@tangle-network/agent-eval 0.122.1 → 0.122.2

package/dist/{chunk-PICTDURQ.js → chunk-OZFTLMGS.js}

@@ -1,7 +1,7 @@
 import {
   fsCampaignStorage,
   runCampaign
-} from "./chunk-XDIRG3TO.js";
+} from "./chunk-N3QPYIVG.js";
 import {
   __export
 } from "./chunk-PZ5AY32C.js";
@@ -763,4 +763,4 @@ export {
   retrievalMetricsAtCutoff,
   benchmarks_exports
 };
-//# sourceMappingURL=chunk-PICTDURQ.js.map
+//# sourceMappingURL=chunk-OZFTLMGS.js.map

package/dist/{chunk-VZ6VKOJT.js → chunk-U4R2AGY5.js}

@@ -1,6 +1,7 @@
 import {
   assertCampaignSplitIdentity,
   campaignCoverage,
+  campaignSplitDigest,
   canonicalJson,
   contentHash,
   createRunCostLedger,
@@ -9,11 +10,11 @@ import {
   resolveRunDir,
   runCampaign,
   summarizeAgentReceiptIntegrity
-} from "./chunk-XDIRG3TO.js";
+} from "./chunk-N3QPYIVG.js";
 import {
   clamp01,
   validatePolicyEditCandidateRecord
-} from "./chunk-S5TT5R3L.js";
+} from "./chunk-5CVUPHJ4.js";
 import {
   detectRewardHacking
 } from "./chunk-ARU2PZFM.js";
@@ -30,7 +31,7 @@ import {
 } from "./chunk-PJQFMIOX.js";
 import {
   CostLedger
-} from "./chunk-JHOJHHU7.js";
+} from "./chunk-ZMXDQ4K7.js";
 import {
   DEFAULT_REDACTION_RULES
 } from "./chunk-GGE4NNQT.js";
@@ -2661,23 +2662,32 @@ async function runOptimization(opts) {
       "runOptimization: promoteTopK must be 1 because the loop has one global incumbent"
     );
   }
-  const baselineCampaign = await runCampaign({
+  const requireJudgeScore = (opts.judges?.length ?? 0) > 0;
+  const reps = opts.reps ?? 1;
+  const premeasuredBaseline = opts.premeasuredBaseline;
+  const baselineCampaign = premeasuredBaseline ? validatedPremeasuredBaseline({
+    input: premeasuredBaseline,
+    baselineSurface: opts.baselineSurface,
+    scenarios: opts.scenarios,
+    reps,
+    seed: opts.seed ?? 42
+  }) : await runCampaign({
     ...opts,
     costLedger,
     costPhase: "search.baseline",
     dispatch: (scenario, ctx) => opts.dispatchWithSurface(opts.baselineSurface, scenario, ctx),
     runDir: `${opts.runDir}/baseline`
   });
-  const requireJudgeScore = (opts.judges?.length ?? 0) > 0;
   const baselineCoverage = campaignCoverage(
     baselineCampaign.cells,
     opts.scenarios,
-    opts.reps ?? 1,
+    reps,
     requireJudgeScore
   );
   if (!baselineCoverage.complete) {
+    const label = opts.premeasuredBaseline ? "premeasured baseline" : "baseline";
     throw new Error(
-      `runOptimization: baseline is incomplete (${baselineCoverage.scorableCellIds.length}/${baselineCoverage.expectedCellIds.length} designed cells scorable) \u2014 ${formatCoverageFailures(baselineCoverage)}. Refusing to optimize against an incomplete incumbent.`
+      `runOptimization: ${label} is incomplete (${baselineCoverage.scorableCellIds.length}/${baselineCoverage.expectedCellIds.length} designed cells scorable) \u2014 ${formatCoverageFailures(baselineCoverage)}. Refusing to optimize against an incomplete incumbent.`
     );
   }
   const generations = [];
@@ -2701,7 +2711,7 @@ async function runOptimization(opts) {
   if (opts.analyzeGeneration && opts.maxGenerations > 0 && baselineCampaign.cells.length > 0) {
     const fresh = await opts.analyzeGeneration({
       generation: -1,
-      runDir: `${opts.runDir}/baseline`,
+      runDir: baselineCampaign.runDir,
       candidates: [
         { surfaceHash: winnerSurfaceHash, campaign: baselineCampaign, composite: winnerComposite }
       ],
@@ -2852,6 +2862,38 @@ async function runOptimization(opts) {
     cost: costLedger.summary()
   };
 }
+function validatedPremeasuredBaseline(args) {
+  const { input } = args;
+  if (input.surfaceHash !== surfaceHash(args.baselineSurface)) {
+    throw new Error(
+      "runOptimization: premeasured baseline surface hash does not match baselineSurface"
+    );
+  }
+  const campaign = input.campaign;
+  if (campaign.reps !== args.reps) {
+    throw new Error(
+      `runOptimization: premeasured baseline reps ${campaign.reps} do not match requested reps ${args.reps}`
+    );
+  }
+  if (campaign.seed !== args.seed) {
+    throw new Error(
+      `runOptimization: premeasured baseline seed ${campaign.seed} does not match requested seed ${args.seed}`
+    );
+  }
+  try {
+    assertCampaignSplitIdentity(campaign.scenarios, campaign.reps, campaign.splitDigest);
+  } catch (error) {
+    throw new Error(
+      `runOptimization: premeasured baseline has an invalid retained split identity \u2014 ${error instanceof Error ? error.message : String(error)}`
+    );
+  }
+  if (campaign.splitDigest !== campaignSplitDigest(args.scenarios, args.reps)) {
+    throw new Error(
+      "runOptimization: premeasured baseline split does not match the requested scenarios"
+    );
+  }
+  return campaign;
+}
 function toParetoParent(surface, hash, campaign, generation, label, rationale) {
   const objectives = {};
   for (const { scenarioId, composite } of campaignBreakdown(campaign).scenarios) {
@@ -2946,7 +2988,7 @@ async function runImprovementLoop(opts) {
   const dispatchTimeoutMs = opts.dispatchTimeoutMs ?? DEFAULT_DISPATCH_TIMEOUT_MS;
   const optimization = await runOptimization({ ...opts, dispatchTimeoutMs, costLedger });
   const winnerIsBaseline = optimization.winnerSurfaceHash === surfaceHash(opts.baselineSurface);
-  const { runCampaign: runCampaign2 } = await import("./run-campaign-HNFPJET4.js");
+  const { runCampaign: runCampaign2 } = await import("./run-campaign-2YJ3Z6JS.js");
   const baselineOnHoldout = await runCampaign2({
     ...opts,
     costLedger,
@@ -3738,4 +3780,4 @@ export {
   provenanceSpansPath,
   emitLoopProvenance
 };
-//# sourceMappingURL=chunk-VZ6VKOJT.js.map
+//# sourceMappingURL=chunk-U4R2AGY5.js.map