@tangle-network/agent-eval 0.202.0 → 0.202.2

package/dist/{claude-jsonl-CxZZrDJ3.js → claude-jsonl-DoQJBCVR.js}

@@ -29,21 +29,33 @@ async function findClaudeTranscripts(cwd, projectsDir = DEFAULT_CLAUDE_PROJECTS_
 }
 const isRecord = (v) => typeof v === "object" && v !== null && !Array.isArray(v);
 /** Parse transcript jsonl text into conversation lines. Non-conversation lines are dropped. */
-function parseClaudeEntries(raw) {
+function parseClaudeEntries(raw, onGap) {
 	const out = [];
-	for (const line of raw.split("\n")) {
+	for (const [index, line] of raw.split("\n").entries()) {
 		if (!line.trim()) continue;
-		let entry;
+		const gap = (reason) => {
+			const detail = `line ${index + 1}: ${reason}`;
+			if (onGap) onGap(detail);
+			else throw new Error(`Claude transcript ${detail}`);
+		};
+		let parsed;
 		try {
-			const parsed = JSON.parse(line);
-			if (!isRecord(parsed)) continue;
-			entry = parsed;
+			parsed = JSON.parse(line);
 		} catch {
+			gap("malformed JSON");
 			continue;
 		}
+		if (!isRecord(parsed)) {
+			gap("JSON value is not an object");
+			continue;
+		}
+		const entry = parsed;
 		if (entry.type !== "user" && entry.type !== "assistant") continue;
 		const message = entry.message;
-		if (!isRecord(message)) continue;
+		if (!isRecord(message)) {
+			gap("conversation message is missing");
+			continue;
+		}
 		out.push({
 			type: entry.type,
 			timestamp: typeof entry.timestamp === "string" ? entry.timestamp : null,
@@ -62,7 +74,11 @@ function blockText(content) {
 }
 /** Parse one transcript jsonl into canonical messages + usage totals. */
 async function readClaudeTranscript(path, options = {}) {
-	return transcriptFromEntries(parseClaudeEntries(await readFile(path, "utf8")), options);
+	const gaps = [];
+	return transcriptFromEntries(parseClaudeEntries(await readFile(path, "utf8"), (gap) => gaps.push(gap)), {
+		...options,
+		sourceGaps: gaps
+	});
 }
 /** The messages+usage projection of already-parsed entries. */
 function transcriptFromEntries(entries, options = {}) {
@@ -74,6 +90,8 @@ function transcriptFromEntries(entries, options = {}) {
 		cacheRead: 0,
 		cacheWrite: 0
 	};
+	const gaps = [...options.sourceGaps ?? []];
+	let usageObserved = false;
 	let startedAt = null;
 	let endedAt = null;
 	let model = null;
@@ -118,11 +136,20 @@ function transcriptFromEntries(entries, options = {}) {
 		const apiId = typeof message.id === "string" ? message.id : null;
 		const continuesTurn = apiId !== null && apiId === lastAssistantApiId && lastAssistantIndex >= 0;
 		const msgUsage = message.usage;
-		if (isRecord(msgUsage) && !continuesTurn) {
-			usage.tokensIn += typeof msgUsage.input_tokens === "number" ? msgUsage.input_tokens : 0;
-			usage.tokensOut += typeof msgUsage.output_tokens === "number" ? msgUsage.output_tokens : 0;
-			usage.cacheRead += typeof msgUsage.cache_read_input_tokens === "number" ? msgUsage.cache_read_input_tokens : 0;
-			usage.cacheWrite += typeof msgUsage.cache_creation_input_tokens === "number" ? msgUsage.cache_creation_input_tokens : 0;
+		if (!continuesTurn) {
+			usageObserved = true;
+			for (const [target, source] of [
+				["tokensIn", "input_tokens"],
+				["tokensOut", "output_tokens"],
+				["cacheRead", "cache_read_input_tokens"],
+				["cacheWrite", "cache_creation_input_tokens"]
+			]) {
+				const value = isRecord(msgUsage) ? msgUsage[source] : void 0;
+				if (typeof value !== "number" || !Number.isFinite(value) || value < 0) {
+					usage[target] = null;
+					gaps.push(`assistant ${apiId ?? "without id"}: ${source} unavailable`);
+				} else if (usage[target] !== null) usage[target] += value;
+			}
 		}
 		const content = message.content;
 		if (!Array.isArray(content)) continue;
@@ -159,15 +186,22 @@ function transcriptFromEntries(entries, options = {}) {
 		lastAssistantApiId = apiId;
 		lastAssistantIndex = messages.length - 1;
 	}
+	if (!usageObserved) {
+		usage.tokensIn = null;
+		usage.tokensOut = null;
+		usage.cacheRead = null;
+		usage.cacheWrite = null;
+	}
 	return {
 		messages,
 		usage,
 		startedAt,
 		endedAt,
-		model
+		model,
+		gaps
 	};
 }
 //#endregion
 export { readClaudeTranscript as a, parseClaudeEntries as i, claudeProjectSlug as n, transcriptFromEntries as o, findClaudeTranscripts as r, DEFAULT_CLAUDE_PROJECTS_DIR as t };
 
-//# sourceMappingURL=claude-jsonl-CxZZrDJ3.js.map
+//# sourceMappingURL=claude-jsonl-DoQJBCVR.js.map

package/dist/claude-jsonl-DoQJBCVR.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-jsonl-DoQJBCVR.js","names":[],"sources":["../src/rollout/readers/claude-jsonl.ts"],"sourcesContent":["/**\n * Backfill reader over Claude Code project transcripts\n * (~/.claude/projects/<cwd-slug>/<sessionId>.jsonl) → canonical\n * chat-with-tools messages plus per-session token usage.\n *\n * Transcript lines consumed: type:\"user\" (string content or content blocks —\n * text + tool_result) and type:\"assistant\" (content blocks — thinking, text,\n * tool_use; message.usage carries tokens). Sidechain lines (isSidechain=true,\n * subagent threads) are separate invocations and are excluded from the main\n * transcript. Everything else (queue-operation, attachment, last-prompt…) is\n * transport metadata, not conversation.\n */\n\nimport { readdir, readFile } from 'node:fs/promises'\nimport { homedir } from 'node:os'\nimport { join } from 'node:path'\nimport type { ChatMessage, ChatToolCall } from '../schema'\n\nexport const DEFAULT_CLAUDE_PROJECTS_DIR = join(homedir(), '.claude', 'projects')\n\n/** Claude Code's project-directory slug for a working directory. */\nexport function claudeProjectSlug(cwd: string): string {\n  return cwd.replace(/[^a-zA-Z0-9-]/g, '-')\n}\n\nexport interface ClaudeTranscriptRef {\n  sessionId: string\n  path: string\n}\n\n/** Transcript files recorded for sessions launched from `cwd`. */\nexport async function findClaudeTranscripts(\n  cwd: string,\n  projectsDir: string = DEFAULT_CLAUDE_PROJECTS_DIR,\n): Promise<ClaudeTranscriptRef[]> {\n  const dir = join(projectsDir, claudeProjectSlug(cwd))\n  const names = await readdir(dir).catch(() => [])\n  return names\n    .filter((n) => n.endsWith('.jsonl'))\n    .sort()\n    .map((n) => ({ sessionId: n.replace(/\\.jsonl$/, ''), path: join(dir, n) }))\n}\n\nexport interface ClaudeUsageTotals {\n  tokensIn: number | null\n  tokensOut: number | null\n  cacheRead: number | null\n  cacheWrite: number | null\n}\n\nexport interface ClaudeTranscript {\n  messages: ChatMessage[]\n  usage: ClaudeUsageTotals\n  /** Timestamp of the first conversation line; null = empty transcript. */\n  startedAt: string | null\n  endedAt: string | null\n  model: string | null\n  /** Every parse or usage gap retained beside this semantic projection. */\n  gaps: string[]\n}\n\nconst isRecord = (v: unknown): v is Record<string, unknown> =>\n  typeof v === 'object' && v !== null && !Array.isArray(v)\n\n/**\n * One conversation line of a transcript, still in Claude Code's own shape.\n *\n * This is the single line-level parse of the format. `readClaudeTranscript`\n * projects it to canonical messages + usage; the supervision-tree reader\n * (`src/supervisor-run/claude-code-reader.ts`) projects the SAME entries to\n * spawn/settle/steer instants. Two projections, one parser — a second\n * transcript parser is how the two views silently disagree.\n */\nexport interface ClaudeEntry {\n  readonly type: 'user' | 'assistant'\n  /** ISO instant of the line; null when the line carried none. */\n  readonly timestamp: string | null\n  /** The Anthropic message body (`role`, `content`, `model`, `usage`). */\n  readonly message: Record<string, unknown>\n  /** Claude Code's structured tool result, when the line carries one. */\n  readonly toolUseResult: unknown\n  /** True on subagent threads — a separate invocation, not this transcript's turn. */\n  readonly isSidechain: boolean\n  /** Subagent id Claude Code stamps on sidechain lines; null on main-thread lines. */\n  readonly agentId: string | null\n}\n\n/** Parse transcript jsonl text into conversation lines. Non-conversation lines are dropped. */\nexport function parseClaudeEntries(raw: string, onGap?: (gap: string) => void): ClaudeEntry[] {\n  const out: ClaudeEntry[] = []\n  for (const [index, line] of raw.split('\\n').entries()) {\n    if (!line.trim()) continue\n    const gap = (reason: string): void => {\n      const detail = `line ${index + 1}: ${reason}`\n      if (onGap) onGap(detail)\n      else throw new Error(`Claude transcript ${detail}`)\n    }\n    let parsed: unknown\n    try {\n      parsed = JSON.parse(line)\n    } catch {\n      gap('malformed JSON')\n      continue\n    }\n    if (!isRecord(parsed)) {\n      gap('JSON value is not an object')\n      continue\n    }\n    const entry = parsed\n    if (entry.type !== 'user' && entry.type !== 'assistant') continue\n    const message = entry.message\n    if (!isRecord(message)) {\n      gap('conversation message is missing')\n      continue\n    }\n    out.push({\n      type: entry.type,\n      timestamp: typeof entry.timestamp === 'string' ? entry.timestamp : null,\n      message,\n      toolUseResult: entry.toolUseResult,\n      isSidechain: entry.isSidechain === true,\n      agentId: typeof entry.agentId === 'string' ? entry.agentId : null,\n    })\n  }\n  return out\n}\n\nexport interface ReadClaudeTranscriptOptions {\n  /**\n   * Read the sidechain (subagent) thread instead of skipping it. Subagent\n   * transcripts under `<session>/subagents/agent-<id>.jsonl` are sidechain\n   * lines end to end, so their usage is invisible without this.\n   */\n  readonly includeSidechain?: boolean\n  /** Parser diagnostics from the retained source; used by the shared reader. */\n  readonly sourceGaps?: readonly string[]\n}\n\nfunction blockText(content: unknown): string {\n  if (typeof content === 'string') return content\n  if (!Array.isArray(content)) return ''\n  return content\n    .filter(\n      (b): b is Record<string, unknown> =>\n        isRecord(b) && b.type === 'text' && typeof b.text === 'string',\n    )\n    .map((b) => b.text as string)\n    .join('\\n')\n}\n\n/** Parse one transcript jsonl into canonical messages + usage totals. */\nexport async function readClaudeTranscript(\n  path: string,\n  options: ReadClaudeTranscriptOptions = {},\n): Promise<ClaudeTranscript> {\n  const gaps: string[] = []\n  const entries = parseClaudeEntries(await readFile(path, 'utf8'), (gap) => gaps.push(gap))\n  return transcriptFromEntries(entries, { ...options, sourceGaps: gaps })\n}\n\n/** The messages+usage projection of already-parsed entries. */\nexport function transcriptFromEntries(\n  entries: readonly ClaudeEntry[],\n  options: ReadClaudeTranscriptOptions = {},\n): ClaudeTranscript {\n  const wantSidechain = options.includeSidechain === true\n  const messages: ChatMessage[] = []\n  const usage: ClaudeUsageTotals = { tokensIn: 0, tokensOut: 0, cacheRead: 0, cacheWrite: 0 }\n  const gaps = [...(options.sourceGaps ?? [])]\n  let usageObserved = false\n  let startedAt: string | null = null\n  let endedAt: string | null = null\n  let model: string | null = null\n  // Claude Code writes one jsonl line PER CONTENT BLOCK of an API message,\n  // repeating message.id and usage on each — merge blocks into one canonical\n  // assistant turn and count usage once per API message id.\n  let lastAssistantApiId: string | null = null\n  let lastAssistantIndex = -1\n\n  for (const entry of entries) {\n    if (entry.isSidechain !== wantSidechain) continue\n    const message = entry.message\n    if (entry.timestamp !== null) {\n      if (startedAt === null) startedAt = entry.timestamp\n      endedAt = entry.timestamp\n    }\n\n    if (entry.type === 'user') {\n      lastAssistantApiId = null\n      lastAssistantIndex = -1\n      const content = message.content\n      if (typeof content === 'string') {\n        messages.push({ role: 'user', content })\n        continue\n      }\n      if (!Array.isArray(content)) continue\n      // A user line may interleave tool_result blocks (answers to the prior\n      // assistant tool_use) with plain text; preserve order.\n      let userText = ''\n      for (const block of content) {\n        if (!isRecord(block)) continue\n        if (block.type === 'tool_result' && typeof block.tool_use_id === 'string') {\n          messages.push({\n            role: 'tool',\n            tool_call_id: block.tool_use_id,\n            content:\n              blockText(block.content) || (typeof block.content === 'string' ? block.content : ''),\n          })\n        } else if (block.type === 'text' && typeof block.text === 'string') {\n          userText += (userText.length > 0 ? '\\n' : '') + block.text\n        }\n      }\n      if (userText.length > 0) messages.push({ role: 'user', content: userText })\n      continue\n    }\n\n    // assistant\n    if (typeof message.model === 'string') model = message.model\n    const apiId = typeof message.id === 'string' ? message.id : null\n    const continuesTurn = apiId !== null && apiId === lastAssistantApiId && lastAssistantIndex >= 0\n    const msgUsage = message.usage\n    if (!continuesTurn) {\n      usageObserved = true\n      const fields = [\n        ['tokensIn', 'input_tokens'],\n        ['tokensOut', 'output_tokens'],\n        ['cacheRead', 'cache_read_input_tokens'],\n        ['cacheWrite', 'cache_creation_input_tokens'],\n      ] as const\n      for (const [target, source] of fields) {\n        const value = isRecord(msgUsage) ? msgUsage[source] : undefined\n        if (typeof value !== 'number' || !Number.isFinite(value) || value < 0) {\n          usage[target] = null\n          gaps.push(`assistant ${apiId ?? 'without id'}: ${source} unavailable`)\n        } else if (usage[target] !== null) {\n          usage[target] += value\n        }\n      }\n    }\n    const content = message.content\n    if (!Array.isArray(content)) continue\n    let reasoning = ''\n    let text = ''\n    const toolCalls: ChatToolCall[] = []\n    for (const block of content) {\n      if (!isRecord(block)) continue\n      if (\n        block.type === 'thinking' &&\n        typeof block.thinking === 'string' &&\n        block.thinking.length > 0\n      ) {\n        reasoning += (reasoning.length > 0 ? '\\n' : '') + block.thinking\n      } else if (block.type === 'text' && typeof block.text === 'string') {\n        text += (text.length > 0 ? '\\n' : '') + block.text\n      } else if (block.type === 'tool_use' && typeof block.id === 'string') {\n        toolCalls.push({\n          id: block.id,\n          type: 'function',\n          function: {\n            name: typeof block.name === 'string' ? block.name : 'unknown',\n            arguments: JSON.stringify(block.input ?? {}),\n          },\n        })\n      }\n    }\n    if (reasoning.length === 0 && text.length === 0 && toolCalls.length === 0) continue\n    if (continuesTurn) {\n      const prev = messages[lastAssistantIndex]!\n      if (text.length > 0) prev.content = prev.content === null ? text : `${prev.content}\\n${text}`\n      if (reasoning.length > 0) {\n        prev.reasoning_content =\n          prev.reasoning_content === undefined\n            ? reasoning\n            : `${prev.reasoning_content}\\n${reasoning}`\n      }\n      if (toolCalls.length > 0) prev.tool_calls = [...(prev.tool_calls ?? []), ...toolCalls]\n      continue\n    }\n    messages.push({\n      role: 'assistant',\n      content: text.length > 0 ? text : null,\n      ...(reasoning.length > 0 ? { reasoning_content: reasoning } : {}),\n      ...(toolCalls.length > 0 ? { tool_calls: toolCalls } : {}),\n    })\n    lastAssistantApiId = apiId\n    lastAssistantIndex = messages.length - 1\n  }\n\n  if (!usageObserved) {\n    usage.tokensIn = null\n    usage.tokensOut = null\n    usage.cacheRead = null\n    usage.cacheWrite = null\n  }\n  return { messages, usage, startedAt, endedAt, model, gaps }\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAkBA,MAAa,8BAA8B,KAAK,QAAQ,GAAG,WAAW,UAAU;;AAGhF,SAAgB,kBAAkB,KAAqB;CACrD,OAAO,IAAI,QAAQ,kBAAkB,GAAG;AAC1C;;AAQA,eAAsB,sBACpB,KACA,cAAsB,6BACU;CAChC,MAAM,MAAM,KAAK,aAAa,kBAAkB,GAAG,CAAC;CAEpD,QAAO,MADa,QAAQ,GAAG,CAAC,CAAC,YAAY,CAAC,CAAC,EAAA,CAE5C,QAAQ,MAAM,EAAE,SAAS,QAAQ,CAAC,CAAC,CACnC,KAAK,CAAC,CACN,KAAK,OAAO;EAAE,WAAW,EAAE,QAAQ,YAAY,EAAE;EAAG,MAAM,KAAK,KAAK,CAAC;CAAE,EAAE;AAC9E;AAoBA,MAAM,YAAY,MAChB,OAAO,MAAM,YAAY,MAAM,QAAQ,CAAC,MAAM,QAAQ,CAAC;;AA0BzD,SAAgB,mBAAmB,KAAa,OAA8C;CAC5F,MAAM,MAAqB,CAAC;CAC5B,KAAK,MAAM,CAAC,OAAO,SAAS,IAAI,MAAM,IAAI,CAAC,CAAC,QAAQ,GAAG;EACrD,IAAI,CAAC,KAAK,KAAK,GAAG;EAClB,MAAM,OAAO,WAAyB;GACpC,MAAM,SAAS,QAAQ,QAAQ,EAAE,IAAI;GACrC,IAAI,OAAO,MAAM,MAAM;QAClB,MAAM,IAAI,MAAM,qBAAqB,QAAQ;EACpD;EACA,IAAI;EACJ,IAAI;GACF,SAAS,KAAK,MAAM,IAAI;EAC1B,QAAQ;GACN,IAAI,gBAAgB;GACpB;EACF;EACA,IAAI,CAAC,SAAS,MAAM,GAAG;GACrB,IAAI,6BAA6B;GACjC;EACF;EACA,MAAM,QAAQ;EACd,IAAI,MAAM,SAAS,UAAU,MAAM,SAAS,aAAa;EACzD,MAAM,UAAU,MAAM;EACtB,IAAI,CAAC,SAAS,OAAO,GAAG;GACtB,IAAI,iCAAiC;GACrC;EACF;EACA,IAAI,KAAK;GACP,MAAM,MAAM;GACZ,WAAW,OAAO,MAAM,cAAc,WAAW,MAAM,YAAY;GACnE;GACA,eAAe,MAAM;GACrB,aAAa,MAAM,gBAAgB;GACnC,SAAS,OAAO,MAAM,YAAY,WAAW,MAAM,UAAU;EAC/D,CAAC;CACH;CACA,OAAO;AACT;AAaA,SAAS,UAAU,SAA0B;CAC3C,IAAI,OAAO,YAAY,UAAU,OAAO;CACxC,IAAI,CAAC,MAAM,QAAQ,OAAO,GAAG,OAAO;CACpC,OAAO,QACJ,QACE,MACC,SAAS,CAAC,KAAK,EAAE,SAAS,UAAU,OAAO,EAAE,SAAS,QAC1D,CAAC,CACA,KAAK,MAAM,EAAE,IAAc,CAAC,CAC5B,KAAK,IAAI;AACd;;AAGA,eAAsB,qBACpB,MACA,UAAuC,CAAC,GACb;CAC3B,MAAM,OAAiB,CAAC;CAExB,OAAO,sBADS,mBAAmB,MAAM,SAAS,MAAM,MAAM,IAAI,QAAQ,KAAK,KAAK,GAAG,CACpD,GAAG;EAAE,GAAG;EAAS,YAAY;CAAK,CAAC;AACxE;;AAGA,SAAgB,sBACd,SACA,UAAuC,CAAC,GACtB;CAClB,MAAM,gBAAgB,QAAQ,qBAAqB;CACnD,MAAM,WAA0B,CAAC;CACjC,MAAM,QAA2B;EAAE,UAAU;EAAG,WAAW;EAAG,WAAW;EAAG,YAAY;CAAE;CAC1F,MAAM,OAAO,CAAC,GAAI,QAAQ,cAAc,CAAC,CAAE;CAC3C,IAAI,gBAAgB;CACpB,IAAI,YAA2B;CAC/B,IAAI,UAAyB;CAC7B,IAAI,QAAuB;CAI3B,IAAI,qBAAoC;CACxC,IAAI,qBAAqB;CAEzB,KAAK,MAAM,SAAS,SAAS;EAC3B,IAAI,MAAM,gBAAgB,eAAe;EACzC,MAAM,UAAU,MAAM;EACtB,IAAI,MAAM,cAAc,MAAM;GAC5B,IAAI,cAAc,MAAM,YAAY,MAAM;GAC1C,UAAU,MAAM;EAClB;EAEA,IAAI,MAAM,SAAS,QAAQ;GACzB,qBAAqB;GACrB,qBAAqB;GACrB,MAAM,UAAU,QAAQ;GACxB,IAAI,OAAO,YAAY,UAAU;IAC/B,SAAS,KAAK;KAAE,MAAM;KAAQ;IAAQ,CAAC;IACvC;GACF;GACA,IAAI,CAAC,MAAM,QAAQ,OAAO,GAAG;GAG7B,IAAI,WAAW;GACf,KAAK,MAAM,SAAS,SAAS;IAC3B,IAAI,CAAC,SAAS,KAAK,GAAG;IACtB,IAAI,MAAM,SAAS,iBAAiB,OAAO,MAAM,gBAAgB,UAC/D,SAAS,KAAK;KACZ,MAAM;KACN,cAAc,MAAM;KACpB,SACE,UAAU,MAAM,OAAO,MAAM,OAAO,MAAM,YAAY,WAAW,MAAM,UAAU;IACrF,CAAC;SACI,IAAI,MAAM,SAAS,UAAU,OAAO,MAAM,SAAS,UACxD,aAAa,SAAS,SAAS,IAAI,OAAO,MAAM,MAAM;GAE1D;GACA,IAAI,SAAS,SAAS,GAAG,SAAS,KAAK;IAAE,MAAM;IAAQ,SAAS;GAAS,CAAC;GAC1E;EACF;EAGA,IAAI,OAAO,QAAQ,UAAU,UAAU,QAAQ,QAAQ;EACvD,MAAM,QAAQ,OAAO,QAAQ,OAAO,WAAW,QAAQ,KAAK;EAC5D,MAAM,gBAAgB,UAAU,QAAQ,UAAU,sBAAsB,sBAAsB;EAC9F,MAAM,WAAW,QAAQ;EACzB,IAAI,CAAC,eAAe;GAClB,gBAAgB;GAOhB,KAAK,MAAM,CAAC,QAAQ,WAAW;IAL7B,CAAC,YAAY,cAAc;IAC3B,CAAC,aAAa,eAAe;IAC7B,CAAC,aAAa,yBAAyB;IACvC,CAAC,cAAc,6BAA6B;GAEV,GAAG;IACrC,MAAM,QAAQ,SAAS,QAAQ,IAAI,SAAS,UAAU,KAAA;IACtD,IAAI,OAAO,UAAU,YAAY,CAAC,OAAO,SAAS,KAAK,KAAK,QAAQ,GAAG;KACrE,MAAM,UAAU;KAChB,KAAK,KAAK,aAAa,SAAS,aAAa,IAAI,OAAO,aAAa;IACvE,OAAO,IAAI,MAAM,YAAY,MAC3B,MAAM,WAAW;GAErB;EACF;EACA,MAAM,UAAU,QAAQ;EACxB,IAAI,CAAC,MAAM,QAAQ,OAAO,GAAG;EAC7B,IAAI,YAAY;EAChB,IAAI,OAAO;EACX,MAAM,YAA4B,CAAC;EACnC,KAAK,MAAM,SAAS,SAAS;GAC3B,IAAI,CAAC,SAAS,KAAK,GAAG;GACtB,IACE,MAAM,SAAS,cACf,OAAO,MAAM,aAAa,YAC1B,MAAM,SAAS,SAAS,GAExB,cAAc,UAAU,SAAS,IAAI,OAAO,MAAM,MAAM;QACnD,IAAI,MAAM,SAAS,UAAU,OAAO,MAAM,SAAS,UACxD,SAAS,KAAK,SAAS,IAAI,OAAO,MAAM,MAAM;QACzC,IAAI,MAAM,SAAS,cAAc,OAAO,MAAM,OAAO,UAC1D,UAAU,KAAK;IACb,IAAI,MAAM;IACV,MAAM;IACN,UAAU;KACR,MAAM,OAAO,MAAM,SAAS,WAAW,MAAM,OAAO;KACpD,WAAW,KAAK,UAAU,MAAM,SAAS,CAAC,CAAC;IAC7C;GACF,CAAC;EAEL;EACA,IAAI,UAAU,WAAW,KAAK,KAAK,WAAW,KAAK,UAAU,WAAW,GAAG;EAC3E,IAAI,eAAe;GACjB,MAAM,OAAO,SAAS;GACtB,IAAI,KAAK,SAAS,GAAG,KAAK,UAAU,KAAK,YAAY,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI;GACvF,IAAI,UAAU,SAAS,GACrB,KAAK,oBACH,KAAK,sBAAsB,KAAA,IACvB,YACA,GAAG,KAAK,kBAAkB,IAAI;GAEtC,IAAI,UAAU,SAAS,GAAG,KAAK,aAAa,CAAC,GAAI,KAAK,cAAc,CAAC,GAAI,GAAG,SAAS;GACrF;EACF;EACA,SAAS,KAAK;GACZ,MAAM;GACN,SAAS,KAAK,SAAS,IAAI,OAAO;GAClC,GAAI,UAAU,SAAS,IAAI,EAAE,mBAAmB,UAAU,IAAI,CAAC;GAC/D,GAAI,UAAU,SAAS,IAAI,EAAE,YAAY,UAAU,IAAI,CAAC;EAC1D,CAAC;EACD,qBAAqB;EACrB,qBAAqB,SAAS,SAAS;CACzC;CAEA,IAAI,CAAC,eAAe;EAClB,MAAM,WAAW;EACjB,MAAM,YAAY;EAClB,MAAM,YAAY;EAClB,MAAM,aAAa;CACrB;CACA,OAAO;EAAE;EAAU;EAAO;EAAW;EAAS;EAAO;CAAK;AAC5D"}

package/dist/cli.js

@@ -3,7 +3,7 @@ import { g as INSUFFICIENT_FROM, h as operatorYield } from "./search-kernel-Dpni
 import { i as replaySearchLedgerText } from "./search-ledger-CXQPwdWz.js";
 import { t as renderSearchSummary } from "./search-summary-C1nAHgkB.js";
 import { c as hostedTenantFromEnv, g as SearchRunKindSchema, r as shipSearchLedger, t as SearchShipConflictError, u as SEARCH_LEDGER_BATCH_MAX_BYTES } from "./search-shipper-DhhofkDX.js";
-import { i as runRolloutReleaseCli } from "./hf-dataset-kjIfIK_e.js";
+import { i as runRolloutReleaseCli } from "./hf-dataset-DdT7nNkf.js";
 import { t as createChatClient } from "./chat-client-DUhWNQQS.js";
 import { t as runAnalystBenchmarkCommand } from "./benchmark-command-DFs_i62N.js";
 import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-B4jMhqQn.js";

package/dist/contract/index.d.ts

@@ -1,14 +1,14 @@
 import { c as CostLedgerHandle } from "../cost-ledger-DbQdN3nO.js";
 import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-CeDfCRXp.js";
 import { A as createChatClient, T as CreateChatClientOpts, x as ChatClient } from "../types-C0bWoSKC.js";
-import { $ as paretoSignificanceGate, G as BuildEvidenceVectorOptions, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, q as ObjectiveSource } from "../statistical-heldout-HswMvzbP.js";
-import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-DfvQOe91.js";
-import { $i as RunImprovementLoopOptions, $r as GepaOptimizationRecipe, Bi as ReferenceEquivalenceJudgeInput, Ci as compareOptimizationMethods, Dr as DefaultProductionGateOptions, Er as DefaultProductionGateCheck, Gi as runReferenceEquivalenceJudge, Gr as SkillOptOptimizationMethodConfig, Hi as ReferenceEquivalenceJudgeResult, Ji as llmJudge, Jr as skillOptOptimizationMethod, Ki as LlmJudgeDimension, Kr as SkillOptRunnerCommand, Na as campaignSplitDigest, Oa as FinalEvidencePolicy, Or as DefaultProductionRewardHackingOptions, Qi as transientDispatchFailure, Qr as GepaOptimizationMethodConfig, Ri as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Si as OptimizationTokenUsage, Ti as ComparisonCost, Tr as heldOutGate, Ui as ReferenceEquivalenceScenario, Vi as ReferenceEquivalenceJudgeOptions, Wi as createReferenceEquivalenceJudge, Wr as composeGate, Xr as GepaEngineOptions, Yr as GepaAdaptiveEngineRun, Zr as GepaEngineRun, _a as runCampaign, _i as OptimizationMethodProvenance, ai as ExternalTextOptimizationMethodConfig, ci as ExternalOptimizationExample, di as CompareOptimizationMethodsOptions, ea as RunImprovementLoopResult, ei as GepaRunnerCommand, fa as RunEvalOptions, fi as OptimizationMethod, ga as RunCampaignOptions, ha as CampaignCellRetryPolicy, hi as OptimizationMethodInput, ii as externalTextOptimizationMethod, ka as FinalEvidenceUse, kr as defaultProductionGate, li as ExternalTextEvaluationResponse, ma as CampaignCellFailureReceipt, ni as OpenAICompatibleOptimizerModel, oi as ExternalTextOptimizerContext, pa as runEval, pi as OptimizationMethodComparison, qi as LlmJudgeOptions, qr as SkillOptTrainerConfig, ri as OptimizerModelBudget, si as ExternalTextOptimizerResult, ta as runImprovementLoop, ti as gepaOptimizationMethod, vi as OptimizationMethodResult, wr as HeldOutGateOptions, xi as OptimizationPackageSource, zi as REFERENCE_EQUIVALENCE_JUDGE_VERSION } from "../index-giYkbitB.js";
-import { Qt as inMemoryCampaignStorage, Yt as CampaignStorage, Zt as fsCampaignStorage } from "../search-kernel-C9CXvFR7.js";
+import { $ as paretoSignificanceGate, G as BuildEvidenceVectorOptions, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, q as ObjectiveSource } from "../statistical-heldout-BAbd_oXC.js";
+import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-BR1LLIxm.js";
+import { $i as RunImprovementLoopOptions, $r as GepaOptimizationRecipe, Bi as ReferenceEquivalenceJudgeInput, Ci as compareOptimizationMethods, Dr as DefaultProductionGateOptions, Er as DefaultProductionGateCheck, Fa as campaignSplitDigest, Gi as runReferenceEquivalenceJudge, Gr as SkillOptOptimizationMethodConfig, Hi as ReferenceEquivalenceJudgeResult, Ji as llmJudge, Jr as skillOptOptimizationMethod, Ki as LlmJudgeDimension, Kr as SkillOptRunnerCommand, Oa as FinalEvidencePolicy, Or as DefaultProductionRewardHackingOptions, Qi as transientDispatchFailure, Qr as GepaOptimizationMethodConfig, Ri as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Si as OptimizationTokenUsage, Ti as ComparisonCost, Tr as heldOutGate, Ui as ReferenceEquivalenceScenario, Vi as ReferenceEquivalenceJudgeOptions, Wi as createReferenceEquivalenceJudge, Wr as composeGate, Xr as GepaEngineOptions, Yr as GepaAdaptiveEngineRun, Zr as GepaEngineRun, _a as runCampaign, _i as OptimizationMethodProvenance, ai as ExternalTextOptimizationMethodConfig, ci as ExternalOptimizationExample, di as CompareOptimizationMethodsOptions, ea as RunImprovementLoopResult, ei as GepaRunnerCommand, fa as RunEvalOptions, fi as OptimizationMethod, ga as RunCampaignOptions, ha as CampaignCellRetryPolicy, hi as OptimizationMethodInput, ii as externalTextOptimizationMethod, ka as FinalEvidenceUse, kr as defaultProductionGate, li as ExternalTextEvaluationResponse, ma as CampaignCellFailureReceipt, ni as OpenAICompatibleOptimizerModel, oi as ExternalTextOptimizerContext, pa as runEval, pi as OptimizationMethodComparison, qi as LlmJudgeOptions, qr as SkillOptTrainerConfig, ri as OptimizerModelBudget, si as ExternalTextOptimizerResult, ta as runImprovementLoop, ti as gepaOptimizationMethod, vi as OptimizationMethodResult, wr as HeldOutGateOptions, xi as OptimizationPackageSource, zi as REFERENCE_EQUIVALENCE_JUDGE_VERSION } from "../index-BPCdNSMW.js";
+import { Qt as inMemoryCampaignStorage, Yt as CampaignStorage, Zt as fsCampaignStorage } from "../search-kernel-zI-rsHoG.js";
 import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-CNt4iZ67.js";
 import { r as HostedTenant } from "../client-BzL9niqi.js";
 import { _ as ReleaseSummary, a as summarizeExecution, c as ExecutionInsight, d as InsightReport, f as InterRaterInsight, g as Recommendation, h as OutcomeCorrelationInsight, i as analyzeRuns, l as FailureClassTally, m as LiftInsight, n as ExecutionReport, o as CostProvenanceSummary, p as JudgeInsight, r as SummarizeExecutionOptions, s as ExecutionErrorOutcomeCell, t as AnalyzeRunsOptions, u as FailureClusterInsight, v as ScalarDistribution, y as TokenUsageInsight } from "../analyze-runs-CPiKLJFs.js";
-import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, m as SelfImproveRunError, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "../define-agent-eval-DQXLPqWc.js";
-import { $ as FromRunRecordDirResult, A as CodeAgentSessionActionStatus, B as AgentTraceFile, C as fromOpenCodeSession, Ct as measuredComparisonFromCandidateExperiment, D as streamCodeAgentJsonlFile, Dt as sealCandidateExperiment, E as parseCodeAgentJsonlFile, Et as sealCandidateBenchmarkTask, F as CodeAgentSessionTerminalStatus, G as PartitionByAuthoringModelResult, H as AgentTraceRange, I as observeCodeAgentSession, J as EvalReportingSuiteInput, K as parseAgentTrace, L as AgentTraceContributor, M as CodeAgentSessionExecutionReceipt, N as CodeAgentSessionObservation, O as CodeAgentSessionAction, Ot as verifyCandidateExperiment, P as CodeAgentSessionSource, Q as FromRunRecordDirOptions, R as AgentTraceContributorType, S as fromKimiCodeSession, St as evaluatePairedMeasurements, T as parseCodeAgentJsonl, Tt as sealCandidateBenchmarkSuite, U as AgentTraceRecord, V as AgentTraceIndex, W as AuthoringProvenance, X as EvalReportingSuiteResult, Y as EvalReportingSuiteOptions, Z as evalReportingSuite, _ as CodeAgentSessionIntakeResult, _t as PairedMeasurement, a as FromFeedbackTableOptions, at as RunAgentProfileImprovementExperimentOptions, b as fromClaudeCodeSession, bt as RunCandidateExperimentOptions, c as CodeAgentStoreObservation, ct as runAgentProfileImprovementExperiment, d as CodeAgentStoreScope, dt as sealAgentProfileImprovementTask, et as RunRecordRejection, f as CodeAgentStoreSession, ft as verifyAgentProfileImprovementExperimentComparison, g as CodeAgentSessionIntakeOptions, gt as EvaluatePairedMeasurementsOptions, h as CodeAgentSessionDiagnostic, ht as CompareCandidateExperimentOptions, i as FeedbackTableRow, it as CompareAgentProfileImprovementExperimentOptions, j as CodeAgentSessionActionSurface, k as CodeAgentSessionActionKind, kt as verifyCandidateExperimentComparison, l as CodeAgentStoreRejection, lt as sealAgentProfileImprovementExperiment, m as CodeAgentJsonlLine, mt as CandidateExperimentRun, n as fromOtelSpans, nt as AgentProfileImprovementExperimentExecutionInput, o as FromFeedbackTableResult, ot as SealAgentProfileImprovementSuiteOptions, p as observeCodeAgentStore, pt as CandidateExperimentExecutionInput, q as partitionRunsByAuthoringModel, r as FeedbackTableMeta, rt as AgentProfileImprovementExperimentRun, s as fromFeedbackTable, st as measuredComparisonFromAgentProfileImprovementExperiment, t as FromOtelSpansOptions, tt as fromRunRecordDir, u as CodeAgentStoreRejections, ut as sealAgentProfileImprovementSuite, v as CodeAgentSessionMetrics, vt as PairedMeasurementAdapter, w as fromPiSession, wt as runCandidateExperiment, x as fromCodexSession, xt as SealCandidateBenchmarkSuiteOptions, y as ParsedCodeAgentJsonl, yt as PairedMeasurementEvaluation, z as AgentTraceConversation } from "../index-D1ZX28rV.js";
-import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-D1rNS2yB.js";
+import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, m as SelfImproveRunError, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "../define-agent-eval-BDPV3ieA.js";
+import { $ as FromRunRecordDirResult, A as CodeAgentSessionActionStatus, B as AgentTraceFile, C as fromOpenCodeSession, Ct as measuredComparisonFromCandidateExperiment, D as streamCodeAgentJsonlFile, Dt as sealCandidateExperiment, E as parseCodeAgentJsonlFile, Et as sealCandidateBenchmarkTask, F as CodeAgentSessionTerminalStatus, G as PartitionByAuthoringModelResult, H as AgentTraceRange, I as observeCodeAgentSession, J as EvalReportingSuiteInput, K as parseAgentTrace, L as AgentTraceContributor, M as CodeAgentSessionExecutionReceipt, N as CodeAgentSessionObservation, O as CodeAgentSessionAction, Ot as verifyCandidateExperiment, P as CodeAgentSessionSource, Q as FromRunRecordDirOptions, R as AgentTraceContributorType, S as fromKimiCodeSession, St as evaluatePairedMeasurements, T as parseCodeAgentJsonl, Tt as sealCandidateBenchmarkSuite, U as AgentTraceRecord, V as AgentTraceIndex, W as AuthoringProvenance, X as EvalReportingSuiteResult, Y as EvalReportingSuiteOptions, Z as evalReportingSuite, _ as CodeAgentSessionIntakeResult, _t as PairedMeasurement, a as FromFeedbackTableOptions, at as RunAgentProfileImprovementExperimentOptions, b as fromClaudeCodeSession, bt as RunCandidateExperimentOptions, c as CodeAgentStoreObservation, ct as runAgentProfileImprovementExperiment, d as CodeAgentStoreScope, dt as sealAgentProfileImprovementTask, et as RunRecordRejection, f as CodeAgentStoreSession, ft as verifyAgentProfileImprovementExperimentComparison, g as CodeAgentSessionIntakeOptions, gt as EvaluatePairedMeasurementsOptions, h as CodeAgentSessionDiagnostic, ht as CompareCandidateExperimentOptions, i as FeedbackTableRow, it as CompareAgentProfileImprovementExperimentOptions, j as CodeAgentSessionActionSurface, k as CodeAgentSessionActionKind, kt as verifyCandidateExperimentComparison, l as CodeAgentStoreRejection, lt as sealAgentProfileImprovementExperiment, m as CodeAgentJsonlLine, mt as CandidateExperimentRun, n as fromOtelSpans, nt as AgentProfileImprovementExperimentExecutionInput, o as FromFeedbackTableResult, ot as SealAgentProfileImprovementSuiteOptions, p as observeCodeAgentStore, pt as CandidateExperimentExecutionInput, q as partitionRunsByAuthoringModel, r as FeedbackTableMeta, rt as AgentProfileImprovementExperimentRun, s as fromFeedbackTable, st as measuredComparisonFromAgentProfileImprovementExperiment, t as FromOtelSpansOptions, tt as fromRunRecordDir, u as CodeAgentStoreRejections, ut as sealAgentProfileImprovementSuite, v as CodeAgentSessionMetrics, vt as PairedMeasurementAdapter, w as fromPiSession, wt as runCandidateExperiment, x as fromCodexSession, xt as SealCandidateBenchmarkSuiteOptions, y as ParsedCodeAgentJsonl, yt as PairedMeasurementEvaluation, z as AgentTraceConversation } from "../index-CnKgaHKA.js";
+import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-DesnDBvl.js";
 export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentJsonlLine, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeAgentStoreObservation, type CodeAgentStoreRejection, type CodeAgentStoreRejections, type CodeAgentStoreScope, type CodeAgentStoreSession, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FinalEvidencePolicy, type FinalEvidenceUse, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, observeCodeAgentStore, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, transientDispatchFailure, verifyAgentProfileImprovementExperimentComparison, verifyCandidateExperiment, verifyCandidateExperimentComparison };

package/dist/contract/index.js

@@ -11,7 +11,7 @@ import { q as campaignSplitDigest } from "../campaign-evidence-Dt_rWPk2.js";
 import { h as mapConcurrentRange } from "../journal-DvKCvjty.js";
 import { C as inMemoryCampaignStorage, S as fsCampaignStorage } from "../external-optimizer-subprocess-CkyUUV7h.js";
 import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-CJN9dTQw.js";
-import { t as buildDefaultAnalystRegistry } from "../default-registry-D4qXx2aG.js";
+import { t as buildDefaultAnalystRegistry } from "../default-registry-6QOiTzqi.js";
 import { t as extractUsage } from "../extract-usage-DFATEuWW.js";
 import { i as makeProposalFinding } from "../types-CW5b90JH.js";
 import { t as createChatClient } from "../chat-client-DUhWNQQS.js";

package/dist/{default-registry-D4qXx2aG.js → default-registry-6QOiTzqi.js}

@@ -7,7 +7,7 @@ import { LLM_CONTEXT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_OUTPUT_TOKEN_ATTR_KE
 import "./otlp-attributes-D9ptWaIJ.js";
 import { r as makeFinding } from "./types-CW5b90JH.js";
 import { t as executionTrackByLane } from "./execution-tracks-CpgFPpS5.js";
-import { t as analyzeSupervisorRunIntegrity } from "./integrity-uwK7XC-8.js";
+import { t as analyzeSupervisorRunIntegrity } from "./integrity-atFlABhR.js";
 import { randomUUID } from "node:crypto";
 import { z } from "zod";
 //#region src/feedback-trajectory-review.ts
@@ -2366,4 +2366,4 @@ function buildDefaultAnalystRegistry(options = {}) {
 //#endregion
 export { completedAnalystReviewQuality as _, KNOWLEDGE_POISONING_KIND_SPEC as a, validateAnalystReviewDecisions as b, FAILURE_MODE_KIND_SPEC as c, emitControlIntegrityFindings as d, behavioralAnalyst as f, assertUniqueFindingIds as g, analystRunDigest as h, DEFAULT_TRACE_ANALYST_KINDS as i, CONTROL_INTEGRITY_ANALYST as l, analystFindingDigest as m, AnalystRegistry as n, KNOWLEDGE_GAP_KIND_SPEC as o, deriveEfficiencyFindings as p, ExactAnalystRunExecutionError as r, IMPROVEMENT_KIND_SPEC as s, buildDefaultAnalystRegistry as t, ControlIntegrityAnalyst as u, readAnalystReview as v, snapshotAnalystRun as y };
 
-//# sourceMappingURL=default-registry-D4qXx2aG.js.map
+//# sourceMappingURL=default-registry-6QOiTzqi.js.map