@tangle-network/agent-eval 0.118.2 → 0.118.3

package/dist/chunk-K4DBDHLK.js

@@ -0,0 +1,158 @@
+// src/trace/attribute-vocabulary.ts
+import {
+  GEN_AI_INPUT_TOKEN_KEYS,
+  GEN_AI_MODEL_KEYS,
+  GEN_AI_OUTPUT_TOKEN_KEYS
+} from "@tangle-network/agent-core/telemetry";
+var OPENINFERENCE_SPAN_KIND = "openinference.span.kind";
+var LLM_MODEL_NAME = "llm.model_name";
+var LLM_INPUT_TOKENS = "llm.token_count.prompt";
+var LLM_OUTPUT_TOKENS = "llm.token_count.completion";
+var LLM_REASONING_TOKENS = "llm.token_count.reasoning";
+var LLM_CACHED_TOKENS = "llm.token_count.prompt_cache_hit";
+var LLM_CACHE_WRITE_TOKENS = "llm.token_count.prompt_cache_write";
+var LLM_CONTEXT_TOKENS = "tangle.llm.context_tokens";
+var LLM_COST_USD = "llm.cost_usd";
+var TOOL_NAME = "tool.name";
+var TOOL_ARGS_CAPTURED = "tool.args_captured";
+var TOOL_LATENCY_MS = "tool.latency_ms";
+var INPUT_VALUE = "input.value";
+var OUTPUT_VALUE = "output.value";
+var SPAN_KIND_ATTR_KEYS = [OPENINFERENCE_SPAN_KIND, "inference.observation_kind"];
+var LLM_MODEL_ATTR_KEYS = [
+  LLM_MODEL_NAME,
+  "inference.llm.model_name",
+  "llm.model",
+  ...GEN_AI_MODEL_KEYS,
+  "tangle.model"
+];
+var LLM_INPUT_TOKEN_ATTR_KEYS = [
+  LLM_INPUT_TOKENS,
+  "inference.llm.input_tokens",
+  "llm.input_tokens",
+  ...GEN_AI_INPUT_TOKEN_KEYS,
+  "tangle.tokens.in",
+  "tokens.in"
+];
+var LLM_OUTPUT_TOKEN_ATTR_KEYS = [
+  LLM_OUTPUT_TOKENS,
+  "inference.llm.output_tokens",
+  "llm.output_tokens",
+  ...GEN_AI_OUTPUT_TOKEN_KEYS,
+  "tangle.tokens.out",
+  "tokens.out"
+];
+var LLM_REASONING_TOKEN_ATTR_KEYS = [
+  LLM_REASONING_TOKENS,
+  "inference.llm.reasoning_tokens",
+  "llm.reasoning_tokens",
+  "gen_ai.usage.reasoning_tokens",
+  "gen_ai.usage.reasoning_output_tokens",
+  "reasoning_tokens",
+  "reasoning_output_tokens",
+  "tangle.tokens.reasoning",
+  "gen_ai.usage.output_tokens_details.reasoning_tokens",
+  "gen_ai.usage.completion_tokens_details.reasoning_tokens"
+];
+var LLM_CACHED_TOKEN_ATTR_KEYS = [
+  LLM_CACHED_TOKENS,
+  "inference.llm.cached_tokens",
+  "llm.cached_tokens",
+  "gen_ai.usage.cached_tokens",
+  "gen_ai.usage.prompt_tokens_details.cached_tokens",
+  "gen_ai.usage.input_tokens_details.cached_tokens",
+  "gen_ai.usage.cache_read_tokens",
+  "gen_ai.usage.cache_read_input_tokens",
+  "cache_read_tokens",
+  "cache_read_input_tokens",
+  "input_cache_read",
+  "tangle.tokens.cached"
+];
+var LLM_CACHE_WRITE_TOKEN_ATTR_KEYS = [
+  LLM_CACHE_WRITE_TOKENS,
+  "inference.llm.cache_write_tokens",
+  "llm.cache_write_tokens",
+  "gen_ai.usage.cache_creation_tokens",
+  "gen_ai.usage.cache_creation_input_tokens",
+  "cache_creation_tokens",
+  "cache_creation_input_tokens",
+  "input_cache_creation",
+  "tangle.tokens.cache_write"
+];
+var LLM_COST_ATTR_KEYS = [
+  LLM_COST_USD,
+  "inference.llm.cost.total",
+  "llm.cost.total",
+  "gen_ai.usage.cost",
+  "gen_ai.usage.cost_usd",
+  "tangle.cost.usd",
+  "cost.usd",
+  "cost"
+];
+var RUN_COST_ATTR_KEYS = ["tangle.cost.usd", "cost.usd"];
+var TOOL_NAME_ATTR_KEYS = [TOOL_NAME, "inference.tool.name"];
+function asNumber(value) {
+  if (typeof value === "number") return Number.isFinite(value) ? value : null;
+  if (typeof value === "string" && value.length > 0) {
+    const parsed = Number(value);
+    return Number.isFinite(parsed) ? parsed : null;
+  }
+  return null;
+}
+function firstNumberAttr(attributes, keys) {
+  for (const key of keys) {
+    const value = asNumber(attributes[key]);
+    if (value !== null) return value;
+  }
+  return null;
+}
+function contextInputTokens(usage) {
+  if (usage.inputTokens === void 0 || usage.inputTokens === null) return void 0;
+  return usage.inputTokens + (usage.cachedTokens ?? 0) + (usage.cacheWriteTokens ?? 0);
+}
+function applyLlmSpanOtlpAttributes(attributes, span) {
+  if (span.model !== void 0) attributes[LLM_MODEL_NAME] = span.model;
+  if (span.inputTokens !== void 0) attributes[LLM_INPUT_TOKENS] = span.inputTokens;
+  if (span.outputTokens !== void 0) attributes[LLM_OUTPUT_TOKENS] = span.outputTokens;
+  if (span.reasoningTokens !== void 0) attributes[LLM_REASONING_TOKENS] = span.reasoningTokens;
+  if (span.cachedTokens !== void 0) attributes[LLM_CACHED_TOKENS] = span.cachedTokens;
+  if (span.cacheWriteTokens !== void 0) {
+    attributes[LLM_CACHE_WRITE_TOKENS] = span.cacheWriteTokens;
+  }
+  const contextTokens = contextInputTokens(span);
+  if (contextTokens !== void 0) attributes[LLM_CONTEXT_TOKENS] = contextTokens;
+  if (span.costUsd !== void 0) attributes[LLM_COST_USD] = span.costUsd;
+  if (span.finishReason) attributes["llm.finish_reason"] = span.finishReason;
+}
+
+export {
+  OPENINFERENCE_SPAN_KIND,
+  LLM_MODEL_NAME,
+  LLM_INPUT_TOKENS,
+  LLM_OUTPUT_TOKENS,
+  LLM_REASONING_TOKENS,
+  LLM_CACHED_TOKENS,
+  LLM_CACHE_WRITE_TOKENS,
+  LLM_CONTEXT_TOKENS,
+  LLM_COST_USD,
+  TOOL_NAME,
+  TOOL_ARGS_CAPTURED,
+  TOOL_LATENCY_MS,
+  INPUT_VALUE,
+  OUTPUT_VALUE,
+  SPAN_KIND_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
+  LLM_COST_ATTR_KEYS,
+  RUN_COST_ATTR_KEYS,
+  TOOL_NAME_ATTR_KEYS,
+  asNumber,
+  firstNumberAttr,
+  contextInputTokens,
+  applyLlmSpanOtlpAttributes
+};
+//# sourceMappingURL=chunk-K4DBDHLK.js.map

package/dist/chunk-K4DBDHLK.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/trace/attribute-vocabulary.ts"],"sourcesContent":["/** Canonical OpenInference-over-OTLP attribute names used at trace boundaries. */\n\nimport {\n  GEN_AI_INPUT_TOKEN_KEYS,\n  GEN_AI_MODEL_KEYS,\n  GEN_AI_OUTPUT_TOKEN_KEYS,\n} from '@tangle-network/agent-core/telemetry'\n\nexport const OPENINFERENCE_SPAN_KIND = 'openinference.span.kind'\nexport const LLM_MODEL_NAME = 'llm.model_name'\nexport const LLM_INPUT_TOKENS = 'llm.token_count.prompt'\nexport const LLM_OUTPUT_TOKENS = 'llm.token_count.completion'\nexport const LLM_REASONING_TOKENS = 'llm.token_count.reasoning'\nexport const LLM_CACHED_TOKENS = 'llm.token_count.prompt_cache_hit'\nexport const LLM_CACHE_WRITE_TOKENS = 'llm.token_count.prompt_cache_write'\n/** Exact prompt context after summing mutually exclusive input/cache categories. */\nexport const LLM_CONTEXT_TOKENS = 'tangle.llm.context_tokens'\nexport const LLM_COST_USD = 'llm.cost_usd'\nexport const TOOL_NAME = 'tool.name'\nexport const TOOL_ARGS_CAPTURED = 'tool.args_captured'\nexport const TOOL_LATENCY_MS = 'tool.latency_ms'\nexport const INPUT_VALUE = 'input.value'\nexport const OUTPUT_VALUE = 'output.value'\n\nexport const SPAN_KIND_ATTR_KEYS = [OPENINFERENCE_SPAN_KIND, 'inference.observation_kind'] as const\n\nexport const LLM_MODEL_ATTR_KEYS = [\n  LLM_MODEL_NAME,\n  'inference.llm.model_name',\n  'llm.model',\n  ...GEN_AI_MODEL_KEYS,\n  'tangle.model',\n] as const\n\nexport const LLM_INPUT_TOKEN_ATTR_KEYS = [\n  LLM_INPUT_TOKENS,\n  'inference.llm.input_tokens',\n  'llm.input_tokens',\n  ...GEN_AI_INPUT_TOKEN_KEYS,\n  'tangle.tokens.in',\n  'tokens.in',\n] as const\n\nexport const LLM_OUTPUT_TOKEN_ATTR_KEYS = [\n  LLM_OUTPUT_TOKENS,\n  'inference.llm.output_tokens',\n  'llm.output_tokens',\n  ...GEN_AI_OUTPUT_TOKEN_KEYS,\n  'tangle.tokens.out',\n  'tokens.out',\n] as const\n\n/** Reasoning-token subset of output, when a producer exposes it separately. */\nexport const LLM_REASONING_TOKEN_ATTR_KEYS = [\n  LLM_REASONING_TOKENS,\n  'inference.llm.reasoning_tokens',\n  'llm.reasoning_tokens',\n  'gen_ai.usage.reasoning_tokens',\n  'gen_ai.usage.reasoning_output_tokens',\n  'reasoning_tokens',\n  'reasoning_output_tokens',\n  'tangle.tokens.reasoning',\n  'gen_ai.usage.output_tokens_details.reasoning_tokens',\n  'gen_ai.usage.completion_tokens_details.reasoning_tokens',\n] as const\n\nexport const LLM_CACHED_TOKEN_ATTR_KEYS = [\n  LLM_CACHED_TOKENS,\n  'inference.llm.cached_tokens',\n  'llm.cached_tokens',\n  'gen_ai.usage.cached_tokens',\n  'gen_ai.usage.prompt_tokens_details.cached_tokens',\n  'gen_ai.usage.input_tokens_details.cached_tokens',\n  'gen_ai.usage.cache_read_tokens',\n  'gen_ai.usage.cache_read_input_tokens',\n  'cache_read_tokens',\n  'cache_read_input_tokens',\n  'input_cache_read',\n  'tangle.tokens.cached',\n] as const\n\nexport const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS = [\n  LLM_CACHE_WRITE_TOKENS,\n  'inference.llm.cache_write_tokens',\n  'llm.cache_write_tokens',\n  'gen_ai.usage.cache_creation_tokens',\n  'gen_ai.usage.cache_creation_input_tokens',\n  'cache_creation_tokens',\n  'cache_creation_input_tokens',\n  'input_cache_creation',\n  'tangle.tokens.cache_write',\n] as const\n\nexport const LLM_COST_ATTR_KEYS = [\n  LLM_COST_USD,\n  'inference.llm.cost.total',\n  'llm.cost.total',\n  'gen_ai.usage.cost',\n  'gen_ai.usage.cost_usd',\n  'tangle.cost.usd',\n  'cost.usd',\n  'cost',\n] as const\n\n/** Explicit run-total cost keys safe to preserve on an untyped span. */\nexport const RUN_COST_ATTR_KEYS = ['tangle.cost.usd', 'cost.usd'] as const\n\nexport const TOOL_NAME_ATTR_KEYS = [TOOL_NAME, 'inference.tool.name'] as const\n\n/** Read a numeric attribute, tolerating numeric strings; `null` if absent or invalid. */\nexport function asNumber(value: unknown): number | null {\n  if (typeof value === 'number') return Number.isFinite(value) ? value : null\n  if (typeof value === 'string' && value.length > 0) {\n    const parsed = Number(value)\n    return Number.isFinite(parsed) ? parsed : null\n  }\n  return null\n}\n\n/** First finite numeric value across a list of candidate attribute keys. */\nexport function firstNumberAttr(\n  attributes: Record<string, unknown>,\n  keys: readonly string[],\n): number | null {\n  for (const key of keys) {\n    const value = asNumber(attributes[key])\n    if (value !== null) return value\n  }\n  return null\n}\n\n/** Sum producer-supplied, mutually exclusive prompt-token categories. */\nexport function contextInputTokens(usage: {\n  inputTokens?: number | null\n  cachedTokens?: number | null\n  cacheWriteTokens?: number | null\n}): number | undefined {\n  if (usage.inputTokens === undefined || usage.inputTokens === null) return undefined\n  return usage.inputTokens + (usage.cachedTokens ?? 0) + (usage.cacheWriteTokens ?? 0)\n}\n\nexport interface LlmSpanOtlpInput {\n  model?: string\n  inputTokens?: number\n  outputTokens?: number\n  reasoningTokens?: number\n  cachedTokens?: number\n  cacheWriteTokens?: number\n  costUsd?: number\n  finishReason?: string\n}\n\n/** Write the canonical LLM attributes shared by trace producers. */\nexport function applyLlmSpanOtlpAttributes(\n  attributes: Record<string, unknown>,\n  span: LlmSpanOtlpInput,\n): void {\n  if (span.model !== undefined) attributes[LLM_MODEL_NAME] = span.model\n  if (span.inputTokens !== undefined) attributes[LLM_INPUT_TOKENS] = span.inputTokens\n  if (span.outputTokens !== undefined) attributes[LLM_OUTPUT_TOKENS] = span.outputTokens\n  if (span.reasoningTokens !== undefined) attributes[LLM_REASONING_TOKENS] = span.reasoningTokens\n  if (span.cachedTokens !== undefined) attributes[LLM_CACHED_TOKENS] = span.cachedTokens\n  if (span.cacheWriteTokens !== undefined) {\n    attributes[LLM_CACHE_WRITE_TOKENS] = span.cacheWriteTokens\n  }\n  const contextTokens = contextInputTokens(span)\n  if (contextTokens !== undefined) attributes[LLM_CONTEXT_TOKENS] = contextTokens\n  if (span.costUsd !== undefined) attributes[LLM_COST_USD] = span.costUsd\n  if (span.finishReason) attributes['llm.finish_reason'] = span.finishReason\n}\n"],"mappings":";AAEA;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEA,IAAM,0BAA0B;AAChC,IAAM,iBAAiB;AACvB,IAAM,mBAAmB;AACzB,IAAM,oBAAoB;AAC1B,IAAM,uBAAuB;AAC7B,IAAM,oBAAoB;AAC1B,IAAM,yBAAyB;AAE/B,IAAM,qBAAqB;AAC3B,IAAM,eAAe;AACrB,IAAM,YAAY;AAClB,IAAM,qBAAqB;AAC3B,IAAM,kBAAkB;AACxB,IAAM,cAAc;AACpB,IAAM,eAAe;AAErB,IAAM,sBAAsB,CAAC,yBAAyB,4BAA4B;AAElF,IAAM,sBAAsB;AAAA,EACjC;AAAA,EACA;AAAA,EACA;AAAA,EACA,GAAG;AAAA,EACH;AACF;AAEO,IAAM,4BAA4B;AAAA,EACvC;AAAA,EACA;AAAA,EACA;AAAA,EACA,GAAG;AAAA,EACH;AAAA,EACA;AACF;AAEO,IAAM,6BAA6B;AAAA,EACxC;AAAA,EACA;AAAA,EACA;AAAA,EACA,GAAG;AAAA,EACH;AAAA,EACA;AACF;AAGO,IAAM,gCAAgC;AAAA,EAC3C;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEO,IAAM,6BAA6B;AAAA,EACxC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEO,IAAM,kCAAkC;AAAA,EAC7C;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAEO,IAAM,qBAAqB;AAAA,EAChC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAGO,IAAM,qBAAqB,CAAC,mBAAmB,UAAU;AAEzD,IAAM,sBAAsB,CAAC,WAAW,qBAAqB;AAG7D,SAAS,SAAS,OAA+B;AACtD,MAAI,OAAO,UAAU,SAAU,QAAO,OAAO,SAAS,KAAK,IAAI,QAAQ;AACvE,MAAI,OAAO,UAAU,YAAY,MAAM,SAAS,GAAG;AACjD,UAAM,SAAS,OAAO,KAAK;AAC3B,WAAO,OAAO,SAAS,MAAM,IAAI,SAAS;AAAA,EAC5C;AACA,SAAO;AACT;AAGO,SAAS,gBACd,YACA,MACe;AACf,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,SAAS,WAAW,GAAG,CAAC;AACtC,QAAI,UAAU,KAAM,QAAO;AAAA,EAC7B;AACA,SAAO;AACT;AAGO,SAAS,mBAAmB,OAIZ;AACrB,MAAI,MAAM,gBAAgB,UAAa,MAAM,gBAAgB,KAAM,QAAO;AAC1E,SAAO,MAAM,eAAe,MAAM,gBAAgB,MAAM,MAAM,oBAAoB;AACpF;AAcO,SAAS,2BACd,YACA,MACM;AACN,MAAI,KAAK,UAAU,OAAW,YAAW,cAAc,IAAI,KAAK;AAChE,MAAI,KAAK,gBAAgB,OAAW,YAAW,gBAAgB,IAAI,KAAK;AACxE,MAAI,KAAK,iBAAiB,OAAW,YAAW,iBAAiB,IAAI,KAAK;AAC1E,MAAI,KAAK,oBAAoB,OAAW,YAAW,oBAAoB,IAAI,KAAK;AAChF,MAAI,KAAK,iBAAiB,OAAW,YAAW,iBAAiB,IAAI,KAAK;AAC1E,MAAI,KAAK,qBAAqB,QAAW;AACvC,eAAW,sBAAsB,IAAI,KAAK;AAAA,EAC5C;AACA,QAAM,gBAAgB,mBAAmB,IAAI;AAC7C,MAAI,kBAAkB,OAAW,YAAW,kBAAkB,IAAI;AAClE,MAAI,KAAK,YAAY,OAAW,YAAW,YAAY,IAAI,KAAK;AAChE,MAAI,KAAK,aAAc,YAAW,mBAAmB,IAAI,KAAK;AAChE;","names":[]}

package/dist/{chunk-I42AAGHL.js → chunk-QWRLW4CT.js}

@@ -9,7 +9,7 @@ import {
 import {
   clamp01,
   validatePolicyEditCandidateRecord
-} from "./chunk-DZBCDDIE.js";
+} from "./chunk-VNLWDTDA.js";
 import {
   detectRewardHacking
 } from "./chunk-ARU2PZFM.js";
@@ -3610,4 +3610,4 @@ export {
   provenanceSpansPath,
   emitLoopProvenance
 };
-//# sourceMappingURL=chunk-I42AAGHL.js.map
+//# sourceMappingURL=chunk-QWRLW4CT.js.map

package/dist/{chunk-DZBCDDIE.js → chunk-VNLWDTDA.js}

@@ -10,7 +10,7 @@ import {
 import {
   TraceFileMissingError,
   buildTraceAnalystTools
-} from "./chunk-BUBU3YHT.js";
+} from "./chunk-6QIM2EAP.js";
 import {
   validateAgentProfileCell
 } from "./chunk-XJYR7XFV.js";
@@ -1940,4 +1940,4 @@ export {
   aggregateRunScore,
   clamp012 as clamp01
 };
-//# sourceMappingURL=chunk-DZBCDDIE.js.map
+//# sourceMappingURL=chunk-VNLWDTDA.js.map

package/dist/{chunk-6JQE3YVE.js → chunk-ZYHJNKI3.js}

@@ -1,23 +1,22 @@
 import {
   recordAggregateMeasurements,
   summarizeExecutionMeasurements
-} from "./chunk-TVXPJJ7H.js";
+} from "./chunk-H5UD2323.js";
 import {
   extractUsage,
   extractUsageFromSse
 } from "./chunk-PXE2VKMX.js";
 import {
   analyzeTraces
-} from "./chunk-63KG3ASG.js";
+} from "./chunk-FIUFRXP3.js";
 import {
-  LLM_MODEL_ATTR_KEYS,
   applyToolSpanOtlpAttributes,
   compareSpanTime,
   firstStringAttr,
   projectOtlpFlatLine,
   spanEpochMillis,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-BUBU3YHT.js";
+} from "./chunk-6QIM2EAP.js";
 import {
   defaultProviderRedactor,
   providerFromBaseUrl
@@ -34,12 +33,13 @@ import {
 } from "./chunk-ONWEPEDO.js";
 import {
   LLM_INPUT_TOKENS,
+  LLM_MODEL_ATTR_KEYS,
   LLM_MODEL_NAME,
   LLM_OUTPUT_TOKENS,
   OPENINFERENCE_SPAN_KIND,
   TOOL_NAME,
   applyLlmSpanOtlpAttributes
-} from "./chunk-NW4AQBXD.js";
+} from "./chunk-K4DBDHLK.js";
 
 // src/trace-analyst/hook.ts
 var DEFAULT_QUESTION = "Summarise what happened in this run. Surface any failure modes, surprising findings, or evidence that the run's verdict is wrong.";
@@ -1897,4 +1897,4 @@ export {
   createReplayFetch,
   iterateRawCalls
 };
-//# sourceMappingURL=chunk-6JQE3YVE.js.map
+//# sourceMappingURL=chunk-ZYHJNKI3.js.map

package/dist/contract/index.js

@@ -34,7 +34,7 @@ import {
   runReferenceEquivalenceJudge,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-I42AAGHL.js";
+} from "../chunk-QWRLW4CT.js";
 import {
   createRunCostLedger,
   fsCampaignStorage,
@@ -45,9 +45,9 @@ import {
 import {
   buildDefaultAnalystRegistry,
   createChatClient
-} from "../chunk-J23QK2U5.js";
+} from "../chunk-267UCQWI.js";
 import "../chunk-HHWE3POT.js";
-import "../chunk-DZBCDDIE.js";
+import "../chunk-VNLWDTDA.js";
 import {
   FileSystemOutcomeStore,
   InMemoryOutcomeStore
@@ -61,13 +61,11 @@ import "../chunk-VI2UW6B6.js";
 import {
   recordAggregateMeasurements,
   summarizeExecutionMeasurements
-} from "../chunk-TVXPJJ7H.js";
+} from "../chunk-H5UD2323.js";
 import "../chunk-PXE2VKMX.js";
 import "../chunk-ZET2UAYW.js";
 import "../chunk-GGE4NNQT.js";
-import {
-  LLM_MODEL_ATTR_KEYS
-} from "../chunk-BUBU3YHT.js";
+import "../chunk-6QIM2EAP.js";
 import "../chunk-PC4UYEBM.js";
 import {
   parseRunRecordSafe
@@ -76,7 +74,9 @@ import "../chunk-MA6HLL3S.js";
 import "../chunk-XJYR7XFV.js";
 import "../chunk-VSMTAMNK.js";
 import "../chunk-ONWEPEDO.js";
-import "../chunk-NW4AQBXD.js";
+import {
+  LLM_MODEL_ATTR_KEYS
+} from "../chunk-K4DBDHLK.js";
 import "../chunk-PZ5AY32C.js";
 
 // src/contract/self-improve.ts

package/dist/index.d.ts

@@ -6053,6 +6053,22 @@ declare const TOOL_ARGS_CAPTURED = "tool.args_captured";
 declare const TOOL_LATENCY_MS = "tool.latency_ms";
 declare const INPUT_VALUE = "input.value";
 declare const OUTPUT_VALUE = "output.value";
+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", ...string[], "tangle.model"];
+declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", ...string[], "tangle.tokens.in", "tokens.in"];
+declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", ...string[], "tangle.tokens.out", "tokens.out"];
+/** Reasoning-token subset of output, when a producer exposes it separately. */
+declare const LLM_REASONING_TOKEN_ATTR_KEYS: readonly ["llm.token_count.reasoning", "inference.llm.reasoning_tokens", "llm.reasoning_tokens", "gen_ai.usage.reasoning_tokens", "gen_ai.usage.reasoning_output_tokens", "reasoning_tokens", "reasoning_output_tokens", "tangle.tokens.reasoning", "gen_ai.usage.output_tokens_details.reasoning_tokens", "gen_ai.usage.completion_tokens_details.reasoning_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", "gen_ai.usage.prompt_tokens_details.cached_tokens", "gen_ai.usage.input_tokens_details.cached_tokens", "gen_ai.usage.cache_read_tokens", "gen_ai.usage.cache_read_input_tokens", "cache_read_tokens", "cache_read_input_tokens", "input_cache_read", "tangle.tokens.cached"];
+declare const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_write", "inference.llm.cache_write_tokens", "llm.cache_write_tokens", "gen_ai.usage.cache_creation_tokens", "gen_ai.usage.cache_creation_input_tokens", "cache_creation_tokens", "cache_creation_input_tokens", "input_cache_creation", "tangle.tokens.cache_write"];
+declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost", "gen_ai.usage.cost_usd", "tangle.cost.usd", "cost.usd", "cost"];
+/** Explicit run-total cost keys safe to preserve on an untyped span. */
+declare const RUN_COST_ATTR_KEYS: readonly ["tangle.cost.usd", "cost.usd"];
+declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
+/** Read a numeric attribute, tolerating numeric strings; `null` if absent or invalid. */
+declare function asNumber(value: unknown): number | null;
+/** First finite numeric value across a list of candidate attribute keys. */
+declare function firstNumberAttr(attributes: Record<string, unknown>, keys: readonly string[]): number | null;
 /** Sum producer-supplied, mutually exclusive prompt-token categories. */
 declare function contextInputTokens(usage: {
     inputTokens?: number | null;
@@ -6074,18 +6090,6 @@ declare function applyLlmSpanOtlpAttributes(attributes: Record<string, unknown>,
 
 /** Canonical OpenInference-over-OTLP attribute vocabulary used at the trace boundary. */
 
-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", ...string[], "tangle.model"];
-declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", ...string[], "tangle.tokens.in", "tokens.in"];
-declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", ...string[], "tangle.tokens.out", "tokens.out"];
-/** Reasoning-token subset of output, when a producer exposes it separately. */
-declare const LLM_REASONING_TOKEN_ATTR_KEYS: readonly ["llm.token_count.reasoning", "inference.llm.reasoning_tokens", "llm.reasoning_tokens", "gen_ai.usage.reasoning_tokens", "gen_ai.usage.reasoning_output_tokens", "reasoning_tokens", "reasoning_output_tokens", "tangle.tokens.reasoning", "gen_ai.usage.output_tokens_details.reasoning_tokens", "gen_ai.usage.completion_tokens_details.reasoning_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", "gen_ai.usage.prompt_tokens_details.cached_tokens", "gen_ai.usage.input_tokens_details.cached_tokens", "gen_ai.usage.cache_read_tokens", "gen_ai.usage.cache_read_input_tokens", "cache_read_tokens", "cache_read_input_tokens", "input_cache_read", "tangle.tokens.cached"];
-declare const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_write", "inference.llm.cache_write_tokens", "llm.cache_write_tokens", "gen_ai.usage.cache_creation_tokens", "gen_ai.usage.cache_creation_input_tokens", "cache_creation_tokens", "cache_creation_input_tokens", "input_cache_creation", "tangle.tokens.cache_write"];
-declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost", "gen_ai.usage.cost_usd", "tangle.cost.usd", "cost.usd", "cost"];
-/** Explicit run-total cost keys safe to preserve on an untyped span. */
-declare const RUN_COST_ATTR_KEYS: readonly ["tangle.cost.usd", "cost.usd"];
-declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
 type ToolSpanOtlpInput = Pick<ToolSpan, 'toolName' | 'args' | 'argsCaptured' | 'result' | 'latencyMs'>;
 declare function applyToolSpanOtlpAttributes(attributes: Record<string, unknown>, span: ToolSpanOtlpInput): void;
 declare function traceSpanKindToOpenInferenceKind(kind: string): string;
@@ -6463,10 +6467,6 @@ declare function inferOtlpKind(attrs: Record<string, unknown>): TraceAnalystSpan
 declare function extractOtlpAttributes(raw: Record<string, unknown>): Record<string, unknown>;
 declare function stringField(raw: Record<string, unknown>, key: string): string | undefined;
 declare function asString(v: unknown): string | null;
-/** Read a numeric attribute, tolerating numeric strings; `null` if absent/NaN. */
-declare function asNumber(v: unknown): number | null;
-/** First finite numeric value across a list of candidate attribute keys. */
-declare function firstNumberAttr(attrs: Record<string, unknown>, keys: readonly string[]): number | null;
 /** First non-empty string value across a list of candidate attribute keys. */
 declare function firstStringAttr(attrs: Record<string, unknown>, keys: readonly string[]): string | null;
 

package/dist/index.js

@@ -53,7 +53,7 @@ import {
   pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-GCT3DQWU.js";
+} from "./chunk-EKHHBKS6.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -91,7 +91,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-I42AAGHL.js";
+} from "./chunk-QWRLW4CT.js";
 import {
   BackendIntegrityError,
   assertRealBackend,
@@ -115,12 +115,12 @@ import {
   defaultIsMaterial,
   diffFindings,
   runSemanticConceptJudge
-} from "./chunk-DRSFTBKU.js";
+} from "./chunk-F5A4GDQW.js";
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics,
   createChatClient
-} from "./chunk-J23QK2U5.js";
+} from "./chunk-267UCQWI.js";
 import "./chunk-HHWE3POT.js";
 import {
   Mutex
@@ -154,7 +154,7 @@ import {
   scorePolicyEditReadiness,
   validatePolicyEdit,
   validatePolicyEditCandidateRecord
-} from "./chunk-DZBCDDIE.js";
+} from "./chunk-VNLWDTDA.js";
 import {
   allCriticalPassed,
   controlFailureClassFromVerification,
@@ -303,8 +303,8 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-6JQE3YVE.js";
-import "./chunk-TVXPJJ7H.js";
+} from "./chunk-ZYHJNKI3.js";
+import "./chunk-H5UD2323.js";
 import {
   extractUsage,
   extractUsageFromResponse,
@@ -326,7 +326,7 @@ import {
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   TRACE_ANALYST_SUBAGENT_DESCRIPTION,
   analyzeTraces
-} from "./chunk-63KG3ASG.js";
+} from "./chunk-FIUFRXP3.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -335,27 +335,15 @@ import {
 } from "./chunk-GGE4NNQT.js";
 import {
   DEFAULT_TRACE_ANALYST_BUDGETS,
-  LLM_CACHED_TOKEN_ATTR_KEYS,
-  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
-  LLM_COST_ATTR_KEYS,
-  LLM_INPUT_TOKEN_ATTR_KEYS,
-  LLM_MODEL_ATTR_KEYS,
-  LLM_OUTPUT_TOKEN_ATTR_KEYS,
-  LLM_REASONING_TOKEN_ATTR_KEYS,
   OtlpFileTraceStore,
-  RUN_COST_ATTR_KEYS,
-  SPAN_KIND_ATTR_KEYS,
   SpanNotFoundError,
-  TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,
   TraceFileMissingError,
   TraceNotFoundError,
   applyToolSpanOtlpAttributes,
-  asNumber,
   asString,
   buildTraceAnalystTools,
   extractOtlpAttributes,
-  firstNumberAttr,
   firstStringAttr,
   inferOtlpKind,
   projectOtlpFlatLine,
@@ -363,7 +351,7 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-BUBU3YHT.js";
+} from "./chunk-6QIM2EAP.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
@@ -432,21 +420,33 @@ import {
 import {
   INPUT_VALUE,
   LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
   LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_CONTEXT_TOKENS,
+  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,
   LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OPENINFERENCE_SPAN_KIND,
   OUTPUT_VALUE,
+  RUN_COST_ATTR_KEYS,
+  SPAN_KIND_ATTR_KEYS,
   TOOL_ARGS_CAPTURED,
   TOOL_LATENCY_MS,
   TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   applyLlmSpanOtlpAttributes,
-  contextInputTokens
-} from "./chunk-NW4AQBXD.js";
+  asNumber,
+  contextInputTokens,
+  firstNumberAttr
+} from "./chunk-K4DBDHLK.js";
 import {
   __export
 } from "./chunk-PZ5AY32C.js";

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.118.2",
+    "version": "0.118.3",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.0.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/trace-attributes.d.ts

@@ -14,6 +14,22 @@ declare const TOOL_ARGS_CAPTURED = "tool.args_captured";
 declare const TOOL_LATENCY_MS = "tool.latency_ms";
 declare const INPUT_VALUE = "input.value";
 declare const OUTPUT_VALUE = "output.value";
+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", ...string[], "tangle.model"];
+declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", ...string[], "tangle.tokens.in", "tokens.in"];
+declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", ...string[], "tangle.tokens.out", "tokens.out"];
+/** Reasoning-token subset of output, when a producer exposes it separately. */
+declare const LLM_REASONING_TOKEN_ATTR_KEYS: readonly ["llm.token_count.reasoning", "inference.llm.reasoning_tokens", "llm.reasoning_tokens", "gen_ai.usage.reasoning_tokens", "gen_ai.usage.reasoning_output_tokens", "reasoning_tokens", "reasoning_output_tokens", "tangle.tokens.reasoning", "gen_ai.usage.output_tokens_details.reasoning_tokens", "gen_ai.usage.completion_tokens_details.reasoning_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", "gen_ai.usage.prompt_tokens_details.cached_tokens", "gen_ai.usage.input_tokens_details.cached_tokens", "gen_ai.usage.cache_read_tokens", "gen_ai.usage.cache_read_input_tokens", "cache_read_tokens", "cache_read_input_tokens", "input_cache_read", "tangle.tokens.cached"];
+declare const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_write", "inference.llm.cache_write_tokens", "llm.cache_write_tokens", "gen_ai.usage.cache_creation_tokens", "gen_ai.usage.cache_creation_input_tokens", "cache_creation_tokens", "cache_creation_input_tokens", "input_cache_creation", "tangle.tokens.cache_write"];
+declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost", "gen_ai.usage.cost_usd", "tangle.cost.usd", "cost.usd", "cost"];
+/** Explicit run-total cost keys safe to preserve on an untyped span. */
+declare const RUN_COST_ATTR_KEYS: readonly ["tangle.cost.usd", "cost.usd"];
+declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
+/** Read a numeric attribute, tolerating numeric strings; `null` if absent or invalid. */
+declare function asNumber(value: unknown): number | null;
+/** First finite numeric value across a list of candidate attribute keys. */
+declare function firstNumberAttr(attributes: Record<string, unknown>, keys: readonly string[]): number | null;
 /** Sum producer-supplied, mutually exclusive prompt-token categories. */
 declare function contextInputTokens(usage: {
     inputTokens?: number | null;
@@ -33,4 +49,4 @@ interface LlmSpanOtlpInput {
 /** Write the canonical LLM attributes shared by trace producers. */
 declare function applyLlmSpanOtlpAttributes(attributes: Record<string, unknown>, span: LlmSpanOtlpInput): void;
 
-export { INPUT_VALUE, LLM_CACHED_TOKENS, LLM_CACHE_WRITE_TOKENS, LLM_CONTEXT_TOKENS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_REASONING_TOKENS, type LlmSpanOtlpInput, OPENINFERENCE_SPAN_KIND, OUTPUT_VALUE, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, applyLlmSpanOtlpAttributes, contextInputTokens };
+export { INPUT_VALUE, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, 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, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, type LlmSpanOtlpInput, OPENINFERENCE_SPAN_KIND, OUTPUT_VALUE, RUN_COST_ATTR_KEYS, SPAN_KIND_ATTR_KEYS, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, applyLlmSpanOtlpAttributes, asNumber, contextInputTokens, firstNumberAttr };

package/dist/trace-attributes.js

@@ -1,38 +1,62 @@
 import {
   INPUT_VALUE,
   LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
   LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_CONTEXT_TOKENS,
+  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,
   LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OPENINFERENCE_SPAN_KIND,
   OUTPUT_VALUE,
+  RUN_COST_ATTR_KEYS,
+  SPAN_KIND_ATTR_KEYS,
   TOOL_ARGS_CAPTURED,
   TOOL_LATENCY_MS,
   TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   applyLlmSpanOtlpAttributes,
-  contextInputTokens
-} from "./chunk-NW4AQBXD.js";
+  asNumber,
+  contextInputTokens,
+  firstNumberAttr
+} from "./chunk-K4DBDHLK.js";
 import "./chunk-PZ5AY32C.js";
 export {
   INPUT_VALUE,
   LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
   LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_CONTEXT_TOKENS,
+  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,
   LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OPENINFERENCE_SPAN_KIND,
   OUTPUT_VALUE,
+  RUN_COST_ATTR_KEYS,
+  SPAN_KIND_ATTR_KEYS,
   TOOL_ARGS_CAPTURED,
   TOOL_LATENCY_MS,
   TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   applyLlmSpanOtlpAttributes,
-  contextInputTokens
+  asNumber,
+  contextInputTokens,
+  firstNumberAttr
 };
 //# sourceMappingURL=trace-attributes.js.map