npm - @tangle-network/agent-eval - Versions diffs - 0.134.1 → 0.135.0 - Mend

@tangle-network/agent-eval 0.134.1 → 0.135.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (50) hide show

package/dist/cli.js CHANGED Viewed

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
 import { o as runRolloutReleaseCli } from "./hf-dataset-DBJXXoY1.js";
-import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-BPqlDBWK.js";
+import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-Dc_lsOYd.js";
 import { writeFileSync } from "node:fs";
 //#region src/cli-config.ts
 function resolveCliLlmConfig(env = process.env) {

package/dist/contract/index.js CHANGED Viewed

@@ -1,11 +1,12 @@
 import { s as ValidationError } from "../errors-8YnH8WlF.js";
-import { F as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-IjYs7T8l.js";
-import { g as isModelPriced, i as CostLedger, m as estimateCost } from "../cost-ledger-BrJxbrMy.js";
+import { F as createChatClient, t as buildDefaultAnalystRegistry } from "../default-registry-CFUZyNeZ.js";
+import { i as estimateCost, o as isModelPriced } from "../metrics-C9YY1OcL.js";
+import { i as CostLedger } from "../cost-ledger-ZAa_P4r0.js";
 import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
 import { b as classifyOtlpSpanRole, x as isOtlpModelCall } from "../tools-BmuN627J.js";
 import { l as makeProposalFinding } from "../proposal-findings-DCawte-y.js";
 import { r as mapConcurrentRange } from "../concurrency-MUjT7VjM.js";
-import { B as surfaceContentHash, Ct as llmJudge, M as compareOptimizationMethods, O as composeGate, Q as inMemoryCampaignStorage, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as createRunCostLedger, Y as runCampaign, Z as fsCampaignStorage, _ as buildEvidenceVector, a as emitLoopProvenance, b as powerPreflight, ct as campaignSplitDigest, d as runImprovementLoop, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, g as gepaOptimizationMethod, j as assertOptimizationResult, k as externalTextOptimizationMethod, mt as runReferenceEquivalenceJudge, o as loopProvenanceArgsFromResult, p as runEval, pt as createReferenceEquivalenceJudge, rt as resolveRunDir, t as skillOptOptimizationMethod, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-BY6vKLJB.js";
+import { B as surfaceContentHash, Ct as llmJudge, M as compareOptimizationMethods, O as composeGate, Q as inMemoryCampaignStorage, S as defaultProductionGate, T as heldoutSignificance, V as surfaceHash, X as createRunCostLedger, Y as runCampaign, Z as fsCampaignStorage, _ as buildEvidenceVector, a as emitLoopProvenance, b as powerPreflight, ct as campaignSplitDigest, d as runImprovementLoop, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, g as gepaOptimizationMethod, j as assertOptimizationResult, k as externalTextOptimizationMethod, mt as runReferenceEquivalenceJudge, o as loopProvenanceArgsFromResult, p as runEval, pt as createReferenceEquivalenceJudge, rt as resolveRunDir, t as skillOptOptimizationMethod, v as paretoPolicy, x as heldOutGate, y as paretoSignificanceGate } from "../skillopt-optimization-method-Cl4XPkLC.js";
 import { C as pairedBootstrap } from "../statistics-RwRNu2__.js";
 import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-BIwU2wdV.js";
 import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-2j25whHw.js";