@ls-stack/agent-eval 0.27.0 → 0.28.0

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.
package/dist/index.mjs CHANGED
@@ -1,4 +1,4 @@
1
- import { $ as llmCallsConfigSchema, $t as traceSpanKindSchema, A as extractApiCalls, An as getCurrentScope, At as evalChartTypeSchema, B as runSummarySchema, Bn as setEvalOutput, Bt as cacheRecordingOpSchema, Cn as evalExpect, Ct as evalChartAggregateSchema, D as sseEnvelopeSchema, Dt as evalChartConfigSchema, E as updateManualScoreRequestSchema, En as appendToEvalOutput, Et as evalChartColorSchema, F as getEvalDisplayStatus, Fn as mergeEvalOutput, Ft as cacheEntryWithDebugKeySchema, G as apiCallMetricPlacementSchema, Gt as traceCacheRefSchema, H as DEFAULT_LLM_CALLS_CONFIG, Hn as startEvalBackgroundJob, Ht as cacheStatusSchema, I as deriveScopedSummaryFromCases, In as nextEvalId, It as cacheFileSchema, J as defaultConfigKeySchema, Jt as traceAttributeDisplayPlacementSchema, K as apiCallMetricSchema, Kt as traceAttributeDisplayFormatSchema, L as deriveStatusFromCaseRows, Ln as runInEvalRuntimeScope, Lt as cacheListItemSchema, M as applyDerivedCallAttributes, Mn as getEvalStartTime, Mt as cacheDebugKeyEntrySchema, N as getNestedAttribute, Nn as incrementEvalOutput, Nt as cacheDebugKeyFileSchema, O as extractCacheEntries, On as evalAssert, Ot as evalChartMetricSchema, P as getEvalTitle, Pn as isInEvalScope, Pt as cacheEntrySchema, Q as llmCallPricingSchema, Qt as traceSpanErrorSchema, R as deriveStatusFromChildStatuses, Rn as runInEvalScope, Rt as cacheModeSchema, Sn as repoFile, St as scoreTraceSchema, T as createRunRequestSchema, Tn as advanceEvalTime, Tt as evalChartBuiltinMetricSchema, U as agentEvalsConfigSchema, Un as defineEval, Ut as serializedCacheSpanSchema, V as DEFAULT_API_CALLS_CONFIG, Vn as setScopeCacheContext, Vt as cacheRecordingSchema, W as apiCallMetricFormatSchema, Wn as getEvalRegistry, Wt as spanCacheOptionsSchema, X as llmCallMetricPlacementSchema, Xt as traceDisplayConfigSchema, Y as llmCallMetricFormatSchema, Yt as traceAttributeDisplaySchema, Z as llmCallMetricSchema, Zt as traceDisplayInputConfigSchema, _n as hashCacheKeySync, _t as evalSummarySchema, an as columnKindSchema, at as buildCaseKey, bn as serializeCacheRecording, bt as runLogLocationSchema, cn as numberDisplayOptionsSchema, ct as getCaseRowEvalKey, dn as z, dt as caseRowSchema, en as traceSpanSchema, et as removeDefaultConfigSchema, fn as buildTraceTree, ft as discoveryIssueSchema, gn as hashCacheKey, gt as evalStatsConfigSchema, hn as evalTracer, ht as evalStatItemSchema, in as columnFormatSchema, it as trialSelectionModeSchema, j as extractLlmCalls, jn as getEvalCaseInput, jt as evalChartsConfigSchema, k as extractCacheHits, kn as evalLog, kt as evalChartTooltipExtraSchema, ln as repoFileRefSchema, lt as assertionFailureSchema, mn as evalSpan, mt as evalStatAggregateSchema, nn as cellValueSchema, nt as resolveLlmCallsConfig, on as fileRefSchema, ot as buildEvalKey, pn as captureEvalSpanError, pt as evalFreshnessStatusSchema, q as apiCallsConfigSchema, qt as traceAttributeDisplayInputSchema, rn as columnDefSchema, rt as runLogsConfigSchema, sn as jsonCellSchema, st as getCaseRowCaseKey, tn as traceSpanWarningSchema, tt as resolveApiCallsConfig, un as runArtifactRefSchema, ut as caseDetailSchema, vn as deserializeCacheRecording, vt as runLogEntrySchema, wn as EvalAssertionError, wt as evalChartAxisSchema, xn as serializeCacheValue, xt as runLogPhaseSchema, yn as deserializeCacheValue, yt as runLogLevelSchema, z as runManifestSchema, zn as runInExistingEvalScope, zt as cacheOperationTypeSchema } from "./runOrchestration-DB0dwGrd.mjs";
2
- import { n as createRunner, t as runCli } from "./cli-DRkwWgTj.mjs";
3
- import "./src-BwKm3sKU.mjs";
4
- export { DEFAULT_API_CALLS_CONFIG, DEFAULT_LLM_CALLS_CONFIG, EvalAssertionError, advanceEvalTime, agentEvalsConfigSchema, apiCallMetricFormatSchema, apiCallMetricPlacementSchema, apiCallMetricSchema, apiCallsConfigSchema, appendToEvalOutput, applyDerivedCallAttributes, assertionFailureSchema, buildCaseKey, buildEvalKey, buildTraceTree, cacheDebugKeyEntrySchema, cacheDebugKeyFileSchema, cacheEntrySchema, cacheEntryWithDebugKeySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheOperationTypeSchema, cacheRecordingOpSchema, cacheRecordingSchema, cacheStatusSchema, captureEvalSpanError, caseDetailSchema, caseRowSchema, cellValueSchema, columnDefSchema, columnFormatSchema, columnKindSchema, createRunRequestSchema, createRunner, defaultConfigKeySchema, defineEval, deriveScopedSummaryFromCases, deriveStatusFromCaseRows, deriveStatusFromChildStatuses, deserializeCacheRecording, deserializeCacheValue, discoveryIssueSchema, evalAssert, evalChartAggregateSchema, evalChartAxisSchema, evalChartBuiltinMetricSchema, evalChartColorSchema, evalChartConfigSchema, evalChartMetricSchema, evalChartTooltipExtraSchema, evalChartTypeSchema, evalChartsConfigSchema, evalExpect, evalFreshnessStatusSchema, evalLog, evalSpan, evalStatAggregateSchema, evalStatItemSchema, evalStatsConfigSchema, evalSummarySchema, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, fileRefSchema, getCaseRowCaseKey, getCaseRowEvalKey, getCurrentScope, getEvalCaseInput, getEvalDisplayStatus, getEvalRegistry, getEvalStartTime, getEvalTitle, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, jsonCellSchema, llmCallMetricFormatSchema, llmCallMetricPlacementSchema, llmCallMetricSchema, llmCallPricingSchema, llmCallsConfigSchema, mergeEvalOutput, nextEvalId, numberDisplayOptionsSchema, removeDefaultConfigSchema, repoFile, repoFileRefSchema, resolveApiCallsConfig, resolveLlmCallsConfig, runArtifactRefSchema, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, runLogEntrySchema, runLogLevelSchema, runLogLocationSchema, runLogPhaseSchema, runLogsConfigSchema, runManifestSchema, runSummarySchema, scoreTraceSchema, serializeCacheRecording, serializeCacheValue, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, startEvalBackgroundJob, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceCacheRefSchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, traceSpanWarningSchema, trialSelectionModeSchema, updateManualScoreRequestSchema, z };
1
+ import { $ as llmCallMetricFormatSchema, $t as traceAttributeDisplayPlacementSchema, A as extractCacheHits, An as advanceEvalTime, At as evalChartBuiltinMetricSchema, B as runManifestSchema, Bn as mergeEvalOutput, Bt as cacheEntryWithDebugKeySchema, Cn as deserializeCacheRecording, Ct as runLogEntrySchema, D as updateManualScoreRequestSchema, Dn as repoFile, Dt as scoreTraceSchema, E as createRunRequestSchema, En as serializeCacheValue, Et as runLogPhaseSchema, F as getEvalTitle, Fn as getCurrentScope, Ft as evalChartTypeSchema, G as apiCallMetricFormatSchema, Gn as setEvalOutput, Gt as cacheRecordingOpSchema, H as DEFAULT_API_CALLS_CONFIG, Hn as runInEvalRuntimeScope, Ht as cacheListItemSchema, I as getEvalDisplayStatus, In as getEvalCaseInput, It as evalChartsConfigSchema, J as apiCallsConfigSchema, Jn as defineEval, Jt as serializedCacheSpanSchema, K as apiCallMetricPlacementSchema, Kn as setScopeCacheContext, Kt as cacheRecordingSchema, L as deriveScopedSummaryFromCases, Ln as getEvalStartTime, Lt as cacheDebugKeyEntrySchema, M as extractLlmCalls, Mt as evalChartConfigSchema, N as applyDerivedCallAttributes, Nn as evalAssert, Nt as evalChartMetricSchema, O as sseEnvelopeSchema, On as evalExpect, Ot as evalChartAggregateSchema, P as getNestedAttribute, Pn as evalLog, Pt as evalChartTooltipExtraSchema, Q as evalDeriveConfigSchema, Qt as traceAttributeDisplayInputSchema, R as deriveStatusFromCaseRows, Rn as incrementEvalOutput, Rt as cacheDebugKeyFileSchema, Sn as hashCacheKeySync, St as evalSummarySchema, Tn as serializeCacheRecording, Tt as runLogLocationSchema, U as DEFAULT_LLM_CALLS_CONFIG, Un as runInEvalScope, Ut as cacheModeSchema, V as runSummarySchema, Vn as nextEvalId, Vt as cacheFileSchema, W as agentEvalsConfigSchema, Wn as runInExistingEvalScope, Wt as cacheOperationTypeSchema, X as evalColumnOverrideSchema, Xt as traceCacheRefSchema, Y as defaultConfigKeySchema, Yn as getEvalRegistry, Yt as spanCacheOptionsSchema, Z as evalColumnsSchema, Zt as traceAttributeDisplayFormatSchema, _n as buildTraceTree, _t as discoveryIssueSchema, an as traceSpanSchema, at as removeDefaultConfigSchema, bn as evalTracer, bt as evalStatItemSchema, cn as columnDefSchema, ct as runLogsConfigSchema, dn as fileRefSchema, dt as buildEvalKey, en as traceAttributeDisplaySchema, et as llmCallMetricPlacementSchema, fn as jsonCellSchema, ft as getCaseRowCaseKey, gn as z, gt as caseRowSchema, hn as runArtifactRefSchema, ht as caseDetailSchema, in as traceSpanKindSchema, it as llmCallsConfigSchema, j as extractApiCalls, jn as appendToEvalOutput, jt as evalChartColorSchema, k as extractCacheEntries, kn as EvalAssertionError, kt as evalChartAxisSchema, ln as columnFormatSchema, lt as trialSelectionModeSchema, mn as repoFileRefSchema, mt as assertionFailureSchema, nn as traceDisplayInputConfigSchema, nt as llmCallPricingRateSchema, on as traceSpanWarningSchema, ot as resolveApiCallsConfig, pn as numberDisplayOptionsSchema, pt as getCaseRowEvalKey, q as apiCallMetricSchema, qn as startEvalBackgroundJob, qt as cacheStatusSchema, rn as traceSpanErrorSchema, rt as llmCallPricingSchema, sn as cellValueSchema, st as resolveLlmCallsConfig, tn as traceDisplayConfigSchema, tt as llmCallMetricSchema, un as columnKindSchema, ut as buildCaseKey, vn as captureEvalSpanError, vt as evalFreshnessStatusSchema, wn as deserializeCacheValue, wt as runLogLevelSchema, xn as hashCacheKey, xt as evalStatsConfigSchema, yn as evalSpan, yt as evalStatAggregateSchema, z as deriveStatusFromChildStatuses, zn as isInEvalScope, zt as cacheEntrySchema } from "./runOrchestration-ClWYWPen.mjs";
2
+ import { n as createRunner, t as runCli } from "./cli-BQwRbqsL.mjs";
3
+ import "./src-CuirVcPY.mjs";
4
+ export { DEFAULT_API_CALLS_CONFIG, DEFAULT_LLM_CALLS_CONFIG, EvalAssertionError, advanceEvalTime, agentEvalsConfigSchema, apiCallMetricFormatSchema, apiCallMetricPlacementSchema, apiCallMetricSchema, apiCallsConfigSchema, appendToEvalOutput, applyDerivedCallAttributes, assertionFailureSchema, buildCaseKey, buildEvalKey, buildTraceTree, cacheDebugKeyEntrySchema, cacheDebugKeyFileSchema, cacheEntrySchema, cacheEntryWithDebugKeySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheOperationTypeSchema, cacheRecordingOpSchema, cacheRecordingSchema, cacheStatusSchema, captureEvalSpanError, caseDetailSchema, caseRowSchema, cellValueSchema, columnDefSchema, columnFormatSchema, columnKindSchema, createRunRequestSchema, createRunner, defaultConfigKeySchema, defineEval, deriveScopedSummaryFromCases, deriveStatusFromCaseRows, deriveStatusFromChildStatuses, deserializeCacheRecording, deserializeCacheValue, discoveryIssueSchema, evalAssert, evalChartAggregateSchema, evalChartAxisSchema, evalChartBuiltinMetricSchema, evalChartColorSchema, evalChartConfigSchema, evalChartMetricSchema, evalChartTooltipExtraSchema, evalChartTypeSchema, evalChartsConfigSchema, evalColumnOverrideSchema, evalColumnsSchema, evalDeriveConfigSchema, evalExpect, evalFreshnessStatusSchema, evalLog, evalSpan, evalStatAggregateSchema, evalStatItemSchema, evalStatsConfigSchema, evalSummarySchema, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, fileRefSchema, getCaseRowCaseKey, getCaseRowEvalKey, getCurrentScope, getEvalCaseInput, getEvalDisplayStatus, getEvalRegistry, getEvalStartTime, getEvalTitle, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, jsonCellSchema, llmCallMetricFormatSchema, llmCallMetricPlacementSchema, llmCallMetricSchema, llmCallPricingRateSchema, llmCallPricingSchema, llmCallsConfigSchema, mergeEvalOutput, nextEvalId, numberDisplayOptionsSchema, removeDefaultConfigSchema, repoFile, repoFileRefSchema, resolveApiCallsConfig, resolveLlmCallsConfig, runArtifactRefSchema, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, runLogEntrySchema, runLogLevelSchema, runLogLocationSchema, runLogPhaseSchema, runLogsConfigSchema, runManifestSchema, runSummarySchema, scoreTraceSchema, serializeCacheRecording, serializeCacheValue, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, startEvalBackgroundJob, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceCacheRefSchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, traceSpanWarningSchema, trialSelectionModeSchema, updateManualScoreRequestSchema, z };
package/dist/runChild.mjs CHANGED
@@ -1,4 +1,4 @@
1
- import { B as runSummarySchema, Dn as configureEvalRunLogs, T as createRunRequestSchema, b as loadConfig, gt as evalStatsConfigSchema, jt as evalChartsConfigSchema, ot as buildEvalKey, r as getTargetEvals$1, rn as columnDefSchema, t as executeRun, v as parseEvalDiscovery, w as createFsCacheStore, z as runManifestSchema } from "./runOrchestration-DB0dwGrd.mjs";
1
+ import { B as runManifestSchema, E as createRunRequestSchema, It as evalChartsConfigSchema, Mn as configureEvalRunLogs, T as createFsCacheStore, V as runSummarySchema, cn as columnDefSchema, dt as buildEvalKey, r as getTargetEvals$1, t as executeRun, x as loadConfig, xt as evalStatsConfigSchema, y as parseEvalDiscovery } from "./runOrchestration-ClWYWPen.mjs";
2
2
  import { createHash } from "node:crypto";
3
3
  import { readFile } from "node:fs/promises";
4
4
  import { relative } from "node:path";