSyntax } from './file-workspace.js'
-import { estimateTokenCost, summarizeMoAUsage } from './pricing.js'
-import { recordMoaRunAsync, candidatesForHistory, isTestEnvironment } from './history.js'
-import { createPromotedPreview } from './live-canvas.js'
-import { slotLabel, isBroadPromptRequiringQuestions, buildQuestionSynthesisPrompt, buildSynthesisPrompt, buildPeerCritiquePrompt } from './moa-prompts.js'
-import { parseWinnerIndex, parseRecommendedAssembler } from './moa-parser.js'
-import { REFERENCE_SYSTEM_PROMPT, callWithTransientRetry, runReferencesParallel } from './moa-candidates.js'
-import { executeTestGateForCandidates } from './moa-test-gate.js'
-import { executeMultiJudgePanel, buildCompositeBlockDirectives, runMultiJudgePhase } from './moa-multi-judge.js'
-import { applyBudgetGuardrails, checkPhaseBudget, estimateRound2Cost, estimateDownstreamCost, calcSynthesisPromptTokens, createBudgetAbortPayload } from './moa-budget.js'
-import { extractPriorTurnBaseline, pruneMultiTurnMessages } from './moa-context.js'
-import { generateMoABenchmarkReport } from './moa-report.js'
-import { formatFileScoresDirective, parseFileScores } from './moa-workbench.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'
@@ -21,573 +5,4 @@ export { REFERENCE_SYSTEM_PROMPT, callWithTransientRetry, runReferencesParallel
 export { executeTestGateForCandidates, runCandidateTestGate } from './moa-test-gate.js'
 export { streamMoATurn, streamPromoteTurn } from './moa-stream.js'
 export { formatFileScoresDirective, parseFileScores } from './moa-workbench.js'
-/**
- * Executes the full Mixture of Agents pipeline.
- */
-export async function runMoAPipeline({
-  userPrompt,
-  messages = [],
-  preset,
-  callLlm,
-  cwd = process.cwd(), prices = {}, onProgress = null,
-  historyFilePath = null, liveCanvas = null, onStreamDelta = null, signal = null,
-  testGateEnabled = null, testCommand = null, execFn = null,
-  webServerPort = null, timeMachineEngine = null, timeMachineUrl = null,
-  force = isTestEnvironment() ? true : false, checkpoint = true, checkpointFn = null, runId: customRunId = null,
-}) {
-  const runId = (customRunId && String(customRunId).trim()) || crypto.randomUUID()
-  const promoteOpts = { webServerPort, timeMachineEngine, timeMachineUrl, force: Boolean(force), checkpoint, checkpointFn, signal }
-  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: '', model: '' }]
-  const primaryJudge = preset?.aggregator?.provider && preset?.aggregator?.model
-    ? preset.aggregator : { provider: '', model: '' }
-  const fallbackJudges = Array.isArray(preset?.aggregator_fallbacks) ? preset.aggregator_fallbacks : []
-  const judgesChain = [primaryJudge, ...fallbackJudges]
-  const refTemp = typeof preset?.reference_temperature === 'number' ? preset.reference_temperature : 0.6
-  const aggTemp = typeof preset?.aggregator_temperature === 'number' ? preset.aggregator_temperature : 0.4
-  const maxTokens = typeof preset?.max_tokens === 'number' ? preset.max_tokens : 4096
-  const judgeCriteria = preset?.judge_criteria || ''
-  let isFastMode = referenceModels.length === 1 && !preset?.curator_synthesis
-  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), isPeerCritiqueEnabled = Boolean(preset?.peer_critique_enabled)
-  const allowCandidateOverride = Boolean(preset?.allow_candidate_override)
-  // 1. Collect project context for refinement tasks
-  const collectedCtx = await collectProjectContext(cwd, 16000)
-  const isRefinement = Boolean(collectedCtx?.files?.length > 0 && isRefinementTask(userPrompt, collectedCtx.files))
-  let projectContext = ''
-  if (collectedCtx?.skippedFiles > 0 && typeof onProgress === 'function') {
-    const preview = collectedCtx.skippedList.slice(0, 3).join(', ')
-    const more = collectedCtx.skippedList.length > 3 ? ` (+${collectedCtx.skippedList.length - 3})` : ''
-    onProgress(`⚠️ *Workspace scan note: ${collectedCtx.skippedFiles} file(s) skipped (${preview}${more})*\n`)
-  }
-  if (isRefinement) {
-    if (typeof onProgress === 'function') {
-      onProgress('🔍 *Reading project files for refinement context...*\n')
-    }
-    projectContext = formatProjectContext(collectedCtx.files, {
-      skippedFiles: collectedCtx.skippedFiles,
-      skippedList: collectedCtx.skippedList,
-    })
-  }
-  // 2. Budget Guardrail check
-  const budgetCheck = applyBudgetGuardrails({
-    references: referenceModels,
-    enabled: preset?.budget_guard_enabled,
-    maxBudgetUsd: preset?.max_budget_usd,
-    action: preset?.budget_action || 'trim',
-    prices,
-    promptLength: (userPrompt?.length || 1000) + (judgeCriteria?.length || 0) + (projectContext ? projectContext.length : 0),
-    judgeCriteria,
-    projectContext,
-    messages,
-    maxTokens,
-    peerCritiqueEnabled: isPeerCritiqueEnabled,
-    aggregator: primaryJudge,
-    judgeModels: preset?.multi_judge_enabled ? preset?.judge_models : [],
-  })
-  if (!budgetCheck.allowed) {
-    return createBudgetAbortPayload({ reason: budgetCheck.reason, primaryJudge, referenceOutputs: [], currentSpentUsd: 0, preset, isRefinement, isFastMode, collectedCtx, startTime })
-  }
-  const effectiveReferences = budgetCheck.references
-  if (effectiveReferences.length === 1 && !preset?.curator_synthesis) {
-    isFastMode = true
-  }
-  if (budgetCheck.action === 'trim' && typeof onProgress === 'function') {
-    onProgress(`✂️ *${budgetCheck.reason}*\n`)
-  }
-  // 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 }]
-  // 4. Parallel fan-out to candidate models with temperature gradient & local fallback
-  const referenceOutputs = await runReferencesParallel(
-    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,
-      quorumEnabled: isQuorumEnabled,
-      gracePeriodSec,
-      maxRetries: candidateRetries,
-      signal,
-    },
-    callLlm,
-    onProgress
-  )
-  if (signal?.aborted) {
-    await cleanMoaWorkspaces(cwd, { runId })
-    return { error: new Error('Turn aborted') }
-  }
-  // Fail fast if all candidates failed
-  const successfulRefs = referenceOutputs.filter((r) => r.ok)
-  if (successfulRefs.length === 0) {
-    const reasons = referenceOutputs.map((r) => `${r.label}: ${r.error || 'unknown error'}`).join('; ')
-    const totalTokens = referenceOutputs.reduce((acc, r) => acc + (r.usage?.totalTokens || 0), 0)
-    const totalCostUsd = Number((referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0)).toFixed(5))
-    return {
-      kind: 'failure', content: `⚠️ All advisor models (${referenceOutputs.length}) failed: ${reasons}`,
-      aggregator: slotLabel(primaryJudge), references: referenceOutputs, presetName: preset?.name || 'default',
-      isRefinement, isFastMode, winningIndex: 0, winnerModel: 'none', promotedFiles: [],
-      usage: { totalTokens, totalCostUsd, candidates: referenceOutputs.map((r) => ({ label: r.label, usage: r.usage, costUsd: r.costUsd })), aggregator: { inputTokens: 0, outputTokens: 0, totalTokens: 0, costUsd: 0 } },
-      skippedFiles: collectedCtx?.skippedFiles || 0, skippedList: collectedCtx?.skippedList || [], durationMs: Date.now() - startTime,
-    }
-  }
-  // 5. Extract file blocks & write candidate workspaces
-  for (let i = 0; i < referenceOutputs.length; i++) {
-    const ref = referenceOutputs[i]
-    if (!ref.ok) continue
-    const files = extractFileBlocks(ref.text)
-    ref.files = files
-    if (files.length > 0) {
-      ref.syntaxWarning = (verifyFileSyntax(files) || []).map((w) => `${w.file}: ${w.error}`).join('; ')
-      await writeCandidateWorkspace(cwd, i + 1, files, { runId })
-    }
-  }
-  if (signal?.aborted) {
-    await cleanMoaWorkspaces(cwd, { runId })
-    return { error: new Error('Turn aborted') }
-  }
-  // 5b. Test Execution Gate
-  await executeTestGateForCandidates({
-    cwd,
-    referenceOutputs,
-    preset,
-    options: { testGateEnabled, testCommand, execFn, runId },
-    onProgress,
-  })
-  // 6. Questionnaire synthesis branch
-  if (needsQuestions) {
-    if (typeof onProgress === 'function') {
-      onProgress('📋 *Judge synthesizes the clarification questionnaire...*\n')
-    }
-    const questionPrompt = buildQuestionSynthesisPrompt(userPrompt, referenceOutputs)
-    let questionsContent = ''
-    let qUsage = { inputTokens: 0, outputTokens: 0, totalTokens: 0, costUsd: 0 }
-    try {
-      const qRes = await callWithTransientRetry(callLlm, {
-        provider: primaryJudge.provider,
-        model: primaryJudge.model,
-        messages: [{ role: 'user', content: questionPrompt }],
-        temperature: 0.3,
-        maxTokens: 2048,
-        timeoutMs: aggTimeoutSec * 1000,
-      }, 1)
-      questionsContent = typeof qRes === 'string' ? qRes : (qRes?.content || qRes?.text || '')
-      const qFallbackUsage = (typeof qRes === 'object' && qRes?.usage) ? qRes.usage : {
-        inputTokens: Math.max(1, Math.round(questionPrompt.length / 4)),
-        outputTokens: Math.max(1, Math.round(questionsContent.length / 4)),
-      }
-      qUsage = estimateTokenCost(primaryJudge, qFallbackUsage, prices)
-    } catch (err) {
-      logger.warn('[dsh-moa] Questionnaire synthesis failed, proceeding with fallback questions:', err)
-      questionsContent = `### Clarification of Requirements: "${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, { runId })
-    const totalTokens = referenceOutputs.reduce((acc, r) => acc + (r.usage?.totalTokens || 0), 0) + qUsage.totalTokens
-    const totalCostUsd = Number((referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0) + qUsage.costUsd).toFixed(5))
-    const durationMs = Date.now() - startTime
-    try {
-      await recordMoaRunAsync({
-        cwd,
-        prompt: userPrompt,
-        preset: preset?.name || 'default',
-        isRefinement,
-        candidates: candidatesForHistory(referenceOutputs),
-        aggregator: { provider: primaryJudge.provider, model: primaryJudge.model, usage: qUsage, costUsd: qUsage.costUsd },
-        winnerIndex: -1,
-        winnerModel: 'none',
-        promotedFiles: [],
-        totalTokens,
-        totalCostUsd,
-        durationMs,
-      }, historyFilePath || undefined)
-    } catch (histErr) {
-      logger.warn('[dsh-moa] Failed to record questionnaire run in history:', histErr?.message || histErr)
-    }
-    return {
-      kind: 'questions', content: questionsContent, aggregator: slotLabel(primaryJudge),
-      references: referenceOutputs, presetName: preset?.name || 'default', isRefinement, isFastMode,
-      winningIndex: 0, winnerModel: 'none', promotedFiles: [],
-      usage: { totalTokens, totalCostUsd, candidates: referenceOutputs.map((r) => ({ label: r.label, usage: r.usage, costUsd: r.costUsd })), aggregator: qUsage },
-      skippedFiles: collectedCtx?.skippedFiles || 0, skippedList: collectedCtx?.skippedList || [], durationMs,
-    }
-  }
-
-  // 7. Fast Mode (1 candidate, bypass judge)
-  if (isFastMode) {
-    if (signal?.aborted) {
-      await cleanMoaWorkspaces(cwd, { runId })
-      return { error: new Error('Turn aborted') }
-    }
-    const single = successfulRefs[0]
-    let promotedFiles = [], promoteErr = null
-    if (single.files?.length > 0) {
-      try {
-        promotedFiles = await promoteCandidateWorkspace(cwd, single.index, { ...promoteOpts, runId, keepMoa: allowCandidateOverride })
-      } catch (e) { promoteErr = e }
-    } else if (!allowCandidateOverride) {
-      await cleanMoaWorkspaces(cwd, { runId })
-    }
-    if (promoteErr) {
-      if (!allowCandidateOverride) await cleanMoaWorkspaces(cwd, { runId })
-      if (signal?.aborted || promoteErr?.message?.includes('aborted') || promoteErr?.name === 'AbortError' || promoteErr?.message === 'Turn aborted') {
-        return { error: new Error('Turn aborted') }
-      }
-      const totalTokens = referenceOutputs.reduce((acc, r) => acc + (r.usage?.totalTokens || 0), 0)
-      const totalCostUsd = Number((referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0)).toFixed(5))
-      const durationMs = Date.now() - startTime
-      try {
-        await recordMoaRunAsync({
-          id: runId, cwd, prompt: userPrompt, preset: preset?.name || 'default',
-          isRefinement, isFastMode: true, candidates: candidatesForHistory(referenceOutputs),
-          aggregator: null, winnerIndex: 0, winnerModel: 'none',
-          promotedFiles: [], totalTokens, totalCostUsd, durationMs,
-          error: promoteErr?.message || String(promoteErr),
-        }, historyFilePath || undefined)
-      } catch (histErr) {
-        logger.warn('[dsh-moa] Failed to record fast synthesis failure in history:', histErr?.message || histErr)
-      }
-      return {
-        kind: 'failure', error: promoteErr,
-        content: '🛑 Checkpoint or Promotion Failure: ' + (promoteErr?.message || String(promoteErr)),
-        aggregator: 'Fast Mode (Direct)', references: referenceOutputs,
-        presetName: preset?.name || 'default', isRefinement, isFastMode: true,
-        winningIndex: 0, winnerModel: 'none', promotedFiles: [], runId,
-        usage: {
-          totalTokens, totalCostUsd,
-          candidates: referenceOutputs.map((r) => ({ label: r.label, usage: r.usage, costUsd: r.costUsd })),
-          aggregator: { inputTokens: 0, outputTokens: 0, totalTokens: 0, costUsd: 0 },
-        },
-        skippedFiles: collectedCtx?.skippedFiles || 0,
-        skippedList: collectedCtx?.skippedList || [],
-        durationMs,
-      }
-    }
-
-    const livePreview = await createPromotedPreview(liveCanvas, cwd, promotedFiles, referenceOutputs, { winningIndex: single.index })
-    const totalTokens = single.usage?.totalTokens || 0
-    const totalCostUsd = Number((single.costUsd || 0).toFixed(5))
-    const durationMs = Date.now() - startTime
-
-    try {
-      await recordMoaRunAsync({
-        id: runId, cwd, prompt: userPrompt, preset: preset?.name || 'default',
-        isRefinement, isFastMode: true, candidates: candidatesForHistory(referenceOutputs),
-        aggregator: null, winnerIndex: single.index, winnerModel: single.label,
-        promotedFiles, totalTokens, totalCostUsd, durationMs,
-      }, historyFilePath || undefined)
-    } catch (histErr) {
-      logger.warn('[dsh-moa] Failed to record fast synthesis run in history:', histErr?.message || histErr)
-    }
-
-    return {
-      kind: 'synthesis', content: single.text, aggregator: 'Fast Mode (Direct)', references: referenceOutputs,
-      presetName: preset?.name || 'default', isRefinement, isFastMode: true, winningIndex: single.index,
-      winnerModel: single.label, promotedFiles, runId, ...(livePreview ? { liveCanvas: livePreview } : {}),
-      usage: { totalTokens, totalCostUsd, candidates: referenceOutputs.map((r) => ({ label: r.label, usage: r.usage, costUsd: r.costUsd })), aggregator: { inputTokens: 0, outputTokens: 0, totalTokens: 0, costUsd: 0 } },
-      skippedFiles: collectedCtx?.skippedFiles || 0, skippedList: collectedCtx?.skippedList || [], durationMs,
-    }
-  }
-
-  if (signal?.aborted) {
-    await cleanMoaWorkspaces(cwd, { runId })
-    return { error: new Error('Turn aborted') }
-  }
-
-  // 7b. Consilium Round 2 (Peer Critique) if enabled
-  let runRound2 = isPeerCritiqueEnabled && successfulRefs.length > 1
-  if (runRound2 && preset?.budget_guard_enabled && preset?.max_budget_usd > 0) {
-    const currentSpent = referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0)
-    const r2Next = estimateRound2Cost(successfulRefs, prices, maxTokens)
-    const downNext = estimateDownstreamCost(primaryJudge, preset?.judge_models, successfulRefs.length, prices, maxTokens)
-    const phaseCheck = checkPhaseBudget({ phaseName: 'Consilium Round 2', currentSpentUsd: currentSpent, nextPhaseCostUsd: r2Next + downNext, maxBudgetUsd: preset.max_budget_usd, action: preset.budget_action || 'trim', enabled: true })
-    if (!phaseCheck.allowed) {
-      if (phaseCheck.action === 'abort') {
-        await cleanMoaWorkspaces(cwd, { runId })
-        return createBudgetAbortPayload({ reason: phaseCheck.reason, primaryJudge, referenceOutputs, currentSpentUsd: currentSpent, preset, isRefinement, isFastMode, collectedCtx, startTime })
-      }
-      runRound2 = false
-      if (typeof onProgress === 'function') {
-        onProgress(`✂️ *Skipping Round 2 to stay within budget ($${preset.max_budget_usd.toFixed(4)})*\n`)
-      }
-    }
-  }
-
-  if (runRound2) {
-    if (typeof onProgress === 'function') {
-      onProgress('🤝 *Consilium Round 2: Candidates reviewing peer proposals in parallel...*\n')
-    }
-    const r2Promises = successfulRefs.map(async (cand) => {
-      const opponents = successfulRefs
-        .filter((c) => c.index !== cand.index)
-        .map((c) => ({
-          label: isBlindEvaluation ? `Candidate ${c.index}` : c.label,
-          role_persona: c.role_persona || c.slot?.role_persona,
-          text: c.text,
-        }))
-      const r2Prompt = buildPeerCritiquePrompt(userPrompt, cand.text, opponents, isBlindEvaluation)
-      try {
-        const slot = cand.slot || { provider: cand.provider, model: cand.model }
-        const res = await callWithTransientRetry(callLlm, {
-          provider: slot.provider,
-          model: slot.model,
-          messages: [{ role: 'user', content: r2Prompt }],
-          temperature: refTemp,
-          maxTokens,
-          timeoutMs: refTimeoutSec * 1000,
-          signal,
-        }, candidateRetries)
-        const refinedText = typeof res === 'string' ? res : (res?.content || res?.text || cand.text)
-        cand.text = refinedText
-        const r2Files = extractFileBlocks(refinedText)
-        if (r2Files.length > 0) {
-          cand.files = r2Files
-          cand.syntaxWarning = (verifyFileSyntax(r2Files) || []).map((w) => `${w.file}: ${w.error}`).join('; ')
-          await writeCandidateWorkspace(cwd, cand.index, r2Files, { runId })
-        }
-        if (typeof res === 'object' && res?.usage) {
-          cand.usage.inputTokens = (cand.usage.inputTokens || 0) + (res.usage.inputTokens || 0)
-          cand.usage.outputTokens = (cand.usage.outputTokens || 0) + (res.usage.outputTokens || 0)
-          cand.usage.totalTokens = (cand.usage.totalTokens || 0) + (res.usage.totalTokens || 0)
-          const extraCost = estimateTokenCost(slot, res.usage, prices)
-          cand.costUsd = Number(((cand.costUsd || 0) + extraCost.costUsd).toFixed(5))
-        }
-      } catch (r2Err) {
-        logger.warn(`[dsh-moa] Round 2 peer critique failed for candidate ${cand.index}:`, r2Err?.message || r2Err)
-      }
-    })
-    await Promise.allSettled(r2Promises)
-    await executeTestGateForCandidates({
-      cwd,
-      referenceOutputs: successfulRefs,
-      preset,
-      options: { testGateEnabled, testCommand, execFn, runId },
-      onProgress,
-    })
-  }
-
-  // 7c. Feature 2: Multi-Judge Panel & Consensus Voting
-  let multiJudgeResult = null
-  if (signal?.aborted) {
-    await cleanMoaWorkspaces(cwd, { runId })
-    return { error: new Error('Turn aborted') }
-  }
-
-  const currentRefSpent = referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0)
-  const panelPhase = await runMultiJudgePhase({
-    preset, successfulRefs, primaryJudge, userPrompt, judgeCriteria, prices, maxTokens, callLlm, signal, onProgress, currentSpent: currentRefSpent,
-  })
-  if (panelPhase.abort) {
-    await cleanMoaWorkspaces(cwd, { runId })
-    return createBudgetAbortPayload({ reason: panelPhase.reason, primaryJudge, referenceOutputs, currentSpentUsd: currentRefSpent, preset, isRefinement, isFastMode, collectedCtx, startTime })
-  }
-  multiJudgeResult = panelPhase.result
-
-  // 7d. Feature 3: Composite Block Merge Directive
-  const compositeMergeDirective = (preset?.composite_merge_enabled && successfulRefs.length > 1)
-    ? buildCompositeBlockDirectives(successfulRefs, { blind: Boolean(preset.blind_evaluation) })
-    : null
-
-  // 7e. Interactive Workbench File Scoring Directive
-  const fileScoringDirective = (preset?.interactive_workbench_enabled)
-    ? formatFileScoresDirective()
-    : null
-
-  // 8. Synthesis phase via primary judge or fallback chain
-  const synthesisPrompt = buildSynthesisPrompt(userPrompt, referenceOutputs, judgeCriteria, {
-    curatorSynthesis: isCuratorSynthesis,
-    blindEvaluation: isBlindEvaluation,
-    priorTurnBaseline,
-    compositeMergeDirective,
-    fileScoringDirective,
-    consensusReport: multiJudgeResult?.consensusReport,
-  })
-  const synthPromptTokens = Math.max(50, Math.ceil(synthesisPrompt.length / 4))
-
-  if (preset?.budget_guard_enabled && preset?.max_budget_usd > 0 && !isFastMode) {
-    const currentSpent = referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0) + (multiJudgeResult?.totalCostUsd || 0)
-    const downNext = estimateDownstreamCost(primaryJudge, [], successfulRefs.length, prices, maxTokens, synthPromptTokens)
-    const phaseCheck = checkPhaseBudget({ phaseName: 'Judge Synthesis', currentSpentUsd: currentSpent, nextPhaseCostUsd: downNext, maxBudgetUsd: preset.max_budget_usd, action: preset.budget_action || 'trim', enabled: true })
-    if (!phaseCheck.allowed && fallbackJudges.length === 0) {
-      await cleanMoaWorkspaces(cwd, { runId })
-      return createBudgetAbortPayload({ reason: phaseCheck.reason, primaryJudge, referenceOutputs, currentSpentUsd: currentSpent, preset, isRefinement, isFastMode, collectedCtx, startTime, multiJudgeResult })
-    }
-  }
-  let synthesizedText = ''
-  let aggUsage = { inputTokens: 0, outputTokens: 0, totalTokens: 0, costUsd: 0 }
-  let chosenJudge = primaryJudge
-  let judgeSuccess = false
-  let lastJudgeError = null
-
-  for (let jIdx = 0; jIdx < judgesChain.length; jIdx++) {
-    const currentJudge = judgesChain[jIdx]
-    const currentLabel = slotLabel(currentJudge)
-
-    if (preset?.budget_guard_enabled && preset?.max_budget_usd > 0 && !isFastMode) {
-      const currentSpent = referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0) + (multiJudgeResult?.totalCostUsd || 0)
-      const judgeCost = estimateDownstreamCost(currentJudge, [], successfulRefs.length, prices, maxTokens, synthPromptTokens)
-      const phaseCheck = checkPhaseBudget({ phaseName: `Judge Synthesis (${currentLabel})`, currentSpentUsd: currentSpent, nextPhaseCostUsd: judgeCost, maxBudgetUsd: preset.max_budget_usd, action: preset.budget_action || 'trim', enabled: true })
-      if (!phaseCheck.allowed) {
-        lastJudgeError = new Error(phaseCheck.reason)
-        logger.warn(`[dsh-moa] Judge ${currentLabel} budget check failed: ${phaseCheck.reason}`)
-        if (typeof onProgress === 'function') onProgress(`⚠️ *Judge ${currentLabel} skipped: exceeds budget limit.*\n`)
-        continue
-      }
-    }
-
-    if (typeof onProgress === 'function') {
-      const modeTitle = isCuratorSynthesis ? 'Curator' : 'Judge'
-      const fallbackBadge = jIdx > 0 ? ` (Fallback #${jIdx})` : ''
-      onProgress(`⚖️ *${modeTitle} (${currentLabel}${fallbackBadge}) synthesizing solution...*\n`)
-    }
-
-    try {
-      const aggRes = await callWithTransientRetry(callLlm, {
-        provider: currentJudge.provider,
-        model: currentJudge.model,
-        messages: [{ role: 'user', content: synthesisPrompt }],
-        temperature: aggTemp,
-        maxTokens,
-        timeoutMs: aggTimeoutSec * 1000,
-        signal,
-        onStreamDelta: (delta) => {
-          if (isStreamAggregator && typeof onStreamDelta === 'function') {
-            onStreamDelta(delta)
-          }
-        },
-      }, 0)
-
-      synthesizedText = typeof aggRes === 'string' ? aggRes : (aggRes?.content || aggRes?.text || '')
-
-      const aggFallbackUsage = (typeof aggRes === 'object' && aggRes?.usage) ? aggRes.usage : {
-        inputTokens: Math.max(1, Math.round(synthesisPrompt.length / 4)),
-        outputTokens: Math.max(1, Math.round(synthesizedText.length / 4)),
-      }
-      aggUsage = estimateTokenCost(currentJudge, aggFallbackUsage, prices)
-      chosenJudge = currentJudge
-      judgeSuccess = true
-      break
-    } catch (err) {
-      lastJudgeError = err
-      logger.warn(`[dsh-moa] Judge ${currentLabel} failed:`, err)
-      if (typeof onProgress === 'function') {
-        const nextJudge = judgesChain[jIdx + 1]
-        const nextHint = nextJudge ? ` Trying fallback ${slotLabel(nextJudge)}...` : ''
-        onProgress(`⚠️ *Judge ${currentLabel} failed: ${err?.message || err}.${nextHint}*\n`)
-      }
-    }
-  }
-
-  if (!judgeSuccess) {
-    if (preset?.budget_guard_enabled && preset?.max_budget_usd > 0 && !isFastMode && (lastJudgeError?.message?.includes('exceeds max budget') || lastJudgeError?.message?.includes('Budget Guardrail') || preset?.budget_action === 'abort')) {
-      await cleanMoaWorkspaces(cwd, { runId })
-      const currentSpent = referenceOutputs.reduce((acc, r) => acc + (r.costUsd || 0), 0) + (multiJudgeResult?.totalCostUsd || 0)
-      return createBudgetAbortPayload({ reason: lastJudgeError?.message || 'Budget exceeded before judge synthesis', primaryJudge, referenceOutputs, currentSpentUsd: currentSpent, preset, isRefinement, isFastMode, collectedCtx, startTime, multiJudgeResult })
-    }
-    const firstErr = lastJudgeError ? (lastJudgeError.message || String(lastJudgeError)) : 'unknown error'
-    synthesizedText = `⚠️ *[Aggregator error: ${firstErr}. Fallback candidate outputs:]*\n\n` +
-      successfulRefs.map((r, i) => `### Candidate ${i + 1} (${r.label})\n${r.text}`).join('\n\n')
-  }
-
-  if (signal?.aborted) {
-    await cleanMoaWorkspaces(cwd, { runId })
-    return { error: new Error('Turn aborted') }
-  }
-
-  // 9. Evaluate winner & promote files
-  const winningIndex = (multiJudgeResult?.consensus && multiJudgeResult?.winningCandidateIndex)
-    ? multiJudgeResult.winningCandidateIndex
-    : parseWinnerIndex(synthesizedText, 1, referenceOutputs.length)
-
-  const recommendedAssembler = isCuratorSynthesis
-    ? parseRecommendedAssembler(synthesizedText, winningIndex, referenceOutputs.length)
-    : null
-
-  // Check if aggregator synthesized unified file blocks directly
-  const synthesizedFiles = extractFileBlocks(synthesizedText)
-  let promotedFiles = []
-
-  const keepWorkspaces = allowCandidateOverride
-  if (signal?.aborted) {
-    if (!keepWorkspaces) await cleanMoaWorkspaces(cwd, { runId })
-    return { error: new Error('Turn aborted') }
-  }
-  try {
-    if (synthesizedFiles.length > 0) {
-      await writeCandidateWorkspace(cwd, 'curator-synthesis', synthesizedFiles, { runId })
-      promotedFiles = await promoteCandidateWorkspace(cwd, 'curator-synthesis', { ...promoteOpts, runId, keepMoa: keepWorkspaces })
-    } else if (referenceOutputs[winningIndex - 1]?.files?.length > 0) {
-      promotedFiles = await promoteCandidateWorkspace(cwd, winningIndex, { ...promoteOpts, runId, keepMoa: keepWorkspaces })
-    } else if (successfulRefs[0]?.files?.length > 0) {
-      const fallbackIdx = successfulRefs[0].index
-      promotedFiles = await promoteCandidateWorkspace(cwd, fallbackIdx, { ...promoteOpts, runId, keepMoa: keepWorkspaces })
-    } else if (!keepWorkspaces) {
-      await cleanMoaWorkspaces(cwd, { runId })
-    }
-  } catch (promoteErr) {
-    if (!keepWorkspaces) await cleanMoaWorkspaces(cwd, { runId })
-    if (signal?.aborted || promoteErr?.message === 'Turn aborted') return { error: new Error('Turn aborted') }
-    throw promoteErr
-  }
-
-  const livePreview = await createPromotedPreview(liveCanvas, cwd, promotedFiles, referenceOutputs, { winningIndex, synthesizedFiles })
-  const durationMs = Date.now() - startTime
-  const usageSummary = summarizeMoAUsage(referenceOutputs, aggUsage, multiJudgeResult?.totalUsage)
-  const winningRef = referenceOutputs[winningIndex - 1]
-  const winnerModel = winningRef?.label || slotLabel(effectiveReferences[0] || 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.filter((r) => r.testResult).map((r) => r.testResult),
-      isComposite: Boolean(preset?.composite_merge_enabled),
-    })
-  }
-
-  // Feature 10: Interactive Workbench candidate file quality scores
-  const fileScores = (preset?.interactive_workbench_enabled)
-    ? parseFileScores(synthesizedText, referenceOutputs)
-    : null
-
-  try {
-    await recordMoaRunAsync({
-      id: runId, cwd, prompt: userPrompt, preset: preset?.name || 'default', isRefinement,
-      candidates: candidatesForHistory(referenceOutputs),
-      aggregator: { provider: chosenJudge.provider, model: chosenJudge.model, usage: aggUsage, costUsd: aggUsage.costUsd },
-      winnerIndex: winningIndex, winnerModel, promotedFiles, benchmarkReport, consensus: multiJudgeResult,
-      fileScores, totalTokens: usageSummary.totalTokens, totalCostUsd: usageSummary.totalCostUsd, durationMs,
-    }, historyFilePath || undefined)
-  } catch (histErr) {
-    logger.warn('[dsh-moa] Failed to record run in history:', histErr)
-  }
-
-  return {
-    kind: 'synthesis', content: synthesizedText, aggregator: isFastMode ? 'Fast Mode (Direct)' : finalAggLabel,
-    references: referenceOutputs, presetName: preset?.name || 'default', isRefinement, isFastMode, isCuratorSynthesis,
-    recommendedAssembler, winningIndex, winnerModel, promotedFiles, runId, allowCandidateOverride,
-    benchmarkReport, consensus: multiJudgeResult, fileScores, ...(livePreview ? { liveCanvas: livePreview } : {}),
-    usage: usageSummary, skippedFiles: collectedCtx?.skippedFiles || 0, skippedList: collectedCtx?.skippedList || [], durationMs,
-  }
-}
+export { runMoAPipeline } from './pipeline/runner.js'