@tangle-network/agent-eval 0.43.2 → 0.44.1

package/dist/campaign/index.js

@@ -1,420 +1,33 @@
+import {
+  composeGate,
+  defaultProductionGate,
+  evolutionaryDriver,
+  gepaDriver,
+  heldOutGate,
+  openAutoPr,
+  runEval,
+  runImprovementLoop,
+  runOptimization,
+  surfaceHash
+} from "../chunk-H5BGRSN4.js";
 import {
   fsCampaignStorage,
   inMemoryCampaignStorage,
   runCampaign
 } from "../chunk-RXK7FXLV.js";
-import {
-  buildReflectionPrompt,
-  parseReflectionResponse,
-  runCanaries,
-  scoreRedTeamOutput
-} from "../chunk-N4SBKEPJ.js";
-import {
-  detectRewardHacking
-} from "../chunk-YV7J7X5N.js";
+import "../chunk-N4SBKEPJ.js";
+import "../chunk-YV7J7X5N.js";
 import "../chunk-WP7SY7AI.js";
 import "../chunk-GGE4NNQT.js";
-import {
-  callLlm
-} from "../chunk-VXNVVBZO.js";
+import "../chunk-VXNVVBZO.js";
 import "../chunk-PC4UYEBM.js";
 import "../chunk-QYJT52YW.js";
 import "../chunk-NSBPE2FW.js";
 
-// src/campaign/auto-pr.ts
-import { execSync } from "child_process";
-import { writeFileSync } from "fs";
-import { tmpdir } from "os";
-import { join } from "path";
-function openAutoPr(options) {
-  if (options.gate.decision !== "ship") {
-    return {
-      opened: false,
-      dryRun: false,
-      reason: `gate verdict was "${options.gate.decision}" \u2014 refusing to open PR`
-    };
-  }
-  const dryRun = options.dryRun ?? !process.env.GH_AUTO_PR_TOKEN;
-  const branch = options.branch ?? `auto/${options.result.manifestHash.slice(0, 12)}`;
-  const title = options.title ?? `auto: campaign ${options.result.manifestHash.slice(0, 8)} promoted by gate`;
-  const body = renderPrBody(options.result, options.gate, options.promotedDiff);
-  const bodyPath = join(tmpdir(), `auto-pr-body-${Date.now()}.md`);
-  writeFileSync(bodyPath, body);
-  if (dryRun) {
-    return {
-      opened: false,
-      dryRun: true,
-      reason: `dry-run (GH_AUTO_PR_TOKEN not set). Would create PR on ${options.ghOwner}/${options.ghRepo} branch ${branch}. Body at ${bodyPath}.`
-    };
-  }
-  const ghExec = options.ghExec ?? defaultGhExec;
-  const result = ghExec([
-    "pr",
-    "create",
-    "--repo",
-    `${options.ghOwner}/${options.ghRepo}`,
-    "--head",
-    branch,
-    "--title",
-    title,
-    "--body-file",
-    bodyPath
-  ]);
-  if (result.status !== 0) {
-    return {
-      opened: false,
-      dryRun: false,
-      reason: `gh pr create failed (exit ${result.status}): ${result.stderr.slice(0, 400)}`
-    };
-  }
-  const prUrl = result.stdout.trim();
-  return { opened: true, prUrl, dryRun: false, reason: "PR opened" };
-}
-function renderPrBody(result, gate, diff) {
-  const lines = [];
-  lines.push(`## Automated promotion by \`runImprovementLoop\``);
-  lines.push("");
-  lines.push(`**Manifest**: \`${result.manifestHash}\``);
-  lines.push(`**Seed**: ${result.seed}`);
-  lines.push(`**Duration**: ${Math.round(result.durationMs / 1e3)}s`);
-  lines.push(
-    `**Cells**: executed ${result.aggregates.cellsExecuted}, cached ${result.aggregates.cellsCached}, skipped ${result.aggregates.cellsSkipped}, failed ${result.aggregates.cellsFailed}`
-  );
-  lines.push(`**Total spend**: $${result.aggregates.totalCostUsd.toFixed(2)}`);
-  lines.push("");
-  lines.push(`### Gate verdict: \`${gate.decision}\``);
-  lines.push("");
-  for (const reason of gate.reasons) lines.push(`- ${reason}`);
-  if (gate.delta !== void 0) lines.push(`- delta: ${gate.delta.toFixed(3)}`);
-  lines.push("");
-  lines.push("### Contributing gates");
-  lines.push("");
-  lines.push("| gate | passed | detail |");
-  lines.push("|---|---|---|");
-  for (const c of gate.contributingGates) {
-    const detail = typeof c.detail === "object" ? JSON.stringify(c.detail).slice(0, 80) : String(c.detail).slice(0, 80);
-    lines.push(`| ${c.name} | ${c.passed ? "\u2713" : "\u2717"} | ${detail} |`);
-  }
-  lines.push("");
-  lines.push("### Promoted surface");
-  lines.push("");
-  lines.push("```diff");
-  lines.push(diff.slice(0, 8e3));
-  lines.push("```");
-  lines.push("");
-  lines.push("### By-judge aggregates");
-  lines.push("");
-  lines.push("| judge | mean | ci95 | n |");
-  lines.push("|---|---|---|---|");
-  for (const [name, agg] of Object.entries(result.aggregates.byJudge)) {
-    lines.push(
-      `| ${name} | ${agg.mean.toFixed(3)} | [${agg.ci95[0].toFixed(3)}, ${agg.ci95[1].toFixed(3)}] | ${agg.n} |`
-    );
-  }
-  return lines.join("\n");
-}
-function defaultGhExec(args) {
-  try {
-    const stdout = execSync(`gh ${args.map(quoteArg).join(" ")}`, {
-      env: { ...process.env, GH_TOKEN: process.env.GH_AUTO_PR_TOKEN ?? process.env.GH_TOKEN ?? "" },
-      stdio: ["ignore", "pipe", "pipe"]
-    }).toString("utf8");
-    return { stdout, stderr: "", status: 0 };
-  } catch (err) {
-    const e = err;
-    return {
-      stdout: e.stdout?.toString("utf8") ?? "",
-      stderr: e.stderr?.toString("utf8") ?? "",
-      status: e.status ?? 1
-    };
-  }
-}
-function quoteArg(arg) {
-  if (/^[a-zA-Z0-9_/\-:.@]+$/.test(arg)) return arg;
-  return `"${arg.replace(/"/g, '\\"')}"`;
-}
-
-// src/campaign/drivers/evolutionary.ts
-function evolutionaryDriver(opts) {
-  return {
-    kind: `evolutionary:${opts.mutator.kind}`,
-    async propose({ currentSurface, findings, populationSize, signal }) {
-      return opts.mutator.mutate({
-        findings: findings.length > 0 ? findings : opts.findings ?? [],
-        currentSurface,
-        populationSize,
-        signal
-      });
-    }
-  };
-}
-
-// src/campaign/drivers/gepa.ts
-var REFLECTION_SYSTEM = 'You are an expert prompt engineer. Output ONLY a JSON object of shape {"proposals":[{"label":string,"rationale":string,"payload":string}]} where each `payload` is the FULL improved surface text. No prose outside the JSON.';
-function gepaDriver(opts) {
-  const evidenceK = opts.evidenceK ?? 3;
-  return {
-    kind: "gepa",
-    async propose(ctx) {
-      const parent = typeof ctx.currentSurface === "string" ? ctx.currentSurface : JSON.stringify(ctx.currentSurface);
-      const { top, bottom, target } = buildEvidence(ctx, evidenceK, opts.target);
-      const userPrompt = buildReflectionPrompt({
-        target,
-        parentPayload: parent,
-        topTrials: top,
-        bottomTrials: bottom,
-        childCount: ctx.populationSize,
-        mutationPrimitives: opts.mutationPrimitives
-      });
-      const result = await callLlm(
-        {
-          model: opts.model,
-          messages: [
-            { role: "system", content: REFLECTION_SYSTEM },
-            { role: "user", content: userPrompt }
-          ],
-          jsonMode: true,
-          temperature: opts.temperature ?? 0.7,
-          maxTokens: opts.maxTokens ?? 6e3
-        },
-        opts.llm
-      );
-      const proposals = parseReflectionResponse(result.content, ctx.populationSize);
-      const out = [];
-      for (const proposal of proposals) {
-        const text = typeof proposal.payload === "string" ? proposal.payload.trim() : "";
-        if (text && text !== parent && !out.includes(text)) out.push(text);
-      }
-      return out;
-    }
-  };
-}
-function buildEvidence(ctx, evidenceK, baseTarget) {
-  const last = ctx.history.at(-1);
-  if (!last || last.candidates.length === 0) {
-    return { top: [], bottom: [], target: baseTarget };
-  }
-  const best = [...last.candidates].sort((a, b) => b.composite - a.composite)[0];
-  if (!best) return { top: [], bottom: [], target: baseTarget };
-  const byScore = [...best.scenarios].sort((a, b) => b.composite - a.composite);
-  const toTrace = (s) => ({
-    id: s.scenarioId,
-    score: s.composite
-  });
-  const top = byScore.slice(0, evidenceK).map(toTrace);
-  const bottom = byScore.slice(-evidenceK).reverse().map(toTrace);
-  const weakest = Object.entries(best.dimensions).sort((a, b) => a[1] - b[1]).slice(0, 3).map(([dim, value]) => `${dim} (${value.toFixed(2)})`);
-  const target = weakest.length > 0 ? `${baseTarget} \u2014 weakest dimensions: ${weakest.join(", ")}` : baseTarget;
-  return { top, bottom, target };
-}
-
-// src/campaign/gates/compose.ts
-function composeGate(...gates) {
-  if (gates.length === 0) {
-    throw new Error("composeGate requires at least one gate");
-  }
-  return {
-    name: `composed(${gates.map((g) => g.name).join(",")})`,
-    async decide(ctx) {
-      const results = [];
-      for (const gate of gates) {
-        const res = await gate.decide(ctx);
-        results.push({ gate, res });
-      }
-      const decisions = results.map((r) => r.res.decision);
-      const overall = decisions.every((d) => d === "ship") ? "ship" : decisions.includes("arch_ceiling") ? "arch_ceiling" : decisions.includes("model_ceiling") ? "model_ceiling" : decisions.includes("hold") ? "hold" : "need_more_work";
-      const contributing = results.flatMap(
-        (r) => r.res.contributingGates.length > 0 ? r.res.contributingGates : [{ name: r.gate.name, passed: r.res.decision === "ship", detail: r.res }]
-      );
-      const reasons = results.flatMap(
-        (r) => r.res.reasons.map((reason) => `[${r.gate.name}] ${reason}`)
-      );
-      return {
-        decision: overall,
-        reasons,
-        contributingGates: contributing,
-        delta: results[0]?.res.delta
-      };
-    }
-  };
-}
-
-// src/campaign/gates/default-production-gate.ts
-function defaultProductionGate(options) {
-  const deltaThreshold = options.deltaThreshold ?? 0.5;
-  const blockOnGaming = options.blockOnRewardHackingGaming ?? true;
-  return {
-    name: "defaultProductionGate",
-    async decide(ctx) {
-      const reasons = [];
-      const contributing = [];
-      const baselineComposite = meanComposite(
-        ctx.baselineArtifacts,
-        ctx.baselineJudgeScores ?? ctx.judgeScores,
-        options.holdoutScenarios
-      );
-      const candidateComposite = meanComposite(
-        ctx.candidateArtifacts,
-        ctx.judgeScores,
-        options.holdoutScenarios
-      );
-      const delta = candidateComposite - baselineComposite;
-      const heldoutPass = delta >= deltaThreshold;
-      contributing.push({
-        name: "heldout-delta",
-        passed: heldoutPass,
-        detail: { baselineComposite, candidateComposite, delta, deltaThreshold }
-      });
-      if (!heldoutPass) {
-        reasons.push(`heldout delta ${delta.toFixed(3)} < threshold ${deltaThreshold}`);
-      }
-      const budgetPass = options.budgetUsd === void 0 || ctx.cost.candidate + ctx.cost.baseline <= options.budgetUsd;
-      contributing.push({
-        name: "budget",
-        passed: budgetPass,
-        detail: {
-          candidateUsd: ctx.cost.candidate,
-          baselineUsd: ctx.cost.baseline,
-          budgetUsd: options.budgetUsd
-        }
-      });
-      if (!budgetPass) {
-        reasons.push(
-          `spend ${(ctx.cost.candidate + ctx.cost.baseline).toFixed(2)} > budget ${options.budgetUsd}`
-        );
-      }
-      const redTeamFindings = options.redTeamBattery ? probeRedTeam(ctx.candidateArtifacts, options.redTeamBattery) : { passed: true, findings: [] };
-      contributing.push({
-        name: "red-team",
-        passed: redTeamFindings.passed,
-        detail: {
-          failures: redTeamFindings.findings.length,
-          sample: redTeamFindings.findings.slice(0, 3)
-        }
-      });
-      if (!redTeamFindings.passed) {
-        reasons.push(`red-team probe failed (${redTeamFindings.findings.length} findings)`);
-      }
-      let rewardHackingReport = null;
-      if (options.recentRuns && options.recentRuns.length >= 10) {
-        rewardHackingReport = detectRewardHacking({ runs: options.recentRuns });
-      }
-      const gamingThreshold = 0.6;
-      const gamingFindings = (rewardHackingReport?.findings ?? []).filter(
-        (f) => f.severity >= gamingThreshold
-      );
-      const rewardHackingPass = !rewardHackingReport || !blockOnGaming || gamingFindings.length === 0 && rewardHackingReport.verdict !== "gaming";
-      contributing.push({
-        name: "reward-hacking",
-        passed: rewardHackingPass,
-        detail: { report: rewardHackingReport, gamingFindingCount: gamingFindings.length }
-      });
-      if (!rewardHackingPass) {
-        reasons.push(
-          `reward-hacking detector flagged ${gamingFindings.length} gaming-severity findings (verdict=${rewardHackingReport.verdict})`
-        );
-      }
-      let canaryReport = null;
-      if (options.recentRuns && options.recentRuns.length >= 10) {
-        canaryReport = runCanaries(options.recentRuns, {});
-      }
-      const errorAlerts = (canaryReport?.alerts ?? []).filter((a) => a.severity === "error");
-      const canaryPass = errorAlerts.length === 0;
-      contributing.push({
-        name: "canary",
-        passed: canaryPass,
-        detail: { totalAlerts: canaryReport?.alerts.length ?? 0, errorAlerts: errorAlerts.length }
-      });
-      if (!canaryPass) {
-        reasons.push(`canary error alerts: ${errorAlerts.length}`);
-      }
-      const allPassed = contributing.every((c) => c.passed);
-      const decision = allPassed ? "ship" : "hold";
-      return {
-        decision,
-        reasons: reasons.length > 0 ? reasons : ["all gates passed"],
-        contributingGates: contributing,
-        delta
-      };
-    }
-  };
-}
-function meanComposite(artifacts, judgeScoresByCell, scenarios) {
-  if (!artifacts || artifacts.size === 0) return 0;
-  const scenarioIds = new Set(scenarios.map((s) => s.id));
-  const composites = [];
-  for (const [cellId, scores] of judgeScoresByCell) {
-    const scenarioId = cellId.split(":")[0] ?? "";
-    if (!scenarioIds.has(scenarioId)) continue;
-    const cellComposites = Object.values(scores).map((s) => s.composite);
-    if (cellComposites.length === 0) continue;
-    composites.push(cellComposites.reduce((a, b) => a + b, 0) / cellComposites.length);
-  }
-  if (composites.length === 0) return 0;
-  return composites.reduce((a, b) => a + b, 0) / composites.length;
-}
-function probeRedTeam(artifacts, battery) {
-  const findings = [];
-  for (const [_cellId, artifact] of artifacts) {
-    const text = extractText(artifact);
-    if (text === void 0) continue;
-    for (const rtCase of battery) {
-      const finding = scoreRedTeamOutput(text, [], rtCase);
-      if (!finding.passed) {
-        findings.push({ scenarioId: rtCase.id, reason: finding.reason ?? "red-team probe failed" });
-      }
-    }
-  }
-  return { passed: findings.length === 0, findings };
-}
-function extractText(artifact) {
-  if (typeof artifact === "string") return artifact;
-  if (artifact && typeof artifact === "object") {
-    const rec = artifact;
-    if (typeof rec.text === "string") return rec.text;
-    if (typeof rec.output === "string") return rec.output;
-    if (typeof rec.content === "string") return rec.content;
-  }
-  return void 0;
-}
-
-// src/campaign/gates/heldout-gate.ts
-function heldOutGate(options) {
-  const deltaThreshold = options.deltaThreshold ?? 0.5;
-  return {
-    name: "heldOutGate",
-    async decide(ctx) {
-      const scenarioIds = new Set(options.scenarios.map((s) => s.id));
-      const baseline = meanForScenarios(ctx.baselineJudgeScores ?? ctx.judgeScores, scenarioIds);
-      const candidate = meanForScenarios(ctx.judgeScores, scenarioIds);
-      const delta = candidate - baseline;
-      const passed = delta >= deltaThreshold;
-      return {
-        decision: passed ? "ship" : "hold",
-        reasons: passed ? [`held-out delta ${delta.toFixed(3)} \u2265 ${deltaThreshold}`] : [`held-out delta ${delta.toFixed(3)} < ${deltaThreshold}`],
-        contributingGates: [
-          { name: "heldOutGate", passed, detail: { baseline, candidate, delta, deltaThreshold } }
-        ],
-        delta
-      };
-    }
-  };
-}
-function meanForScenarios(judgeScoresByCell, scenarioIds) {
-  const composites = [];
-  for (const [cellId, scores] of judgeScoresByCell) {
-    const scenarioId = cellId.split(":")[0] ?? "";
-    if (!scenarioIds.has(scenarioId)) continue;
-    const vals = Object.values(scores).map((s) => s.composite);
-    if (vals.length > 0) composites.push(vals.reduce((a, b) => a + b, 0) / vals.length);
-  }
-  return composites.length === 0 ? 0 : composites.reduce((a, b) => a + b, 0) / composites.length;
-}
-
 // src/campaign/labeled-store/fs-adapter.ts
 import { createHash } from "crypto";
-import { existsSync, mkdirSync, readFileSync, writeFileSync as writeFileSync2 } from "fs";
-import { join as join2 } from "path";
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from "fs";
+import { join } from "path";
 var LabeledScenarioStoreError = class extends Error {
   constructor(code, message) {
     super(message);
@@ -561,7 +174,7 @@ var FsLabeledScenarioStore = class {
     };
   }
   pathForSource(source) {
-    return join2(this.options.root, `${source}.jsonl`);
+    return join(this.options.root, `${source}.jsonl`);
   }
 };
 var ALL_SOURCES = [
@@ -600,239 +213,16 @@ function sha256(input) {
 function appendLine(path, line) {
   if (existsSync(path)) {
     const existing = readFileSync(path, "utf8");
-    writeFileSync2(path, existing + line);
+    writeFileSync(path, existing + line);
   } else {
-    writeFileSync2(path, line);
-  }
-}
-
-// src/campaign/presets/run-eval.ts
-async function runEval(opts) {
-  return runCampaign(opts);
-}
-
-// src/campaign/presets/run-optimization.ts
-import { createHash as createHash2 } from "crypto";
-async function runOptimization(opts) {
-  const promoteTopK = opts.promoteTopK ?? 2;
-  const baselineCampaign = await runCampaign({
-    ...opts,
-    dispatch: (scenario, ctx) => opts.dispatchWithSurface(opts.baselineSurface, scenario, ctx),
-    runDir: `${opts.runDir}/baseline`
-  });
-  const generations = [];
-  const history = [];
-  let currentSurfaces = [opts.baselineSurface];
-  let winnerSurface = opts.baselineSurface;
-  let winnerSurfaceHash = surfaceHash(opts.baselineSurface);
-  let winnerComposite = meanComposite2(baselineCampaign);
-  for (let gen = 0; gen < opts.maxGenerations; gen++) {
-    if (opts.driver.decide?.({ history }).stop) break;
-    const candidates = await opts.driver.propose({
-      currentSurface: currentSurfaces[0] ?? opts.baselineSurface,
-      history,
-      findings: [],
-      populationSize: opts.populationSize,
-      generation: gen,
-      signal: new AbortController().signal,
-      report: opts.report,
-      dataset: opts.labeledStore && opts.labeledStore !== "off" ? opts.labeledStore : void 0,
-      maxImprovementShots: opts.maxImprovementShots
-    });
-    const surfaceResults = [];
-    for (let i = 0; i < candidates.length; i++) {
-      const surface = candidates[i];
-      const hash = surfaceHash(surface);
-      const campaign = await runCampaign({
-        ...opts,
-        dispatch: (scenario, ctx) => opts.dispatchWithSurface(surface, scenario, ctx),
-        runDir: `${opts.runDir}/gen-${gen}/candidate-${i}`
-      });
-      const composite = meanComposite2(campaign);
-      surfaceResults.push({ surfaceHash: hash, surface, campaign, composite });
-    }
-    surfaceResults.sort((a, b) => b.composite - a.composite);
-    const promoted = surfaceResults.slice(0, promoteTopK);
-    currentSurfaces = promoted.map((p) => p.surface);
-    const top = surfaceResults[0];
-    if (top && top.composite > winnerComposite) {
-      winnerSurface = top.surface;
-      winnerSurfaceHash = top.surfaceHash;
-      winnerComposite = top.composite;
-    }
-    const record = {
-      generationIndex: gen,
-      candidates: surfaceResults.map((s) => {
-        const breakdown = candidateBreakdown(s.campaign);
-        return {
-          surfaceHash: s.surfaceHash,
-          composite: s.composite,
-          ci95: [s.composite, s.composite],
-          dimensions: breakdown.dimensions,
-          scenarios: breakdown.scenarios
-        };
-      }),
-      promoted: promoted.map((p) => p.surfaceHash)
-    };
-    history.push(record);
-    generations.push({
-      record,
-      surfaces: surfaceResults.map((s) => ({
-        surfaceHash: s.surfaceHash,
-        surface: s.surface,
-        campaign: s.campaign
-      }))
-    });
-  }
-  return {
-    generations,
-    winnerSurface,
-    winnerSurfaceHash,
-    baselineCampaign
-  };
-}
-function surfaceHash(surface) {
-  const material = typeof surface === "string" ? surface : JSON.stringify({
-    kind: surface.kind,
-    worktreeRef: surface.worktreeRef,
-    baseRef: surface.baseRef ?? null
-  });
-  return createHash2("sha256").update(material).digest("hex").slice(0, 16);
-}
-function meanComposite2(campaign) {
-  const composites = [];
-  for (const cell of campaign.cells) {
-    const cellComposites = Object.values(cell.judgeScores).map((s) => s.composite);
-    if (cellComposites.length > 0) {
-      composites.push(cellComposites.reduce((a, b) => a + b, 0) / cellComposites.length);
-    }
-  }
-  return composites.length === 0 ? 0 : composites.reduce((a, b) => a + b, 0) / composites.length;
-}
-function candidateBreakdown(campaign) {
-  const dimSums = {};
-  const dimCounts = {};
-  const byScenario = /* @__PURE__ */ new Map();
-  for (const cell of campaign.cells) {
-    const judgeScores = Object.values(cell.judgeScores);
-    if (judgeScores.length === 0) continue;
-    const cellComposite = judgeScores.reduce((a, s) => a + s.composite, 0) / judgeScores.length;
-    const arr = byScenario.get(cell.scenarioId) ?? [];
-    arr.push(cellComposite);
-    byScenario.set(cell.scenarioId, arr);
-    for (const score of judgeScores) {
-      for (const [key, value] of Object.entries(score.dimensions)) {
-        dimSums[key] = (dimSums[key] ?? 0) + value;
-        dimCounts[key] = (dimCounts[key] ?? 0) + 1;
-      }
-    }
-  }
-  const dimensions = {};
-  for (const key of Object.keys(dimSums)) {
-    const count = dimCounts[key] ?? 0;
-    dimensions[key] = count > 0 ? (dimSums[key] ?? 0) / count : 0;
-  }
-  const scenarios = [...byScenario.entries()].map(([scenarioId, comps]) => ({
-    scenarioId,
-    composite: comps.reduce((a, b) => a + b, 0) / comps.length
-  }));
-  return { dimensions, scenarios };
-}
-
-// src/campaign/presets/run-improvement-loop.ts
-async function runImprovementLoop(opts) {
-  if (opts.autoOnPromote === "config") {
-    throw new Error(
-      "runImprovementLoop: autoOnPromote='config' is deferred to Pass B (requires shadow deploy + rollback + ensemble judges). Use 'pr' or 'none' in v0.40."
-    );
-  }
-  if (opts.tracing === "off" && opts.driver) {
-    throw new Error(
-      "runImprovementLoop: tracing='off' is forbidden when a driver is wired. The improvement loop without traces is unattributable; candidate surfaces cannot be cited back to spans and the optimization dataset goes unfed."
-    );
-  }
-  if (opts.autoOnPromote === "pr" && (!opts.ghOwner || !opts.ghRepo)) {
-    throw new Error("runImprovementLoop: autoOnPromote='pr' requires ghOwner + ghRepo.");
-  }
-  const optimization = await runOptimization(opts);
-  const { runCampaign: runCampaign2 } = await import("../run-campaign-GNDO66B4.js");
-  const baselineOnHoldout = await runCampaign2({
-    ...opts,
-    scenarios: opts.holdoutScenarios,
-    dispatch: (scenario, ctx) => opts.dispatchWithSurface(opts.baselineSurface, scenario, ctx),
-    runDir: `${opts.runDir}/holdout-baseline`
-  });
-  const winnerOnHoldout = await runCampaign2({
-    ...opts,
-    scenarios: opts.holdoutScenarios,
-    dispatch: (scenario, ctx) => opts.dispatchWithSurface(optimization.winnerSurface, scenario, ctx),
-    runDir: `${opts.runDir}/holdout-winner`
-  });
-  const candidateArtifacts = /* @__PURE__ */ new Map();
-  const baselineArtifacts = /* @__PURE__ */ new Map();
-  const judgeScores = /* @__PURE__ */ new Map();
-  const baselineJudgeScores = /* @__PURE__ */ new Map();
-  for (const cell of winnerOnHoldout.cells) {
-    candidateArtifacts.set(cell.cellId, cell.artifact);
-    judgeScores.set(cell.cellId, cell.judgeScores);
-  }
-  for (const cell of baselineOnHoldout.cells) {
-    baselineArtifacts.set(cell.cellId, cell.artifact);
-    baselineJudgeScores.set(cell.cellId, cell.judgeScores);
-  }
-  const gateResult = await opts.gate.decide({
-    candidateArtifacts,
-    baselineArtifacts,
-    judgeScores,
-    baselineJudgeScores,
-    scenarios: opts.holdoutScenarios,
-    cost: {
-      candidate: winnerOnHoldout.aggregates.totalCostUsd,
-      baseline: baselineOnHoldout.aggregates.totalCostUsd
-    },
-    signal: new AbortController().signal
-  });
-  let prResult;
-  if (opts.autoOnPromote === "pr" && gateResult.decision === "ship") {
-    const render = opts.renderPromotedDiff ?? defaultRenderDiff;
-    const promotedDiff = render(optimization.winnerSurface, opts.baselineSurface);
-    prResult = openAutoPr({
-      result: winnerOnHoldout,
-      gate: gateResult,
-      promotedDiff,
-      ghOwner: opts.ghOwner,
-      ghRepo: opts.ghRepo
-    });
-  }
-  return {
-    ...optimization,
-    baselineOnHoldout,
-    winnerOnHoldout,
-    gateResult,
-    prResult
-  };
-}
-function defaultRenderDiff(winnerSurface, baselineSurface) {
-  if (typeof winnerSurface !== "string" || typeof baselineSurface !== "string") {
-    const fmt = (s) => typeof s === "string" ? "(prompt surface)" : `worktree=${s.worktreeRef}${s.baseRef ? ` base=${s.baseRef}` : ""}${s.summary ? `
-${s.summary}` : ""}`;
-    return `--- baseline
-${fmt(baselineSurface)}
-+++ winner
-${fmt(winnerSurface)}`;
+    writeFileSync(path, line);
   }
-  const lines = [];
-  lines.push("--- baseline");
-  lines.push("+++ winner");
-  for (const l of baselineSurface.split("\n")) lines.push(`- ${l}`);
-  for (const l of winnerSurface.split("\n")) lines.push(`+ ${l}`);
-  return lines.join("\n");
 }
 
 // src/campaign/worktree/index.ts
 import { execFileSync } from "child_process";
 import { existsSync as existsSync2 } from "fs";
-import { basename, isAbsolute, join as join3 } from "path";
+import { basename, isAbsolute, join as join2 } from "path";
 var WorktreeAdapterError = class extends Error {
   constructor(message, cause) {
     super(message);
@@ -854,13 +244,13 @@ function slug(label) {
 }
 function gitWorktreeAdapter(opts) {
   const git = opts.git ?? defaultGit;
-  const worktreeDir = opts.worktreeDir ?? join3(opts.repoRoot, ".worktrees");
+  const worktreeDir = opts.worktreeDir ?? join2(opts.repoRoot, ".worktrees");
   const branchPrefix = opts.branchPrefix ?? "improve";
   return {
     async create({ baseRef, label }) {
       const id = `${slug(label)}-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 6)}`;
       const branch = `${branchPrefix}/${id}`;
-      const path = join3(worktreeDir, id);
+      const path = join2(worktreeDir, id);
       git(["worktree", "add", "-b", branch, path, baseRef], opts.repoRoot);
       return { path, branch, baseRef };
     },
@@ -885,7 +275,7 @@ function gitWorktreeAdapter(opts) {
 }
 function resolveWorktreePath(surface, worktreeDir) {
   if (isAbsolute(surface.worktreeRef) && existsSync2(surface.worktreeRef)) return surface.worktreeRef;
-  if (worktreeDir) return join3(worktreeDir, basename(surface.worktreeRef));
+  if (worktreeDir) return join2(worktreeDir, basename(surface.worktreeRef));
   return surface.worktreeRef;
 }
 export {