@tangle-network/agent-eval 0.94.0 → 0.95.1

package/dist/{rubric-predictive-validity-Cy_W-hWZ.d.ts → rubric-predictive-validity-C2hDKM8Z.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-e7vj1uZQ.js';
+import { R as RunRecord } from './run-record-CP2ObebC.js';
 import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
 
 /**

package/dist/{run-campaign-7WNXMDSN.js → run-campaign-WXY7KI67.js}

@@ -1,10 +1,10 @@
 import {
   runCampaign
-} from "./chunk-2K6UUZ7P.js";
+} from "./chunk-2T4EZACH.js";
 import "./chunk-56GC6VYO.js";
 import "./chunk-3BFEG2F6.js";
 import "./chunk-PZ5AY32C.js";
 export {
   runCampaign
 };
-//# sourceMappingURL=run-campaign-7WNXMDSN.js.map
+//# sourceMappingURL=run-campaign-WXY7KI67.js.map

package/dist/{run-record-e7vj1uZQ.d.ts → run-record-CP2ObebC.d.ts}

@@ -1,3 +1,4 @@
+import { AgentProfile } from '@tangle-network/agent-interface';
 import { V as ValidationError } from './errors-CzMUYo7b.js';
 import { F as FailureClass } from './schema-m0gsnbt3.js';
 
@@ -7,7 +8,7 @@ type AgentProfileJson = string | number | boolean | null | AgentProfileJson[] |
 };
 type AgentProfileDimensionValue = string | number | boolean | null;
 interface AgentProfileSource {
-    /** Runtime/profile contract being fingerprinted, e.g. `sandbox-agent-profile`. */
+    /** Runtime/profile contract being fingerprinted, e.g. `agent-interface-profile`. */
     kind: string;
     /** sha256 over the canonical source profile object. */
     hash: string;
@@ -74,9 +75,9 @@ declare function groupRunsByAgentProfileCell<T extends {
  *  new kind is a deliberate cross-product schema change. */
 declare const AGENT_PROFILE_KINDS: {
     /** A profile declared via `defineAgentProfile(...)` from
-     *  `@tangle-network/sandbox`. The default kind for sandbox-hosted
-     *  products (gtm-agent, blueprint-agent, sandbox, evals). */
-    readonly SANDBOX_AGENT_PROFILE: "sandbox-agent-profile";
+     *  `@tangle-network/agent-interface`. The default kind for router-backed
+     *  and sandbox-backed products. */
+    readonly AGENT_INTERFACE_PROFILE: "agent-interface-profile";
 };
 type AgentProfileKind = (typeof AGENT_PROFILE_KINDS)[keyof typeof AGENT_PROFILE_KINDS];
 /** Canonicalize an arbitrary value into `AgentProfileJson` by JSON
@@ -84,26 +85,21 @@ type AgentProfileKind = (typeof AGENT_PROFILE_KINDS)[keyof typeof AGENT_PROFILE_
  *  as JSON (functions, BigInt, cycles) — non-portable profiles fail loud
  *  rather than silently dropping fields. */
 declare function toAgentProfileJson(value: unknown): AgentProfileJson;
-/** Minimal shape required of any sandbox-SDK `AgentProfile` — anything
- *  with a non-empty `name` and `version` plus JSON-serializable contents.
- *  Compatible with `defineAgentProfile(...)` output from
- *  `@tangle-network/sandbox`; products that have not yet declared a real
- *  profile can pass a `{ name, version, ...metadata }` stub. */
-interface SandboxAgentProfileLike {
+/** Canonical AgentProfile shape required when deriving a stable cell id. */
+type AgentInterfaceProfileLike = AgentProfile & {
     name: string;
     version: string;
-    [key: string]: unknown;
-}
+};
 /** Higher-level helper that hard-codes the canonical
- *  `sandbox-agent-profile` kind plus the JSON canonicalization. Equivalent
+ *  `agent-interface-profile` kind plus the JSON canonicalization. Equivalent
  *  to calling `buildAgentProfileCell` with `profileId = \`${name}@${version}\``
- *  and `sourceProfile = { kind: SANDBOX_AGENT_PROFILE, profile: <round-tripped> }`.
+ *  and `sourceProfile = { kind: AGENT_INTERFACE_PROFILE, profile: <round-tripped> }`.
  *
- *  Use this from any product consuming a sandbox-SDK `AgentProfile`; the
+ *  Use this from any product consuming an agent-interface `AgentProfile`; the
  *  manual `buildAgentProfileCell` call is reserved for advanced cases
  *  (custom kinds, pre-computed source hashes, alternate profileId
  *  conventions). */
-declare function buildSandboxAgentProfileCell(profile: SandboxAgentProfileLike, input: Omit<AgentProfileCellInput, 'profileId' | 'sourceProfile'>): Promise<AgentProfileCell>;
+declare function buildAgentInterfaceProfileCell(profile: AgentInterfaceProfileLike, input: Omit<AgentProfileCellInput, 'profileId' | 'sourceProfile'>): Promise<AgentProfileCell>;
 
 /**
  * Paper-grade RunRecord schema + runtime validator.
@@ -221,7 +217,7 @@ interface RunOutcome {
  *     model (after any steering bundle merge).
  *   - `configHash` is the sha256 of the effective run config (model,
  *     temperature, tools, judges, splits). The pair (promptHash,
- *     configHash) uniquely identifies an experimental cell.
+ *     configHash) uniquely identifies an experiment cell.
  *
  * Model snapshot discipline:
  *   - `model` MUST encode a snapshot version. Bare aliases like
@@ -315,4 +311,4 @@ declare function parseRunRecordSafe(input: unknown): {
 /** Round-trip helper — `JSON.parse(JSON.stringify(record))` then validate. */
 declare function roundTripRunRecord(record: RunRecord): RunRecord;
 
-export { type AgentProfileCell as A, validateAgentProfileCell as B, validateRunRecord as C, verifyAgentProfileCell as D, type JudgeScoresRecord as J, type RunRecord as R, type SandboxAgentProfileLike as S, type RunSplitTag as a, type RunTokenUsage as b, type RunJudgeMetadata as c, type AgentProfileCellInput as d, AGENT_PROFILE_KINDS as e, type AgentProfileCellSchemaVersion as f, AgentProfileCellValidationError as g, type AgentProfileDimensionValue as h, type AgentProfileHarness as i, type AgentProfileJson as j, type AgentProfileKind as k, type AgentProfileSource as l, type AgentProfileSourceInput as m, type RunOutcome as n, RunRecordValidationError as o, agentProfileCellHashMaterial as p, agentProfileCellKey as q, assertRunAgentProfileCell as r, buildAgentProfileCell as s, buildSandboxAgentProfileCell as t, groupRunsByAgentProfileCell as u, isRunRecord as v, parseRunRecordSafe as w, requireAgentProfileCell as x, roundTripRunRecord as y, toAgentProfileJson as z };
+export { type AgentProfileCell as A, toAgentProfileJson as B, validateAgentProfileCell as C, validateRunRecord as D, verifyAgentProfileCell as E, type JudgeScoresRecord as J, type RunRecord as R, type RunSplitTag as a, type RunTokenUsage as b, type RunJudgeMetadata as c, type AgentProfileCellInput as d, AGENT_PROFILE_KINDS as e, type AgentInterfaceProfileLike as f, type AgentProfileCellSchemaVersion as g, AgentProfileCellValidationError as h, type AgentProfileDimensionValue as i, type AgentProfileHarness as j, type AgentProfileJson as k, type AgentProfileKind as l, type AgentProfileSource as m, type AgentProfileSourceInput as n, type RunOutcome as o, RunRecordValidationError as p, agentProfileCellHashMaterial as q, agentProfileCellKey as r, assertRunAgentProfileCell as s, buildAgentInterfaceProfileCell as t, buildAgentProfileCell as u, groupRunsByAgentProfileCell as v, isRunRecord as w, parseRunRecordSafe as x, requireAgentProfileCell as y, roundTripRunRecord as z };

package/dist/{runtime-trajectory-BDgfGZSr.d.ts → runtime-trajectory-BOUUjI0y.d.ts}

@@ -1,4 +1,4 @@
-import { a as RunSplitTag } from './run-record-e7vj1uZQ.js';
+import { a as RunSplitTag } from './run-record-CP2ObebC.js';
 
 interface RuntimeTrajectoryHookEvent {
     id: string;

package/dist/{semantic-concept-judge-B9MgmBnM.d.ts → semantic-concept-judge-DSBB2Cfp.d.ts}

@@ -1,7 +1,7 @@
 import { AxAIService } from '@ax-llm/ax';
 import { z } from 'zod';
-import { c as AnalystFinding, A as Analyst, a as AnalystContext } from './types-Ce17tDlG.js';
-import { T as TraceAnalystKindSpec } from './kind-factory-0BhLSI27.js';
+import { c as AnalystFinding, A as Analyst, a as AnalystContext } from './types-B5x54y6n.js';
+import { T as TraceAnalystKindSpec } from './kind-factory-X3eDYbKn.js';
 import { a as TraceAnalystSpan } from './store-C1YxJDEK.js';
 import { L as LlmClientOptions } from './llm-client-Bj7g0rqu.js';
 import { S as Severity } from './multi-layer-verifier-DUZXrPDA.js';

package/dist/{summary-report-BDOFevaT.d.ts → summary-report-CInXwsza.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-e7vj1uZQ.js';
+import { R as RunRecord } from './run-record-CP2ObebC.js';
 import { F as FailureClusterReport } from './failure-cluster-DH9Flgcf.js';
 
 /**

package/dist/testing-C21CHsq2.d.ts

@@ -0,0 +1,20 @@
+/**
+ * LockedJsonlAppender — mutex-serialized JSONL append helper for arbitrary
+ * payloads. The reference-replay store does the same thing for typed
+ * `ReferenceReplayRun` rows; this is the generic version used by
+ * `MutationTelemetry`, `TrialTelemetry`, and any other consumer that wants
+ * append-only durable telemetry without rolling its own lock.
+ *
+ * Locks are per absolute file path (process-local). Cross-process
+ * concurrency is NOT addressed — that's an fcntl/flock problem.
+ */
+declare class LockedJsonlAppender {
+    readonly path: string;
+    private readonly mutex;
+    constructor(path: string);
+    append(entry: unknown): Promise<void>;
+}
+/** Reset all internal mutex state — tests only. */
+declare function resetLockedAppendersForTesting(): void;
+
+export { LockedJsonlAppender as L, resetLockedAppendersForTesting as r };

package/dist/testing.d.ts

@@ -0,0 +1 @@
+export { r as resetLockedAppendersForTesting } from './testing-C21CHsq2.js';

package/dist/testing.js

@@ -0,0 +1,8 @@
+import {
+  resetLockedAppendersForTesting
+} from "./chunk-DJWX3GVS.js";
+import "./chunk-PZ5AY32C.js";
+export {
+  resetLockedAppendersForTesting
+};
+//# sourceMappingURL=testing.js.map

package/dist/testing.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}

package/dist/traces.d.ts

@@ -14,8 +14,26 @@ import { A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-C
 export { a as AnalyzeTracesInput, c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
 import { h as TraceAnalystSpanKind, i as TraceAnalystSpanStatus, T as TraceAnalysisStore, g as TraceAnalystFilters, b as DatasetOverview, Q as QueryTracesPage, l as ViewTraceResult, V as ViewSpansResult, c as SearchTraceResult, S as SearchSpanResult } from './store-C1YxJDEK.js';
 export { D as DEFAULT_TRACE_ANALYST_BUDGETS, E as ErrorCluster, d as SpanMatchRecord, e as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, f as TraceAnalystByteBudgets, a as TraceAnalystSpan, j as TraceAnalystTraceSummary, k as ViewTraceOversized } from './store-C1YxJDEK.js';
-import { a as RunSplitTag, b as RunTokenUsage, R as RunRecord } from './run-record-e7vj1uZQ.js';
+import { a as RunSplitTag, b as RunTokenUsage, R as RunRecord } from './run-record-CP2ObebC.js';
 import { AxFunction } from '@ax-llm/ax';
+import '@tangle-network/agent-interface';
+
+/** Canonical OpenInference-over-OTLP attribute vocabulary used at the trace boundary. */
+declare const OPENINFERENCE_SPAN_KIND = "openinference.span.kind";
+declare const LLM_MODEL_NAME = "llm.model_name";
+declare const LLM_INPUT_TOKENS = "llm.token_count.prompt";
+declare const LLM_OUTPUT_TOKENS = "llm.token_count.completion";
+declare const LLM_CACHED_TOKENS = "llm.token_count.prompt_cache_hit";
+declare const LLM_COST_USD = "llm.cost_usd";
+declare const TOOL_NAME = "tool.name";
+declare const SPAN_KIND_ATTR_KEYS: readonly ["openinference.span.kind", "inference.observation_kind"];
+declare const LLM_MODEL_ATTR_KEYS: readonly ["llm.model_name", "inference.llm.model_name", "llm.model", "gen_ai.request.model", "gen_ai.response.model"];
+declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", "gen_ai.usage.input_tokens", "gen_ai.usage.prompt_tokens"];
+declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", "gen_ai.usage.output_tokens", "gen_ai.usage.completion_tokens"];
+declare const LLM_CACHED_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_hit", "inference.llm.cached_tokens", "llm.cached_tokens", "gen_ai.usage.cached_tokens"];
+declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost"];
+declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
+declare function traceSpanKindToOpenInferenceKind(kind: string): string;
 
 /**
  * Trace-analyst auto-execution hook.
@@ -232,11 +250,9 @@ interface OtlpFlatLine {
     }>;
 }
 interface FlattenOtlpOptions {
-    /** `'openinference'` (default) mirrors per-span attributes into the
-     *  OpenInference vocabulary the analyst's `inferKind` reads
-     *  (`llm.model`→`llm.model_name`, `tool.name`→`inference.tool.name`,
-     *  `span.kind`→`openinference.span.kind` uppercased). `'none'` passes
-     *  attributes through untouched. */
+    /** `'openinference'` (default) mirrors legacy per-span attributes into the
+     *  canonical OpenInference vocabulary the analyst readers consume. `'none'`
+     *  passes attributes through untouched. */
     attributeVocabulary?: 'openinference' | 'none';
     /** Override the numeric-kind → otlp-string mapping. */
     kindMap?: Partial<Record<number, string>>;
@@ -321,9 +337,9 @@ declare function firstStringAttr(attrs: Record<string, unknown>, keys: readonly
  * same shape `flattenOtlpExportToNdjson` produces) into validated
  * `RunRecord[]` — one record per `trace_id` (one trace == one task).
  *
- * This is the offline ingestion primitive the AppWorld driver bench and the
+ * This is the offline ingestion primitive the AppWorld proposer bench and the
  * hosted Intelligence product both stand on: traces in, paper-grade rows
- * out, ready for `compareDrivers` / `analyzeRuns` / the promotion gate.
+ * out, ready for `compareProposers` / `analyzeRuns` / the promotion gate.
  *
  * Aggregation per trace:
  *   - tokenUsage: sum LLM-span `input` / `output` (+ `cached` when present)
@@ -354,7 +370,7 @@ interface OtlpToRunRecordsOptions {
     /** Logical experiment grouping for every produced record. */
     experimentId: string;
     /** Candidate (variant) id — the surface these traces exercised. The
-     *  bench passes the driver label here so `compareDrivers` can pair rows. */
+     *  bench passes the proposer label here so `compareProposers` can pair rows. */
     candidateId: string;
     /** Split assignment for every produced record. Default `'holdout'` —
      *  ingested traces are evidence, not the optimizer's training pool. */
@@ -957,4 +973,4 @@ declare function iterateRawCalls(sink: RawProviderSink, filter?: {
     spanId?: string;
 }): AsyncGenerator<ReplayCacheEntry>;
 
-export { AnalyzeTracesOptions, AnalyzeTracesResult, type CaptureFetchContext, type CaptureFetchOptions, DatasetOverview, type ExportableSpan, type ExtractedUsage, type FlattenOtlpOptions, OTEL_AGENT_EVAL_SCOPE, type OtelExportConfig, type OtelExporter, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, type ProjectedOtlpSpan, ProviderRedactor, QueryTracesPage, RawProviderEvent, RawProviderSink, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, type ReplayFetchOptions, Run, RunCompleteHook, RunCompleteHookContext, SearchSpanResult, SearchTraceResult, SpanNotFoundError, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, type TraceAggregate, TraceAnalysisStore, TraceAnalystFilters, type TraceAnalystHookOptions, TraceAnalystSpanKind, TraceAnalystSpanStatus, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracesToOtlpResult, ViewSpansResult, ViewTraceResult, asNumber, asString, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, convertTraceStoresToOtlp, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, inferDomainKeywords, inferOtlpKind, iterateRawCalls, otelRunCompleteHook, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, readOtlpStatus, scoreTraceInsightReadiness, stringField, tokenizeDomainWords, traceAnalystFunctionGroup, traceAnalystOnRunComplete };
+export { AnalyzeTracesOptions, AnalyzeTracesResult, type CaptureFetchContext, type CaptureFetchOptions, DatasetOverview, type ExportableSpan, type ExtractedUsage, type FlattenOtlpOptions, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, type OtelExportConfig, type OtelExporter, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, type ProjectedOtlpSpan, ProviderRedactor, QueryTracesPage, RawProviderEvent, RawProviderSink, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, type ReplayFetchOptions, Run, RunCompleteHook, RunCompleteHookContext, SPAN_KIND_ATTR_KEYS, SearchSpanResult, SearchTraceResult, SpanNotFoundError, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, type TraceAggregate, TraceAnalysisStore, TraceAnalystFilters, type TraceAnalystHookOptions, TraceAnalystSpanKind, TraceAnalystSpanStatus, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracesToOtlpResult, ViewSpansResult, ViewTraceResult, asNumber, asString, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, convertTraceStoresToOtlp, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, inferDomainKeywords, inferOtlpKind, iterateRawCalls, otelRunCompleteHook, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, readOtlpStatus, scoreTraceInsightReadiness, stringField, tokenizeDomainWords, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind };

package/dist/traces.js

@@ -28,7 +28,13 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-CTBHKLEU.js";
+} from "./chunk-77T4STFI.js";
+import {
+  TRACE_ANALYST_ACTOR_DESCRIPTION,
+  TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
+  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
+  analyzeTraces
+} from "./chunk-IZCEK2HR.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -54,16 +60,24 @@ import {
   isSandboxSpan,
   isToolSpan
 } from "./chunk-5BKGXME7.js";
-import {
-  TRACE_ANALYST_ACTOR_DESCRIPTION,
-  TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
-  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
-  analyzeTraces
-} from "./chunk-TBDR6PAI.js";
 import {
   DEFAULT_TRACE_ANALYST_BUDGETS,
+  LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_COST_ATTR_KEYS,
+  LLM_COST_USD,
+  LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
+  LLM_MODEL_NAME,
+  LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
+  OPENINFERENCE_SPAN_KIND,
   OtlpFileTraceStore,
+  SPAN_KIND_ATTR_KEYS,
   SpanNotFoundError,
+  TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,
   TraceFileMissingError,
   TraceNotFoundError,
@@ -77,8 +91,9 @@ import {
   projectOtlpFlatLine,
   readOtlpStatus,
   stringField,
-  traceAnalystFunctionGroup
-} from "./chunk-QMUEXQJS.js";
+  traceAnalystFunctionGroup,
+  traceSpanKindToOpenInferenceKind
+} from "./chunk-YBIGNSCZ.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
@@ -91,7 +106,7 @@ import {
   defaultProviderRedactor,
   providerFromBaseUrl
 } from "./chunk-PC4UYEBM.js";
-import "./chunk-KWRRMR3J.js";
+import "./chunk-LO6IOIJ2.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
@@ -107,14 +122,28 @@ export {
   FileSystemTraceStore,
   InMemoryRawProviderSink,
   InMemoryTraceStore,
+  LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_COST_ATTR_KEYS,
+  LLM_COST_USD,
+  LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
+  LLM_MODEL_NAME,
+  LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
   NoopRawProviderSink,
+  OPENINFERENCE_SPAN_KIND,
   OTEL_AGENT_EVAL_SCOPE,
   OtlpFileTraceStore,
   REDACTION_VERSION,
   ReplayCache,
   ReplayCacheMissError,
   RunIntegrityError,
+  SPAN_KIND_ATTR_KEYS,
   SpanNotFoundError,
+  TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   TRACE_ANALYST_SUBAGENT_DESCRIPTION,
@@ -178,6 +207,7 @@ export {
   tokenizeDomainWords,
   toolSpans,
   traceAnalystFunctionGroup,
-  traceAnalystOnRunComplete
+  traceAnalystOnRunComplete,
+  traceSpanKindToOpenInferenceKind
 };
 //# sourceMappingURL=traces.js.map

package/dist/{types-Ce17tDlG.d.ts → types-B5x54y6n.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-e7vj1uZQ.js';
+import { R as RunRecord } from './run-record-CP2ObebC.js';
 import { T as TraceAnalysisStore } from './store-C1YxJDEK.js';
 import { a as JudgeInput } from './types-C7DGg5ex.js';
 import { b as LlmCallRequest, c as LlmCallResult } from './llm-client-Bj7g0rqu.js';

package/dist/{types-mn5Aqk7x.d.ts → types-BUxNaJ8c.d.ts}

@@ -1,13 +1,11 @@
 import { D as DefaultVerdict } from './verdict-C9MlYujm.js';
 
 /**
- * @experimental
- *
  * N-axis cartesian matrix over substrate types — types module.
  *
  * The matrix is a runner + aggregator. It iterates the cartesian product of
- * caller-provided axes (any value type — `AgentProfile` from sandbox, `Driver`
- * / `Validator` from agent-runtime, rubric records, thinking levels, anything)
+ * caller-provided axes (any value type — `AgentProfile` from agent-interface,
+ * `Driver` / `Validator` from agent-runtime, rubric records, thinking levels, anything)
  * and aggregates per-axis pass/score/cost summaries. Substrate types are
  * imported at the boundary by JSDoc only; the matrix never wraps them.
  */