@percepta/kaizen 0.8.0 → 0.9.0

package/dist/src/commands/ideas.js

@@ -0,0 +1,133 @@
+import { workspaceRoot } from "../lib/paths.js";
+import { pad } from "../lib/cli.js";
+import { boolFlag, parseFlags, strFlag } from "../lib/parse-args.js";
+import { readEnvFile } from "../../shared/env-file.js";
+import { ISSUE_QUERY, KAIZEN_LINEAR_LABEL, buildIssueFilter, linearHttpMessage, linearProjectRefFromFields, parseLinearBody, resolveLinearEnv, resolveLinearProject } from "../../shared/linear-ideas.js";
+import { loadSystem } from "../lib/system.js";
+import { join } from "node:path";
+//#region src/commands/ideas.ts
+async function runIdeas(argv) {
+	const { flags } = parseFlags(argv);
+	const root = workspaceRoot();
+	const systemId = strFlag(flags, "system");
+	const json = boolFlag(flags, "json");
+	const limit = parseLimit(strFlag(flags, "limit") ?? strFlag(flags, "n"));
+	if (!systemId) {
+		process.stderr.write("kaizen ideas: --system <id> is required\n");
+		return 1;
+	}
+	if (limit === null) {
+		process.stderr.write("kaizen ideas: --limit must be a positive integer\n");
+		return 1;
+	}
+	const system = loadSystem(root, systemId);
+	const projectRef = linearProjectRefFromFields({
+		project: stringFrontmatter(system.frontmatter.linear_project),
+		projectId: stringFrontmatter(system.frontmatter.linear_project_id),
+		projectUrl: stringFrontmatter(system.frontmatter.linear_project_url)
+	});
+	if (!projectRef) {
+		process.stderr.write(`kaizen ideas: system "${systemId}" is missing linear_project in frontmatter.\nAdd the Linear project URL or project ID to scope ideas for this system.
+`);
+		return 1;
+	}
+	const envPath = join(root, ".env.local");
+	const linearEnv = resolveLinearEnv(readEnvFile(envPath), {});
+	const token = linearEnv.apiKey;
+	if (!token) {
+		process.stderr.write(`kaizen ideas: LINEAR_API_KEY is missing. Add it to ${envPath}.\n`);
+		return 1;
+	}
+	const config = {
+		system: systemId,
+		projectRef,
+		team: stringFrontmatter(system.frontmatter.linear_team) ?? linearEnv.teamKey,
+		label: KAIZEN_LINEAR_LABEL,
+		limit
+	};
+	const { project, ideas } = await fetchLinearIdeas(token, config);
+	const outputConfig = {
+		...config,
+		project
+	};
+	if (json) {
+		process.stdout.write(JSON.stringify({
+			config: outputConfig,
+			ideas
+		}, null, 2) + "\n");
+		return 0;
+	}
+	printIdeas(outputConfig, ideas);
+	return 0;
+}
+function parseLimit(raw) {
+	if (!raw) return 50;
+	const n = Number(raw);
+	return Number.isInteger(n) && n > 0 ? n : null;
+}
+async function fetchLinearIdeas(token, config) {
+	const controller = new AbortController();
+	const timeout = setTimeout(() => controller.abort(), 3e4);
+	try {
+		const project = await resolveLinearProject(token, config.projectRef, controller.signal);
+		if (!project) throw new Error(`Linear project ${config.projectRef.value} was not found`);
+		const res = await fetch("https://api.linear.app/graphql", {
+			method: "POST",
+			headers: {
+				Authorization: token,
+				"Content-Type": "application/json"
+			},
+			body: JSON.stringify({
+				query: ISSUE_QUERY,
+				variables: {
+					first: config.limit,
+					filter: buildIssueFilter({
+						projectId: project.id,
+						teamKey: config.team,
+						label: config.label
+					})
+				}
+			}),
+			signal: controller.signal
+		});
+		const body = parseLinearBody(await res.text());
+		if (!res.ok || body.errors?.length) throw new Error(linearHttpMessage(res, body.errors));
+		return {
+			project,
+			ideas: body.data?.issues?.nodes ?? []
+		};
+	} finally {
+		clearTimeout(timeout);
+	}
+}
+function printIdeas(config, ideas) {
+	process.stdout.write(`system: ${config.system}   project: ${config.project.name}   label: ${config.label}\n`);
+	if (config.team) process.stdout.write(`team: ${config.team}\n`);
+	if (ideas.length === 0) {
+		process.stdout.write("no ideas found.\n");
+		return;
+	}
+	process.stdout.write("\n");
+	process.stdout.write(pad("issue", 12) + pad("priority", 10) + pad("status", 18) + pad("assignee", 18) + pad("updated", 13) + "title\n");
+	for (const idea of ideas) process.stdout.write(pad(idea.identifier, 12) + pad(priorityLabel(idea.priority), 10) + pad(idea.state?.name ?? "No status", 18) + pad(idea.assignee?.name ?? "Unassigned", 18) + pad(formatDate(idea.updatedAt), 13) + idea.title + "\n");
+	process.stdout.write("\nUse `kaizen ideas --system " + config.system + " --json` for URLs and full issue details.\n");
+}
+function stringFrontmatter(value) {
+	return typeof value === "string" && value.trim() ? value.trim() : null;
+}
+function priorityLabel(priority) {
+	if (priority === 1) return "Urgent";
+	if (priority === 2) return "High";
+	if (priority === 3) return "Normal";
+	if (priority === 4) return "Low";
+	return "None";
+}
+function formatDate(value) {
+	const date = new Date(value);
+	if (Number.isNaN(date.getTime())) return value.slice(0, 10);
+	return date.toISOString().slice(0, 10);
+}
+//#endregion
+export { runIdeas };
+
+//# sourceMappingURL=ideas.js.map

package/dist/src/commands/ideas.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ideas.js","names":[],"sources":["../../../src/commands/ideas.ts"],"sourcesContent":["import { join } from \"node:path\";\nimport { readEnvFile } from \"../../shared/env-file.js\";\nimport {\n  buildIssueFilter,\n  ISSUE_QUERY,\n  KAIZEN_LINEAR_LABEL,\n  linearProjectRefFromFields,\n  linearHttpMessage,\n  parseLinearBody,\n  resolveLinearEnv,\n  resolveLinearProject,\n  type LinearIdea,\n  type LinearProject,\n  type LinearProjectRef,\n} from \"../../shared/linear-ideas.js\";\nimport { pad } from \"../lib/cli.js\";\nimport { boolFlag, parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport { workspaceRoot } from \"../lib/paths.js\";\nimport { loadSystem } from \"../lib/system.js\";\n\nexport async function runIdeas(argv: string[]): Promise<number> {\n  const { flags } = parseFlags(argv);\n  const root = workspaceRoot();\n  const systemId = strFlag(flags, \"system\");\n  const json = boolFlag(flags, \"json\");\n  const limit = parseLimit(strFlag(flags, \"limit\") ?? strFlag(flags, \"n\"));\n\n  if (!systemId) {\n    process.stderr.write(\"kaizen ideas: --system <id> is required\\n\");\n    return 1;\n  }\n  if (limit === null) {\n    process.stderr.write(\"kaizen ideas: --limit must be a positive integer\\n\");\n    return 1;\n  }\n\n  const system = loadSystem(root, systemId);\n  const projectRef = linearProjectRefFromFields({\n    project: stringFrontmatter(system.frontmatter.linear_project),\n    projectId: stringFrontmatter(system.frontmatter.linear_project_id),\n    projectUrl: stringFrontmatter(system.frontmatter.linear_project_url),\n  });\n  if (!projectRef) {\n    process.stderr.write(\n      `kaizen ideas: system \"${systemId}\" is missing linear_project in frontmatter.\\n` +\n        \"Add the Linear project URL or project ID to scope ideas for this system.\\n\",\n    );\n    return 1;\n  }\n\n  const envPath = join(root, \".env.local\");\n  const linearEnv = resolveLinearEnv(readEnvFile(envPath), {});\n  const token = linearEnv.apiKey;\n  if (!token) {\n    process.stderr.write(\n      `kaizen ideas: LINEAR_API_KEY is missing. Add it to ${envPath}.\\n`,\n    );\n    return 1;\n  }\n\n  const config = {\n    system: systemId,\n    projectRef,\n    team:\n      stringFrontmatter(system.frontmatter.linear_team) ?? linearEnv.teamKey,\n    label: KAIZEN_LINEAR_LABEL,\n    limit,\n  };\n\n  const { project, ideas } = await fetchLinearIdeas(token, config);\n  const outputConfig = {\n    ...config,\n    project,\n  };\n\n  if (json) {\n    process.stdout.write(\n      JSON.stringify({ config: outputConfig, ideas }, null, 2) + \"\\n\",\n    );\n    return 0;\n  }\n\n  printIdeas(outputConfig, ideas);\n  return 0;\n}\n\nfunction parseLimit(raw: string | undefined): number | null {\n  if (!raw) return 50;\n  const n = Number(raw);\n  return Number.isInteger(n) && n > 0 ? n : null;\n}\n\nasync function fetchLinearIdeas(\n  token: string,\n  config: {\n    projectRef: LinearProjectRef;\n    team: string | null;\n    label: string;\n    limit: number;\n  },\n): Promise<{ project: LinearProject; ideas: LinearIdea[] }> {\n  const controller = new AbortController();\n  const timeout = setTimeout(() => controller.abort(), 30_000);\n  try {\n    const project = await resolveLinearProject(\n      token,\n      config.projectRef,\n      controller.signal,\n    );\n    if (!project) {\n      throw new Error(\n        `Linear project ${config.projectRef.value} was not found`,\n      );\n    }\n\n    const res = await fetch(\"https://api.linear.app/graphql\", {\n      method: \"POST\",\n      headers: {\n        Authorization: token,\n        \"Content-Type\": \"application/json\",\n      },\n      body: JSON.stringify({\n        query: ISSUE_QUERY,\n        variables: {\n          first: config.limit,\n          filter: buildIssueFilter({\n            projectId: project.id,\n            teamKey: config.team,\n            label: config.label,\n          }),\n        },\n      }),\n      signal: controller.signal,\n    });\n\n    const text = await res.text();\n    const body = parseLinearBody(text);\n    if (!res.ok || body.errors?.length) {\n      throw new Error(linearHttpMessage(res, body.errors));\n    }\n    return { project, ideas: body.data?.issues?.nodes ?? [] };\n  } finally {\n    clearTimeout(timeout);\n  }\n}\n\nfunction printIdeas(\n  config: {\n    system: string;\n    project: LinearProject;\n    team: string | null;\n    label: string;\n  },\n  ideas: LinearIdea[],\n): void {\n  process.stdout.write(\n    `system: ${config.system}   project: ${config.project.name}   label: ${config.label}\\n`,\n  );\n  if (config.team) process.stdout.write(`team: ${config.team}\\n`);\n  if (ideas.length === 0) {\n    process.stdout.write(\"no ideas found.\\n\");\n    return;\n  }\n\n  process.stdout.write(\"\\n\");\n  process.stdout.write(\n    pad(\"issue\", 12) +\n      pad(\"priority\", 10) +\n      pad(\"status\", 18) +\n      pad(\"assignee\", 18) +\n      pad(\"updated\", 13) +\n      \"title\\n\",\n  );\n  for (const idea of ideas) {\n    process.stdout.write(\n      pad(idea.identifier, 12) +\n        pad(priorityLabel(idea.priority), 10) +\n        pad(idea.state?.name ?? \"No status\", 18) +\n        pad(idea.assignee?.name ?? \"Unassigned\", 18) +\n        pad(formatDate(idea.updatedAt), 13) +\n        idea.title +\n        \"\\n\",\n    );\n  }\n  process.stdout.write(\n    \"\\nUse `kaizen ideas --system \" +\n      config.system +\n      \" --json` for URLs and full issue details.\\n\",\n  );\n}\n\nfunction stringFrontmatter(value: unknown): string | null {\n  return typeof value === \"string\" && value.trim() ? value.trim() : null;\n}\n\nfunction priorityLabel(priority: number): string {\n  if (priority === 1) return \"Urgent\";\n  if (priority === 2) return \"High\";\n  if (priority === 3) return \"Normal\";\n  if (priority === 4) return \"Low\";\n  return \"None\";\n}\n\nfunction formatDate(value: string): string {\n  const date = new Date(value);\n  if (Number.isNaN(date.getTime())) return value.slice(0, 10);\n  return date.toISOString().slice(0, 10);\n}\n"],"mappings":";;;;;;;;AAoBA,eAAsB,SAAS,MAAiC;CAC9D,MAAM,EAAE,UAAU,WAAW,KAAK;CAClC,MAAM,OAAO,eAAe;CAC5B,MAAM,WAAW,QAAQ,OAAO,SAAS;CACzC,MAAM,OAAO,SAAS,OAAO,OAAO;CACpC,MAAM,QAAQ,WAAW,QAAQ,OAAO,QAAQ,IAAI,QAAQ,OAAO,IAAI,CAAC;AAExE,KAAI,CAAC,UAAU;AACb,UAAQ,OAAO,MAAM,4CAA4C;AACjE,SAAO;;AAET,KAAI,UAAU,MAAM;AAClB,UAAQ,OAAO,MAAM,qDAAqD;AAC1E,SAAO;;CAGT,MAAM,SAAS,WAAW,MAAM,SAAS;CACzC,MAAM,aAAa,2BAA2B;EAC5C,SAAS,kBAAkB,OAAO,YAAY,eAAe;EAC7D,WAAW,kBAAkB,OAAO,YAAY,kBAAkB;EAClE,YAAY,kBAAkB,OAAO,YAAY,mBAAmB;EACrE,CAAC;AACF,KAAI,CAAC,YAAY;AACf,UAAQ,OAAO,MACb,yBAAyB,SAAS;EAEnC;AACD,SAAO;;CAGT,MAAM,UAAU,KAAK,MAAM,aAAa;CACxC,MAAM,YAAY,iBAAiB,YAAY,QAAQ,EAAE,EAAE,CAAC;CAC5D,MAAM,QAAQ,UAAU;AACxB,KAAI,CAAC,OAAO;AACV,UAAQ,OAAO,MACb,sDAAsD,QAAQ,KAC/D;AACD,SAAO;;CAGT,MAAM,SAAS;EACb,QAAQ;EACR;EACA,MACE,kBAAkB,OAAO,YAAY,YAAY,IAAI,UAAU;EACjE,OAAO;EACP;EACD;CAED,MAAM,EAAE,SAAS,UAAU,MAAM,iBAAiB,OAAO,OAAO;CAChE,MAAM,eAAe;EACnB,GAAG;EACH;EACD;AAED,KAAI,MAAM;AACR,UAAQ,OAAO,MACb,KAAK,UAAU;GAAE,QAAQ;GAAc;GAAO,EAAE,MAAM,EAAE,GAAG,KAC5D;AACD,SAAO;;AAGT,YAAW,cAAc,MAAM;AAC/B,QAAO;;AAGT,SAAS,WAAW,KAAwC;AAC1D,KAAI,CAAC,IAAK,QAAO;CACjB,MAAM,IAAI,OAAO,IAAI;AACrB,QAAO,OAAO,UAAU,EAAE,IAAI,IAAI,IAAI,IAAI;;AAG5C,eAAe,iBACb,OACA,QAM0D;CAC1D,MAAM,aAAa,IAAI,iBAAiB;CACxC,MAAM,UAAU,iBAAiB,WAAW,OAAO,EAAE,IAAO;AAC5D,KAAI;EACF,MAAM,UAAU,MAAM,qBACpB,OACA,OAAO,YACP,WAAW,OACZ;AACD,MAAI,CAAC,QACH,OAAM,IAAI,MACR,kBAAkB,OAAO,WAAW,MAAM,gBAC3C;EAGH,MAAM,MAAM,MAAM,MAAM,kCAAkC;GACxD,QAAQ;GACR,SAAS;IACP,eAAe;IACf,gBAAgB;IACjB;GACD,MAAM,KAAK,UAAU;IACnB,OAAO;IACP,WAAW;KACT,OAAO,OAAO;KACd,QAAQ,iBAAiB;MACvB,WAAW,QAAQ;MACnB,SAAS,OAAO;MAChB,OAAO,OAAO;MACf,CAAC;KACH;IACF,CAAC;GACF,QAAQ,WAAW;GACpB,CAAC;EAGF,MAAM,OAAO,gBAAgB,MADV,IAAI,MAAM,CACK;AAClC,MAAI,CAAC,IAAI,MAAM,KAAK,QAAQ,OAC1B,OAAM,IAAI,MAAM,kBAAkB,KAAK,KAAK,OAAO,CAAC;AAEtD,SAAO;GAAE;GAAS,OAAO,KAAK,MAAM,QAAQ,SAAS,EAAE;GAAE;WACjD;AACR,eAAa,QAAQ;;;AAIzB,SAAS,WACP,QAMA,OACM;AACN,SAAQ,OAAO,MACb,WAAW,OAAO,OAAO,cAAc,OAAO,QAAQ,KAAK,YAAY,OAAO,MAAM,IACrF;AACD,KAAI,OAAO,KAAM,SAAQ,OAAO,MAAM,SAAS,OAAO,KAAK,IAAI;AAC/D,KAAI,MAAM,WAAW,GAAG;AACtB,UAAQ,OAAO,MAAM,oBAAoB;AACzC;;AAGF,SAAQ,OAAO,MAAM,KAAK;AAC1B,SAAQ,OAAO,MACb,IAAI,SAAS,GAAG,GACd,IAAI,YAAY,GAAG,GACnB,IAAI,UAAU,GAAG,GACjB,IAAI,YAAY,GAAG,GACnB,IAAI,WAAW,GAAG,GAClB,UACH;AACD,MAAK,MAAM,QAAQ,MACjB,SAAQ,OAAO,MACb,IAAI,KAAK,YAAY,GAAG,GACtB,IAAI,cAAc,KAAK,SAAS,EAAE,GAAG,GACrC,IAAI,KAAK,OAAO,QAAQ,aAAa,GAAG,GACxC,IAAI,KAAK,UAAU,QAAQ,cAAc,GAAG,GAC5C,IAAI,WAAW,KAAK,UAAU,EAAE,GAAG,GACnC,KAAK,QACL,KACH;AAEH,SAAQ,OAAO,MACb,kCACE,OAAO,SACP,8CACH;;AAGH,SAAS,kBAAkB,OAA+B;AACxD,QAAO,OAAO,UAAU,YAAY,MAAM,MAAM,GAAG,MAAM,MAAM,GAAG;;AAGpE,SAAS,cAAc,UAA0B;AAC/C,KAAI,aAAa,EAAG,QAAO;AAC3B,KAAI,aAAa,EAAG,QAAO;AAC3B,KAAI,aAAa,EAAG,QAAO;AAC3B,KAAI,aAAa,EAAG,QAAO;AAC3B,QAAO;;AAGT,SAAS,WAAW,OAAuB;CACzC,MAAM,OAAO,IAAI,KAAK,MAAM;AAC5B,KAAI,OAAO,MAAM,KAAK,SAAS,CAAC,CAAE,QAAO,MAAM,MAAM,GAAG,GAAG;AAC3D,QAAO,KAAK,aAAa,CAAC,MAAM,GAAG,GAAG"}

package/dist/src/commands/init-system.js

@@ -0,0 +1,86 @@
+import { kaizenSystemDir } from "../../shared/workspace-paths.js";
+import { templatesDir, workspaceRoot } from "../lib/paths.js";
+import { isSystemId, requireKaizenWorkspace } from "../lib/cli.js";
+import { applyVars, ensureDir, writeFileSafely } from "../lib/fs-utils.js";
+import { boolFlag, parseFlags, strFlag } from "../lib/parse-args.js";
+import { prompt, promptChoice } from "../lib/prompt.js";
+import { existsSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+//#region src/commands/init-system.ts
+const EVAL_STYLES = [
+	"ground-truth",
+	"llm-as-judge",
+	"hybrid"
+];
+const EVAL_LANGUAGES = ["python", "typescript"];
+async function runInitSystem(argv) {
+	const { positional, flags } = parseFlags(argv);
+	const root = workspaceRoot();
+	const force = boolFlag(flags, "force");
+	if (!requireKaizenWorkspace(root)) return 1;
+	const name = strFlag(flags, "name") ?? positional[0] ?? await prompt("system name (kebab-case)");
+	if (!name || !isSystemId(name)) {
+		process.stderr.write(`invalid system name: "${name}". use kebab-case (e.g. cost-savings).\n`);
+		return 1;
+	}
+	const description = strFlag(flags, "description") ?? await prompt("one-line description", `${name} system`);
+	const evalStyle = strFlag(flags, "eval-style") ?? await promptChoice("eval style", [...EVAL_STYLES], "ground-truth");
+	const evalLanguage = normalizeEvalLanguage(strFlag(flags, "eval-language") ?? strFlag(flags, "language") ?? "python");
+	if (!evalLanguage) {
+		process.stderr.write(`invalid eval language. use one of: ${EVAL_LANGUAGES.join(", ")}, py, ts.\n`);
+		return 1;
+	}
+	const vars = {
+		name,
+		description,
+		eval_style: evalStyle,
+		primary_metric: strFlag(flags, "metric") ?? await prompt("primary metric name", evalStyle === "ground-truth" ? "accuracy" : "score"),
+		target: strFlag(flags, "target") ?? await prompt("target score (0-1)", "0.90"),
+		eval_ext: evalLanguage === "typescript" ? "ts" : "py",
+		eval_file: `eval.${evalLanguage === "typescript" ? "ts" : "py"}`,
+		rubric_frontmatter: evalStyle === "ground-truth" ? "" : `rubric: kaizen/systems/${name}/rubric.md\n`,
+		iso_now: (/* @__PURE__ */ new Date()).toISOString()
+	};
+	const systemDir = kaizenSystemDir(root, name);
+	const systemPath = join(systemDir, "system.md");
+	const evalPath = join(systemDir, vars.eval_file);
+	const rubricPath = join(systemDir, "rubric.md");
+	if (!force) {
+		for (const p of [
+			systemPath,
+			evalPath,
+			...evalStyle !== "ground-truth" ? [rubricPath] : []
+		]) if (existsSync(p)) {
+			process.stderr.write(`refusing to overwrite ${p}. pass --force to replace.\n`);
+			return 1;
+		}
+	}
+	ensureDir(systemDir);
+	const tpl = templatesDir();
+	writeFileSafely(systemPath, applyVars(readFileSync(join(tpl, "system", "system.md"), "utf-8"), vars), { overwrite: force });
+	writeFileSafely(evalPath, applyVars(readFileSync(join(tpl, "system", evalLanguage === "typescript" ? "eval.ts" : "eval.py"), "utf-8"), vars), { overwrite: force });
+	if (evalStyle !== "ground-truth") writeFileSafely(rubricPath, applyVars(readFileSync(join(tpl, "system", "rubric.md"), "utf-8"), vars), { overwrite: force });
+	process.stdout.write(`scaffolded system "${name}":\n`);
+	process.stdout.write(`  kaizen/systems/${name}/system.md\n`);
+	process.stdout.write(`  kaizen/systems/${name}/${vars.eval_file}\n`);
+	if (evalStyle !== "ground-truth") process.stdout.write(`  kaizen/systems/${name}/rubric.md\n`);
+	process.stdout.write([
+		"",
+		"the starter eval emits a working NDJSON event stream against 3 dummy items.",
+		"edit it to load your dataset and produce real scores, then:",
+		"",
+		`  kaizen run --system ${name} --variant baseline --hypothesis "starting baseline"`,
+		""
+	].join("\n"));
+	return 0;
+}
+function normalizeEvalLanguage(raw) {
+	const normalized = raw.toLowerCase();
+	if (normalized === "python" || normalized === "py") return "python";
+	if (normalized === "typescript" || normalized === "ts") return "typescript";
+	return null;
+}
+//#endregion
+export { runInitSystem };
+
+//# sourceMappingURL=init-system.js.map

package/dist/src/commands/init-system.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"init-system.js","names":[],"sources":["../../../src/commands/init-system.ts"],"sourcesContent":["import { existsSync, readFileSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { isSystemId, requireKaizenWorkspace } from \"../lib/cli.js\";\nimport { applyVars, ensureDir, writeFileSafely } from \"../lib/fs-utils.js\";\nimport { boolFlag, parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport { kaizenSystemDir, templatesDir, workspaceRoot } from \"../lib/paths.js\";\nimport { prompt, promptChoice } from \"../lib/prompt.js\";\n\nconst EVAL_STYLES = [\"ground-truth\", \"llm-as-judge\", \"hybrid\"] as const;\ntype EvalStyle = (typeof EVAL_STYLES)[number];\nconst EVAL_LANGUAGES = [\"python\", \"typescript\"] as const;\ntype EvalLanguage = (typeof EVAL_LANGUAGES)[number];\n\nexport async function runInitSystem(argv: string[]): Promise<number> {\n  const { positional, flags } = parseFlags(argv);\n  const root = workspaceRoot();\n  const force = boolFlag(flags, \"force\");\n\n  if (!requireKaizenWorkspace(root)) return 1;\n\n  const name =\n    strFlag(flags, \"name\") ??\n    positional[0] ??\n    (await prompt(\"system name (kebab-case)\"));\n  if (!name || !isSystemId(name)) {\n    process.stderr.write(\n      `invalid system name: \"${name}\". use kebab-case (e.g. cost-savings).\\n`,\n    );\n    return 1;\n  }\n\n  const description =\n    strFlag(flags, \"description\") ??\n    (await prompt(\"one-line description\", `${name} system`));\n\n  const evalStyle = (strFlag(flags, \"eval-style\") ??\n    (await promptChoice(\n      \"eval style\",\n      [...EVAL_STYLES],\n      \"ground-truth\",\n    ))) as EvalStyle;\n\n  const evalLanguage = normalizeEvalLanguage(\n    strFlag(flags, \"eval-language\") ?? strFlag(flags, \"language\") ?? \"python\",\n  );\n  if (!evalLanguage) {\n    process.stderr.write(\n      `invalid eval language. use one of: ${EVAL_LANGUAGES.join(\", \")}, py, ts.\\n`,\n    );\n    return 1;\n  }\n\n  const primaryMetric =\n    strFlag(flags, \"metric\") ??\n    (await prompt(\n      \"primary metric name\",\n      evalStyle === \"ground-truth\" ? \"accuracy\" : \"score\",\n    ));\n\n  const target =\n    strFlag(flags, \"target\") ?? (await prompt(\"target score (0-1)\", \"0.90\"));\n\n  const vars: Record<string, string> = {\n    name,\n    description,\n    eval_style: evalStyle,\n    primary_metric: primaryMetric,\n    target,\n    eval_ext: evalLanguage === \"typescript\" ? \"ts\" : \"py\",\n    eval_file: `eval.${evalLanguage === \"typescript\" ? \"ts\" : \"py\"}`,\n    rubric_frontmatter:\n      evalStyle === \"ground-truth\"\n        ? \"\"\n        : `rubric: kaizen/systems/${name}/rubric.md\\n`,\n    iso_now: new Date().toISOString(),\n  };\n\n  const systemDir = kaizenSystemDir(root, name);\n  const systemPath = join(systemDir, \"system.md\");\n  const evalPath = join(systemDir, vars.eval_file);\n  const rubricPath = join(systemDir, \"rubric.md\");\n\n  if (!force) {\n    for (const p of [\n      systemPath,\n      evalPath,\n      ...(evalStyle !== \"ground-truth\" ? [rubricPath] : []),\n    ]) {\n      if (existsSync(p)) {\n        process.stderr.write(\n          `refusing to overwrite ${p}. pass --force to replace.\\n`,\n        );\n        return 1;\n      }\n    }\n  }\n\n  ensureDir(systemDir);\n\n  const tpl = templatesDir();\n  writeFileSafely(\n    systemPath,\n    applyVars(readFileSync(join(tpl, \"system\", \"system.md\"), \"utf-8\"), vars),\n    { overwrite: force },\n  );\n  writeFileSafely(\n    evalPath,\n    applyVars(\n      readFileSync(\n        join(\n          tpl,\n          \"system\",\n          evalLanguage === \"typescript\" ? \"eval.ts\" : \"eval.py\",\n        ),\n        \"utf-8\",\n      ),\n      vars,\n    ),\n    { overwrite: force },\n  );\n\n  if (evalStyle !== \"ground-truth\") {\n    writeFileSafely(\n      rubricPath,\n      applyVars(readFileSync(join(tpl, \"system\", \"rubric.md\"), \"utf-8\"), vars),\n      { overwrite: force },\n    );\n  }\n\n  process.stdout.write(`scaffolded system \"${name}\":\\n`);\n  process.stdout.write(`  kaizen/systems/${name}/system.md\\n`);\n  process.stdout.write(`  kaizen/systems/${name}/${vars.eval_file}\\n`);\n  if (evalStyle !== \"ground-truth\")\n    process.stdout.write(`  kaizen/systems/${name}/rubric.md\\n`);\n  process.stdout.write(\n    [\n      \"\",\n      \"the starter eval emits a working NDJSON event stream against 3 dummy items.\",\n      \"edit it to load your dataset and produce real scores, then:\",\n      \"\",\n      `  kaizen run --system ${name} --variant baseline --hypothesis \"starting baseline\"`,\n      \"\",\n    ].join(\"\\n\"),\n  );\n\n  return 0;\n}\n\nfunction normalizeEvalLanguage(raw: string): EvalLanguage | null {\n  const normalized = raw.toLowerCase();\n  if (normalized === \"python\" || normalized === \"py\") return \"python\";\n  if (normalized === \"typescript\" || normalized === \"ts\") return \"typescript\";\n  return null;\n}\n"],"mappings":";;;;;;;;;AAQA,MAAM,cAAc;CAAC;CAAgB;CAAgB;CAAS;AAE9D,MAAM,iBAAiB,CAAC,UAAU,aAAa;AAG/C,eAAsB,cAAc,MAAiC;CACnE,MAAM,EAAE,YAAY,UAAU,WAAW,KAAK;CAC9C,MAAM,OAAO,eAAe;CAC5B,MAAM,QAAQ,SAAS,OAAO,QAAQ;AAEtC,KAAI,CAAC,uBAAuB,KAAK,CAAE,QAAO;CAE1C,MAAM,OACJ,QAAQ,OAAO,OAAO,IACtB,WAAW,MACV,MAAM,OAAO,2BAA2B;AAC3C,KAAI,CAAC,QAAQ,CAAC,WAAW,KAAK,EAAE;AAC9B,UAAQ,OAAO,MACb,yBAAyB,KAAK,0CAC/B;AACD,SAAO;;CAGT,MAAM,cACJ,QAAQ,OAAO,cAAc,IAC5B,MAAM,OAAO,wBAAwB,GAAG,KAAK,SAAS;CAEzD,MAAM,YAAa,QAAQ,OAAO,aAAa,IAC5C,MAAM,aACL,cACA,CAAC,GAAG,YAAY,EAChB,eACD;CAEH,MAAM,eAAe,sBACnB,QAAQ,OAAO,gBAAgB,IAAI,QAAQ,OAAO,WAAW,IAAI,SAClE;AACD,KAAI,CAAC,cAAc;AACjB,UAAQ,OAAO,MACb,sCAAsC,eAAe,KAAK,KAAK,CAAC,aACjE;AACD,SAAO;;CAaT,MAAM,OAA+B;EACnC;EACA;EACA,YAAY;EACZ,gBAbA,QAAQ,OAAO,SAAS,IACvB,MAAM,OACL,uBACA,cAAc,iBAAiB,aAAa,QAC7C;EAUD,QAPA,QAAQ,OAAO,SAAS,IAAK,MAAM,OAAO,sBAAsB,OAAO;EAQvE,UAAU,iBAAiB,eAAe,OAAO;EACjD,WAAW,QAAQ,iBAAiB,eAAe,OAAO;EAC1D,oBACE,cAAc,iBACV,KACA,0BAA0B,KAAK;EACrC,0BAAS,IAAI,MAAM,EAAC,aAAa;EAClC;CAED,MAAM,YAAY,gBAAgB,MAAM,KAAK;CAC7C,MAAM,aAAa,KAAK,WAAW,YAAY;CAC/C,MAAM,WAAW,KAAK,WAAW,KAAK,UAAU;CAChD,MAAM,aAAa,KAAK,WAAW,YAAY;AAE/C,KAAI,CAAC;OACE,MAAM,KAAK;GACd;GACA;GACA,GAAI,cAAc,iBAAiB,CAAC,WAAW,GAAG,EAAE;GACrD,CACC,KAAI,WAAW,EAAE,EAAE;AACjB,WAAQ,OAAO,MACb,yBAAyB,EAAE,8BAC5B;AACD,UAAO;;;AAKb,WAAU,UAAU;CAEpB,MAAM,MAAM,cAAc;AAC1B,iBACE,YACA,UAAU,aAAa,KAAK,KAAK,UAAU,YAAY,EAAE,QAAQ,EAAE,KAAK,EACxE,EAAE,WAAW,OAAO,CACrB;AACD,iBACE,UACA,UACE,aACE,KACE,KACA,UACA,iBAAiB,eAAe,YAAY,UAC7C,EACD,QACD,EACD,KACD,EACD,EAAE,WAAW,OAAO,CACrB;AAED,KAAI,cAAc,eAChB,iBACE,YACA,UAAU,aAAa,KAAK,KAAK,UAAU,YAAY,EAAE,QAAQ,EAAE,KAAK,EACxE,EAAE,WAAW,OAAO,CACrB;AAGH,SAAQ,OAAO,MAAM,sBAAsB,KAAK,MAAM;AACtD,SAAQ,OAAO,MAAM,oBAAoB,KAAK,cAAc;AAC5D,SAAQ,OAAO,MAAM,oBAAoB,KAAK,GAAG,KAAK,UAAU,IAAI;AACpE,KAAI,cAAc,eAChB,SAAQ,OAAO,MAAM,oBAAoB,KAAK,cAAc;AAC9D,SAAQ,OAAO,MACb;EACE;EACA;EACA;EACA;EACA,yBAAyB,KAAK;EAC9B;EACD,CAAC,KAAK,KAAK,CACb;AAED,QAAO;;AAGT,SAAS,sBAAsB,KAAkC;CAC/D,MAAM,aAAa,IAAI,aAAa;AACpC,KAAI,eAAe,YAAY,eAAe,KAAM,QAAO;AAC3D,KAAI,eAAe,gBAAgB,eAAe,KAAM,QAAO;AAC/D,QAAO"}

package/dist/src/commands/init.js

@@ -0,0 +1,78 @@
+import { kaizenConfigPath, kaizenDir, kaizenSystemsDir } from "../../shared/workspace-paths.js";
+import { resolveStateDir, workspaceRoot } from "../lib/paths.js";
+import { appendIfMissing, ensureDir, writeFileSafely } from "../lib/fs-utils.js";
+import { boolFlag, parseFlags, strFlag } from "../lib/parse-args.js";
+import { prompt } from "../lib/prompt.js";
+import { existsSync } from "node:fs";
+import { join } from "node:path";
+//#region src/commands/init.ts
+async function runInit(argv) {
+	const { positional, flags } = parseFlags(argv);
+	if (positional.length > 0) {
+		process.stderr.write(`kaizen init does not take positional arguments. Run \`kaizen create system <name>\` to add a system.\n`);
+		return 1;
+	}
+	const force = boolFlag(flags, "force");
+	const root = workspaceRoot();
+	const configPath = kaizenConfigPath(root);
+	const stateDir = resolveStateDir(root);
+	if (existsSync(configPath) && !force) {
+		process.stdout.write(`kaizen/config.ts already exists at ${configPath}\nre-run with --force to overwrite, or run \`kaizen create system <name>\` to add a system.\n`);
+		return 1;
+	}
+	process.stdout.write("kaizen init — scaffolding a workspace\n\n");
+	const vars = { customer_name: strFlag(flags, "customer-name") ?? strFlag(flags, "customer") ?? await prompt("customer display name", inferCustomerName(root)) };
+	ensureDir(kaizenDir(root));
+	writeFileSafely(configPath, renderConfig(vars), { overwrite: force });
+	ensureDir(kaizenSystemsDir(root));
+	writeFileSafely(join(kaizenSystemsDir(root), ".gitkeep"), "");
+	ensureDir(stateDir);
+	writeFileSafely(join(stateDir, ".gitignore"), [
+		"# Kaizen-managed run state. Runs are large and per-machine; commit only the durable summaries.",
+		"runs/",
+		"dist/",
+		"!hypotheses/",
+		""
+	].join("\n"));
+	appendIfMissing(join(root, ".gitignore"), "# Kaizen");
+	appendIfMissing(join(root, ".gitignore"), "kaizen/.kaizen/runs/");
+	appendIfMissing(join(root, ".gitignore"), "kaizen/.kaizen/dist/");
+	process.stdout.write("\nscaffolded:\n");
+	process.stdout.write(`  ${rel(root, configPath)}\n`);
+	process.stdout.write(`  kaizen/systems/\n`);
+	process.stdout.write(`  ${rel(root, join(stateDir, ".gitignore"))}\n`);
+	process.stdout.write(`  package-owned agent guide available via \`kaizen guide\`\n`);
+	process.stdout.write([
+		"",
+		"next:",
+		"  kaizen guide                       # print agent instructions from the installed package",
+		"  kaizen create system <name>        # add your first system",
+		"  kaizen create view <name> --type trace",
+		"  kaizen create view <name> --type dataset-item",
+		"  kaizen studio                      # open the dashboard",
+		""
+	].join("\n"));
+	return 0;
+}
+function inferCustomerName(root) {
+	return (root.split("/").pop() ?? "") || "Customer";
+}
+function rel(root, p) {
+	return p.startsWith(root) ? p.slice(root.length + 1) : p;
+}
+function renderConfig(vars) {
+	return `import type { KaizenConfig } from "@percepta/kaizen";
+
+const config: KaizenConfig = {
+  customer: {
+    name: ${JSON.stringify(vars.customer_name)},
+  },
+};
+
+export default config;
+`;
+}
+//#endregion
+export { runInit };
+
+//# sourceMappingURL=init.js.map

package/dist/src/commands/init.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"init.js","names":[],"sources":["../../../src/commands/init.ts"],"sourcesContent":["import { existsSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport {\n  appendIfMissing,\n  ensureDir,\n  writeFileSafely,\n} from \"../lib/fs-utils.js\";\nimport { boolFlag, parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport {\n  kaizenConfigPath,\n  kaizenDir,\n  kaizenSystemsDir,\n  resolveStateDir,\n  workspaceRoot,\n} from \"../lib/paths.js\";\nimport { prompt } from \"../lib/prompt.js\";\n\nexport async function runInit(argv: string[]): Promise<number> {\n  const { positional, flags } = parseFlags(argv);\n  if (positional.length > 0) {\n    process.stderr.write(\n      `kaizen init does not take positional arguments. Run \\`kaizen create system <name>\\` to add a system.\\n`,\n    );\n    return 1;\n  }\n  const force = boolFlag(flags, \"force\");\n  const root = workspaceRoot();\n  const configPath = kaizenConfigPath(root);\n  const stateDir = resolveStateDir(root);\n\n  if (existsSync(configPath) && !force) {\n    process.stdout.write(\n      `kaizen/config.ts already exists at ${configPath}\\n` +\n        `re-run with --force to overwrite, or run \\`kaizen create system <name>\\` to add a system.\\n`,\n    );\n    return 1;\n  }\n\n  process.stdout.write(\"kaizen init — scaffolding a workspace\\n\\n\");\n\n  const customerName =\n    strFlag(flags, \"customer-name\") ??\n    strFlag(flags, \"customer\") ??\n    (await prompt(\"customer display name\", inferCustomerName(root)));\n\n  const vars: Record<string, string> = {\n    customer_name: customerName,\n  };\n\n  // 1) kaizen/config.ts\n  ensureDir(kaizenDir(root));\n  writeFileSafely(configPath, renderConfig(vars), { overwrite: force });\n\n  // 2) directory tree\n  ensureDir(kaizenSystemsDir(root));\n  writeFileSafely(join(kaizenSystemsDir(root), \".gitkeep\"), \"\");\n  ensureDir(stateDir);\n\n  // 3) kaizen/.kaizen/.gitignore — ignore bulky generated state but keep durable summaries.\n  writeFileSafely(\n    join(stateDir, \".gitignore\"),\n    [\n      \"# Kaizen-managed run state. Runs are large and per-machine; commit only the durable summaries.\",\n      \"runs/\",\n      \"dist/\",\n      \"!hypotheses/\",\n      \"\",\n    ].join(\"\\n\"),\n  );\n\n  // 4) project .gitignore additions\n  appendIfMissing(join(root, \".gitignore\"), \"# Kaizen\");\n  appendIfMissing(join(root, \".gitignore\"), \"kaizen/.kaizen/runs/\");\n  appendIfMissing(join(root, \".gitignore\"), \"kaizen/.kaizen/dist/\");\n\n  process.stdout.write(\"\\nscaffolded:\\n\");\n  process.stdout.write(`  ${rel(root, configPath)}\\n`);\n  process.stdout.write(`  kaizen/systems/\\n`);\n  process.stdout.write(`  ${rel(root, join(stateDir, \".gitignore\"))}\\n`);\n  process.stdout.write(\n    `  package-owned agent guide available via \\`kaizen guide\\`\\n`,\n  );\n\n  process.stdout.write(\n    [\n      \"\",\n      \"next:\",\n      \"  kaizen guide                       # print agent instructions from the installed package\",\n      \"  kaizen create system <name>        # add your first system\",\n      \"  kaizen create view <name> --type trace\",\n      \"  kaizen create view <name> --type dataset-item\",\n      \"  kaizen studio                      # open the dashboard\",\n      \"\",\n    ].join(\"\\n\"),\n  );\n\n  return 0;\n}\n\nfunction inferCustomerName(root: string): string {\n  const name = root.split(\"/\").pop() ?? \"\";\n  return name || \"Customer\";\n}\n\nfunction rel(root: string, p: string): string {\n  return p.startsWith(root) ? p.slice(root.length + 1) : p;\n}\n\nfunction renderConfig(vars: Record<string, string>): string {\n  // JSON.stringify escapes quotes/backslashes/newlines correctly so user input\n  // with characters like O'Brien or paths with backslashes don't produce\n  // invalid TypeScript.\n  return `import type { KaizenConfig } from \"@percepta/kaizen\";\n\nconst config: KaizenConfig = {\n  customer: {\n    name: ${JSON.stringify(vars.customer_name)},\n  },\n};\n\nexport default config;\n`;\n}\n"],"mappings":";;;;;;;;AAiBA,eAAsB,QAAQ,MAAiC;CAC7D,MAAM,EAAE,YAAY,UAAU,WAAW,KAAK;AAC9C,KAAI,WAAW,SAAS,GAAG;AACzB,UAAQ,OAAO,MACb,yGACD;AACD,SAAO;;CAET,MAAM,QAAQ,SAAS,OAAO,QAAQ;CACtC,MAAM,OAAO,eAAe;CAC5B,MAAM,aAAa,iBAAiB,KAAK;CACzC,MAAM,WAAW,gBAAgB,KAAK;AAEtC,KAAI,WAAW,WAAW,IAAI,CAAC,OAAO;AACpC,UAAQ,OAAO,MACb,sCAAsC,WAAW,+FAElD;AACD,SAAO;;AAGT,SAAQ,OAAO,MAAM,4CAA4C;CAOjE,MAAM,OAA+B,EACnC,eALA,QAAQ,OAAO,gBAAgB,IAC/B,QAAQ,OAAO,WAAW,IACzB,MAAM,OAAO,yBAAyB,kBAAkB,KAAK,CAAC,EAIhE;AAGD,WAAU,UAAU,KAAK,CAAC;AAC1B,iBAAgB,YAAY,aAAa,KAAK,EAAE,EAAE,WAAW,OAAO,CAAC;AAGrE,WAAU,iBAAiB,KAAK,CAAC;AACjC,iBAAgB,KAAK,iBAAiB,KAAK,EAAE,WAAW,EAAE,GAAG;AAC7D,WAAU,SAAS;AAGnB,iBACE,KAAK,UAAU,aAAa,EAC5B;EACE;EACA;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK,CACb;AAGD,iBAAgB,KAAK,MAAM,aAAa,EAAE,WAAW;AACrD,iBAAgB,KAAK,MAAM,aAAa,EAAE,uBAAuB;AACjE,iBAAgB,KAAK,MAAM,aAAa,EAAE,uBAAuB;AAEjE,SAAQ,OAAO,MAAM,kBAAkB;AACvC,SAAQ,OAAO,MAAM,KAAK,IAAI,MAAM,WAAW,CAAC,IAAI;AACpD,SAAQ,OAAO,MAAM,sBAAsB;AAC3C,SAAQ,OAAO,MAAM,KAAK,IAAI,MAAM,KAAK,UAAU,aAAa,CAAC,CAAC,IAAI;AACtE,SAAQ,OAAO,MACb,+DACD;AAED,SAAQ,OAAO,MACb;EACE;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK,CACb;AAED,QAAO;;AAGT,SAAS,kBAAkB,MAAsB;AAE/C,SADa,KAAK,MAAM,IAAI,CAAC,KAAK,IAAI,OACvB;;AAGjB,SAAS,IAAI,MAAc,GAAmB;AAC5C,QAAO,EAAE,WAAW,KAAK,GAAG,EAAE,MAAM,KAAK,SAAS,EAAE,GAAG;;AAGzD,SAAS,aAAa,MAAsC;AAI1D,QAAO;;;;YAIG,KAAK,UAAU,KAAK,cAAc,CAAC"}

package/dist/src/commands/log.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"log.js","names":[],"sources":["../../../src/commands/log.ts"],"sourcesContent":["import { existsSync, readdirSync, statSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { pad, requireKaizenWorkspace } from \"../lib/cli.js\";\nimport {\n  type RunSummary,\n  currentBaseline,\n  listRuns,\n} from \"../lib/leaderboard.js\";\nimport { boolFlag, parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport { resolveStateDir, workspaceRoot } from \"../lib/paths.js\";\nimport { reapStaleRuns } from \"../lib/runner.js\";\n\nexport async function runLog(argv: string[]): Promise<number> {\n  const { flags } = parseFlags(argv);\n  const root = workspaceRoot();\n  const stateDir = resolveStateDir(root);\n\n  if (!requireKaizenWorkspace(root)) return 1;\n\n  const single = strFlag(flags, \"system\");\n  const n = Number(strFlag(flags, \"n\") ?? \"10\");\n  const json = boolFlag(flags, \"json\");\n\n  const systems = single ? [single] : discoverSystems(stateDir);\n  if (systems.length === 0) {\n    process.stdout.write(\"no systems found.\\n\");\n    return 0;\n  }\n\n  if (json) {\n    const out = systems.map((s) => buildSystemReport(stateDir, s, n));\n    process.stdout.write(JSON.stringify(out, null, 2) + \"\\n\");\n    return 0;\n  }\n\n  for (let i = 0; i < systems.length; i++) {\n    if (i > 0) process.stdout.write(\"\\n\");\n    printSystem(stateDir, systems[i], n);\n  }\n  return 0;\n}\n\nfunction discoverSystems(stateDir: string): string[] {\n  const runsRoot = join(stateDir, \"runs\");\n  if (!existsSync(runsRoot)) return [];\n  return readdirSync(runsRoot).filter((d) => {\n    try {\n      return statSync(join(runsRoot, d)).isDirectory();\n    } catch {\n      return false;\n    }\n  });\n}\n\ninterface SystemReport {\n  system: string;\n  baseline: {\n    run_id: string;\n    variant: string;\n    score: number;\n    n: number | null;\n    eval_version: number;\n    dataset_version: string;\n  } | null;\n  totals: {\n    all: number;\n    complete: number;\n    crashed: number;\n    aborted: number;\n    running: number;\n  };\n  recent: RunSummary[];\n}\n\nfunction buildSystemReport(\n  stateDir: string,\n  systemId: string,\n  n: number,\n): SystemReport {\n  reapStaleRuns(stateDir, systemId);\n  const runs = listRuns(stateDir, systemId);\n  const baseline = pickBaseline(runs);\n  const totals = countByStatus(runs);\n  const recent = runs.slice(0, n);\n  return {\n    system: systemId,\n    baseline: baseline\n      ? {\n          run_id: baseline.run_id,\n          variant: baseline.variant,\n          score: baseline.score!,\n          n: baseline.n,\n          eval_version: baseline.eval_version,\n          dataset_version: baseline.dataset_version,\n        }\n      : null,\n    totals,\n    recent,\n  };\n}\n\nfunction pickBaseline(runs: RunSummary[]): RunSummary | null {\n  // Use the eval/dataset versions of the most recent complete run as the \"current\" anchor.\n  const newest = runs.find((r) => r.status === \"complete\");\n  if (!newest) return null;\n  return currentBaseline(runs, newest.eval_version, newest.dataset_version);\n}\n\nfunction countByStatus(runs: RunSummary[]) {\n  const c = {\n    all: runs.length,\n    complete: 0,\n    crashed: 0,\n    aborted: 0,\n    running: 0,\n  };\n  for (const r of runs) {\n    if (r.status === \"complete\") c.complete++;\n    else if (r.status === \"crashed\") c.crashed++;\n    else if (r.status === \"aborted\") c.aborted++;\n    else if (r.status === \"running\") c.running++;\n  }\n  return c;\n}\n\nfunction printSystem(stateDir: string, systemId: string, n: number): void {\n  const report = buildSystemReport(stateDir, systemId, n);\n  const baseHeader = report.baseline\n    ? `promoted baseline: ${report.baseline.run_id} ${report.baseline.variant} (score ${report.baseline.score.toFixed(4)}, n=${report.baseline.n ?? \"?\"}, eval_v=${report.baseline.eval_version}, dataset=${report.baseline.dataset_version})`\n    : \"promoted baseline: <none>\";\n  process.stdout.write(`system: ${systemId}   ${baseHeader}\\n`);\n  const t = report.totals;\n  process.stdout.write(\n    `total runs: ${t.all}   completed: ${t.complete}   crashed: ${t.crashed}   aborted: ${t.aborted}   running: ${t.running}\\n`,\n  );\n  if (report.recent.length === 0) {\n    process.stdout.write(\"(no runs yet)\\n\");\n    return;\n  }\n  process.stdout.write(\"\\n\");\n  process.stdout.write(\n    pad(\"#\", 4) +\n      pad(\"run_id\", 14) +\n      pad(\"variant\", 22) +\n      pad(\"score\", 8) +\n      pad(\"status\", 11) +\n      pad(\"promoted\", 9) +\n      pad(\"started_at\", 21) +\n      \"hypothesis\\n\",\n  );\n  for (let i = 0; i < report.recent.length; i++) {\n    const r = report.recent[i];\n    const idx = report.totals.all - i;\n    const score = r.score === null ? \"—\" : r.score.toFixed(3);\n    const promoted = r.promoted === null ? \"—\" : r.promoted ? \"yes\" : \"no\";\n    process.stdout.write(\n      pad(String(idx), 4) +\n        pad(r.run_id, 14) +\n        pad(r.variant, 22) +\n        pad(score, 8) +\n        pad(r.status, 11) +\n        pad(promoted, 9) +\n        pad(r.started_at.replace(\"T\", \" \").slice(0, 19), 21) +\n        (r.hypothesis || \"\") +\n        \"\\n\",\n    );\n  }\n}\n"],"mappings":";;;;;;;;AAYA,eAAsB,OAAO,MAAiC;CAC5D,MAAM,EAAE,UAAU,WAAW,KAAK;CAClC,MAAM,OAAO,eAAe;CAC5B,MAAM,WAAW,gBAAgB,KAAK;AAEtC,KAAI,CAAC,uBAAuB,KAAK,CAAE,QAAO;CAE1C,MAAM,SAAS,QAAQ,OAAO,SAAS;CACvC,MAAM,IAAI,OAAO,QAAQ,OAAO,IAAI,IAAI,KAAK;CAC7C,MAAM,OAAO,SAAS,OAAO,OAAO;CAEpC,MAAM,UAAU,SAAS,CAAC,OAAO,GAAG,gBAAgB,SAAS;AAC7D,KAAI,QAAQ,WAAW,GAAG;AACxB,UAAQ,OAAO,MAAM,sBAAsB;AAC3C,SAAO;;AAGT,KAAI,MAAM;EACR,MAAM,MAAM,QAAQ,KAAK,MAAM,kBAAkB,UAAU,GAAG,EAAE,CAAC;AACjE,UAAQ,OAAO,MAAM,KAAK,UAAU,KAAK,MAAM,EAAE,GAAG,KAAK;AACzD,SAAO;;AAGT,MAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK;AACvC,MAAI,IAAI,EAAG,SAAQ,OAAO,MAAM,KAAK;AACrC,cAAY,UAAU,QAAQ,IAAI,EAAE;;AAEtC,QAAO;;AAGT,SAAS,gBAAgB,UAA4B;CACnD,MAAM,WAAW,KAAK,UAAU,OAAO;AACvC,KAAI,CAAC,WAAW,SAAS,CAAE,QAAO,EAAE;AACpC,QAAO,YAAY,SAAS,CAAC,QAAQ,MAAM;AACzC,MAAI;AACF,UAAO,SAAS,KAAK,UAAU,EAAE,CAAC,CAAC,aAAa;UAC1C;AACN,UAAO;;GAET;;AAuBJ,SAAS,kBACP,UACA,UACA,GACc;AACd,eAAc,UAAU,SAAS;CACjC,MAAM,OAAO,SAAS,UAAU,SAAS;CACzC,MAAM,WAAW,aAAa,KAAK;CACnC,MAAM,SAAS,cAAc,KAAK;CAClC,MAAM,SAAS,KAAK,MAAM,GAAG,EAAE;AAC/B,QAAO;EACL,QAAQ;EACR,UAAU,WACN;GACE,QAAQ,SAAS;GACjB,SAAS,SAAS;GAClB,OAAO,SAAS;GAChB,GAAG,SAAS;GACZ,cAAc,SAAS;GACvB,iBAAiB,SAAS;GAC3B,GACD;EACJ;EACA;EACD;;AAGH,SAAS,aAAa,MAAuC;CAE3D,MAAM,SAAS,KAAK,MAAM,MAAM,EAAE,WAAW,WAAW;AACxD,KAAI,CAAC,OAAQ,QAAO;AACpB,QAAO,gBAAgB,MAAM,OAAO,cAAc,OAAO,gBAAgB;;AAG3E,SAAS,cAAc,MAAoB;CACzC,MAAM,IAAI;EACR,KAAK,KAAK;EACV,UAAU;EACV,SAAS;EACT,SAAS;EACT,SAAS;EACV;AACD,MAAK,MAAM,KAAK,KACd,KAAI,EAAE,WAAW,WAAY,GAAE;UACtB,EAAE,WAAW,UAAW,GAAE;UAC1B,EAAE,WAAW,UAAW,GAAE;UAC1B,EAAE,WAAW,UAAW,GAAE;AAErC,QAAO;;AAGT,SAAS,YAAY,UAAkB,UAAkB,GAAiB;CACxE,MAAM,SAAS,kBAAkB,UAAU,UAAU,EAAE;CACvD,MAAM,aAAa,OAAO,WACtB,sBAAsB,OAAO,SAAS,OAAO,GAAG,OAAO,SAAS,QAAQ,UAAU,OAAO,SAAS,MAAM,QAAQ,EAAE,CAAC,MAAM,OAAO,SAAS,KAAK,IAAI,WAAW,OAAO,SAAS,aAAa,YAAY,OAAO,SAAS,gBAAgB,KACtO;AACJ,SAAQ,OAAO,MAAM,WAAW,SAAS,KAAK,WAAW,IAAI;CAC7D,MAAM,IAAI,OAAO;AACjB,SAAQ,OAAO,MACb,eAAe,EAAE,IAAI,gBAAgB,EAAE,SAAS,cAAc,EAAE,QAAQ,cAAc,EAAE,QAAQ,cAAc,EAAE,QAAQ,IACzH;AACD,KAAI,OAAO,OAAO,WAAW,GAAG;AAC9B,UAAQ,OAAO,MAAM,kBAAkB;AACvC;;AAEF,SAAQ,OAAO,MAAM,KAAK;AAC1B,SAAQ,OAAO,MACb,IAAI,KAAK,EAAE,GACT,IAAI,UAAU,GAAG,GACjB,IAAI,WAAW,GAAG,GAClB,IAAI,SAAS,EAAE,GACf,IAAI,UAAU,GAAG,GACjB,IAAI,YAAY,EAAE,GAClB,IAAI,cAAc,GAAG,GACrB,eACH;AACD,MAAK,IAAI,IAAI,GAAG,IAAI,OAAO,OAAO,QAAQ,KAAK;EAC7C,MAAM,IAAI,OAAO,OAAO;EACxB,MAAM,MAAM,OAAO,OAAO,MAAM;EAChC,MAAM,QAAQ,EAAE,UAAU,OAAO,MAAM,EAAE,MAAM,QAAQ,EAAE;EACzD,MAAM,WAAW,EAAE,aAAa,OAAO,MAAM,EAAE,WAAW,QAAQ;AAClE,UAAQ,OAAO,MACb,IAAI,OAAO,IAAI,EAAE,EAAE,GACjB,IAAI,EAAE,QAAQ,GAAG,GACjB,IAAI,EAAE,SAAS,GAAG,GAClB,IAAI,OAAO,EAAE,GACb,IAAI,EAAE,QAAQ,GAAG,GACjB,IAAI,UAAU,EAAE,GAChB,IAAI,EAAE,WAAW,QAAQ,KAAK,IAAI,CAAC,MAAM,GAAG,GAAG,EAAE,GAAG,IACnD,EAAE,cAAc,MACjB,KACH"}

package/dist/src/commands/rebuild.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"rebuild.js","names":[],"sources":["../../../src/commands/rebuild.ts"],"sourcesContent":["import { existsSync, readdirSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { requireKaizenWorkspace } from \"../lib/cli.js\";\nimport { type Event, findTerminal } from \"../lib/events.js\";\nimport type { ManifestFile, StateFile } from \"../lib/leaderboard.js\";\nimport { parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport { resolveStateDir, workspaceRoot } from \"../lib/paths.js\";\nimport {\n  readJsonIfExists,\n  readNdjson,\n  writeJsonAtomic,\n} from \"../lib/run-dir.js\";\n\nexport async function runRebuild(argv: string[]): Promise<number> {\n  const { flags } = parseFlags(argv);\n  const root = workspaceRoot();\n  const stateDir = resolveStateDir(root);\n\n  if (!requireKaizenWorkspace(root)) return 1;\n\n  const single = strFlag(flags, \"system\");\n  const runsRoot = join(stateDir, \"runs\");\n  if (!existsSync(runsRoot)) {\n    process.stdout.write(\"nothing to rebuild (no kaizen/.kaizen/runs/).\\n\");\n    return 0;\n  }\n\n  const systems = single ? [single] : readdirSync(runsRoot);\n  let touched = 0;\n  let scanned = 0;\n  for (const system of systems) {\n    const pdir = join(runsRoot, system);\n    if (!existsSync(pdir)) continue;\n    for (const runId of readdirSync(pdir)) {\n      const dir = join(pdir, runId);\n      const manifest = readJsonIfExists<ManifestFile>(\n        join(dir, \"manifest.json\"),\n      );\n      if (!manifest) continue;\n      scanned++;\n      const events = readNdjson<Event>(join(dir, \"events.jsonl\"));\n      const newState = deriveState(manifest, events);\n      const oldState = readJsonIfExists<StateFile>(join(dir, \"state.json\"));\n      if (JSON.stringify(oldState) === JSON.stringify(newState)) continue;\n      writeJsonAtomic(join(dir, \"state.json\"), newState);\n      touched++;\n    }\n  }\n  process.stdout.write(\n    `rebuild: scanned ${scanned} run(s), updated ${touched} state.json file(s).\\n`,\n  );\n  return 0;\n}\n\nfunction deriveState(\n  manifest: ManifestFile,\n  events: Event[],\n): StateFile & {\n  system: string;\n  variant: string;\n  eval_version: number;\n  dataset_version: string;\n  started_at: string;\n  updated_at: string;\n} {\n  let nTotal: number | null = null;\n  let nDone = 0;\n  let score: number | null = null;\n  let promoted: boolean | null = null;\n  let lastTs: number | null = null;\n  for (const e of events) {\n    if (e.type === \"start\") nTotal = e.n;\n    else if (e.type === \"item\") nDone++;\n    else if (e.type === \"complete\") score = e.score;\n    else if (e.type === \"promotion\")\n      promoted = (e as { promoted?: boolean }).promoted ?? null;\n    if (typeof e.ts === \"number\") lastTs = e.ts;\n  }\n  const terminal = findTerminal(events);\n  let status: StateFile[\"status\"] = \"running\";\n  if (terminal) {\n    if (terminal.type === \"complete\") status = \"complete\";\n    else if (terminal.type === \"crashed\") status = \"crashed\";\n    else if (terminal.type === \"aborted\") status = \"aborted\";\n  }\n  const endedAt =\n    terminal && lastTs ? new Date(lastTs * 1000).toISOString() : null;\n  return {\n    run_id: manifest.run_id,\n    system: manifest.system,\n    variant: manifest.variant,\n    status,\n    score,\n    n_total: nTotal,\n    n_done: nDone,\n    promoted,\n    started_at: manifest.started_at,\n    updated_at: endedAt ?? new Date().toISOString(),\n    ended_at: endedAt,\n    eval_version: manifest.eval_version,\n    dataset_version: manifest.dataset_version,\n  };\n}\n"],"mappings":";;;;;;;;AAaA,eAAsB,WAAW,MAAiC;CAChE,MAAM,EAAE,UAAU,WAAW,KAAK;CAClC,MAAM,OAAO,eAAe;CAC5B,MAAM,WAAW,gBAAgB,KAAK;AAEtC,KAAI,CAAC,uBAAuB,KAAK,CAAE,QAAO;CAE1C,MAAM,SAAS,QAAQ,OAAO,SAAS;CACvC,MAAM,WAAW,KAAK,UAAU,OAAO;AACvC,KAAI,CAAC,WAAW,SAAS,EAAE;AACzB,UAAQ,OAAO,MAAM,kDAAkD;AACvE,SAAO;;CAGT,MAAM,UAAU,SAAS,CAAC,OAAO,GAAG,YAAY,SAAS;CACzD,IAAI,UAAU;CACd,IAAI,UAAU;AACd,MAAK,MAAM,UAAU,SAAS;EAC5B,MAAM,OAAO,KAAK,UAAU,OAAO;AACnC,MAAI,CAAC,WAAW,KAAK,CAAE;AACvB,OAAK,MAAM,SAAS,YAAY,KAAK,EAAE;GACrC,MAAM,MAAM,KAAK,MAAM,MAAM;GAC7B,MAAM,WAAW,iBACf,KAAK,KAAK,gBAAgB,CAC3B;AACD,OAAI,CAAC,SAAU;AACf;GAEA,MAAM,WAAW,YAAY,UADd,WAAkB,KAAK,KAAK,eAAe,CACb,CAAC;GAC9C,MAAM,WAAW,iBAA4B,KAAK,KAAK,aAAa,CAAC;AACrE,OAAI,KAAK,UAAU,SAAS,KAAK,KAAK,UAAU,SAAS,CAAE;AAC3D,mBAAgB,KAAK,KAAK,aAAa,EAAE,SAAS;AAClD;;;AAGJ,SAAQ,OAAO,MACb,oBAAoB,QAAQ,mBAAmB,QAAQ,wBACxD;AACD,QAAO;;AAGT,SAAS,YACP,UACA,QAQA;CACA,IAAI,SAAwB;CAC5B,IAAI,QAAQ;CACZ,IAAI,QAAuB;CAC3B,IAAI,WAA2B;CAC/B,IAAI,SAAwB;AAC5B,MAAK,MAAM,KAAK,QAAQ;AACtB,MAAI,EAAE,SAAS,QAAS,UAAS,EAAE;WAC1B,EAAE,SAAS,OAAQ;WACnB,EAAE,SAAS,WAAY,SAAQ,EAAE;WACjC,EAAE,SAAS,YAClB,YAAY,EAA6B,YAAY;AACvD,MAAI,OAAO,EAAE,OAAO,SAAU,UAAS,EAAE;;CAE3C,MAAM,WAAW,aAAa,OAAO;CACrC,IAAI,SAA8B;AAClC,KAAI;MACE,SAAS,SAAS,WAAY,UAAS;WAClC,SAAS,SAAS,UAAW,UAAS;WACtC,SAAS,SAAS,UAAW,UAAS;;CAEjD,MAAM,UACJ,YAAY,0BAAS,IAAI,KAAK,SAAS,IAAK,EAAC,aAAa,GAAG;AAC/D,QAAO;EACL,QAAQ,SAAS;EACjB,QAAQ,SAAS;EACjB,SAAS,SAAS;EAClB;EACA;EACA,SAAS;EACT,QAAQ;EACR;EACA,YAAY,SAAS;EACrB,YAAY,4BAAW,IAAI,MAAM,EAAC,aAAa;EAC/C,UAAU;EACV,cAAc,SAAS;EACvB,iBAAiB,SAAS;EAC3B"}

package/dist/src/commands/run.js

@@ -0,0 +1,68 @@
+import { workspaceRoot } from "../lib/paths.js";
+import { requireKaizenWorkspace } from "../lib/cli.js";
+import { boolFlag, parseFlags, strFlag } from "../lib/parse-args.js";
+import { runExperiment } from "../lib/runner.js";
+import { normalizeLinearIssue } from "../../shared/linear-issue.js";
+import { readFileSync } from "node:fs";
+import { dirname, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+//#region src/commands/run.ts
+async function runRun(argv) {
+	const { flags } = parseFlags(argv);
+	const root = workspaceRoot();
+	if (!requireKaizenWorkspace(root)) return 1;
+	const system = strFlag(flags, "system");
+	const variant = strFlag(flags, "variant");
+	if (!system) {
+		process.stderr.write("kaizen run: --system <id> is required\n");
+		return 1;
+	}
+	if (!variant) {
+		process.stderr.write("kaizen run: --variant <id> is required\n");
+		return 1;
+	}
+	const parent = strFlag(flags, "parent") ?? null;
+	const hypothesis = strFlag(flags, "hypothesis") ?? "";
+	const diagnostic = boolFlag(flags, "diagnostic");
+	const noAutoPromote = boolFlag(flags, "no-auto-promote");
+	const maxItemsFlag = strFlag(flags, "max-items");
+	const maxItems = maxItemsFlag ? Number(maxItemsFlag) : null;
+	if (maxItems !== null && (!Number.isFinite(maxItems) || maxItems <= 0)) {
+		process.stderr.write(`kaizen run: --max-items must be a positive integer\n`);
+		return 1;
+	}
+	const linearIssueInput = strFlag(flags, "idea") ?? strFlag(flags, "linear-issue") ?? strFlag(flags, "linear-ticket") ?? null;
+	const linearIssue = normalizeLinearIssue(linearIssueInput);
+	if (linearIssueInput && !linearIssue) {
+		process.stderr.write("kaizen run: --idea must be a Linear issue id like KZN-123 or a Linear issue URL\n");
+		return 1;
+	}
+	const result = await runExperiment({
+		workspace: root,
+		systemId: system,
+		variant,
+		parent,
+		hypothesis,
+		diagnostic,
+		noAutoPromote,
+		maxItems,
+		kaizenVersion: getKaizenVersion(),
+		linearIssue
+	});
+	const score = result.score === null ? "null" : result.score.toFixed(4);
+	const promoted = result.promoted === null ? "null" : String(result.promoted);
+	process.stdout.write(`score=${score} run_id=${result.runId} status=${result.status} promoted=${promoted}\n`);
+	return result.exitCode;
+}
+function getKaizenVersion() {
+	try {
+		const pkgPath = resolve(dirname(fileURLToPath(import.meta.url)), "..", "..", "package.json");
+		return JSON.parse(readFileSync(pkgPath, "utf-8")).version ?? "0.0.0-unknown";
+	} catch {
+		return "0.0.0-unknown";
+	}
+}
+//#endregion
+export { runRun };
+
+//# sourceMappingURL=run.js.map

package/dist/src/commands/run.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run.js","names":[],"sources":["../../../src/commands/run.ts"],"sourcesContent":["import { readFileSync } from \"node:fs\";\nimport { dirname, resolve } from \"node:path\";\nimport { fileURLToPath } from \"node:url\";\nimport { normalizeLinearIssue } from \"../../shared/linear-issue.js\";\nimport { requireKaizenWorkspace } from \"../lib/cli.js\";\nimport { boolFlag, parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport { workspaceRoot } from \"../lib/paths.js\";\nimport { runExperiment } from \"../lib/runner.js\";\n\nexport async function runRun(argv: string[]): Promise<number> {\n  const { flags } = parseFlags(argv);\n  const root = workspaceRoot();\n\n  if (!requireKaizenWorkspace(root)) return 1;\n\n  const system = strFlag(flags, \"system\");\n  const variant = strFlag(flags, \"variant\");\n  if (!system) {\n    process.stderr.write(\"kaizen run: --system <id> is required\\n\");\n    return 1;\n  }\n  if (!variant) {\n    process.stderr.write(\"kaizen run: --variant <id> is required\\n\");\n    return 1;\n  }\n\n  const parent = strFlag(flags, \"parent\") ?? null;\n  const hypothesis = strFlag(flags, \"hypothesis\") ?? \"\";\n  const diagnostic = boolFlag(flags, \"diagnostic\");\n  const noAutoPromote = boolFlag(flags, \"no-auto-promote\");\n  const maxItemsFlag = strFlag(flags, \"max-items\");\n  const maxItems = maxItemsFlag ? Number(maxItemsFlag) : null;\n  if (maxItems !== null && (!Number.isFinite(maxItems) || maxItems <= 0)) {\n    process.stderr.write(\n      `kaizen run: --max-items must be a positive integer\\n`,\n    );\n    return 1;\n  }\n  const linearIssueInput =\n    strFlag(flags, \"idea\") ??\n    strFlag(flags, \"linear-issue\") ??\n    strFlag(flags, \"linear-ticket\") ??\n    null;\n  const linearIssue = normalizeLinearIssue(linearIssueInput);\n  if (linearIssueInput && !linearIssue) {\n    process.stderr.write(\n      \"kaizen run: --idea must be a Linear issue id like KZN-123 or a Linear issue URL\\n\",\n    );\n    return 1;\n  }\n\n  const result = await runExperiment({\n    workspace: root,\n    systemId: system,\n    variant,\n    parent,\n    hypothesis,\n    diagnostic,\n    noAutoPromote,\n    maxItems,\n    kaizenVersion: getKaizenVersion(),\n    linearIssue,\n  });\n\n  // Single summary line on stdout — the agent reads this.\n  const score = result.score === null ? \"null\" : result.score.toFixed(4);\n  const promoted = result.promoted === null ? \"null\" : String(result.promoted);\n  process.stdout.write(\n    `score=${score} run_id=${result.runId} status=${result.status} promoted=${promoted}\\n`,\n  );\n  return result.exitCode;\n}\n\nfunction getKaizenVersion(): string {\n  try {\n    const here = dirname(fileURLToPath(import.meta.url));\n    // src/commands -> package root\n    const pkgPath = resolve(here, \"..\", \"..\", \"package.json\");\n    return (\n      JSON.parse(readFileSync(pkgPath, \"utf-8\")).version ?? \"0.0.0-unknown\"\n    );\n  } catch {\n    return \"0.0.0-unknown\";\n  }\n}\n"],"mappings":";;;;;;;;;AASA,eAAsB,OAAO,MAAiC;CAC5D,MAAM,EAAE,UAAU,WAAW,KAAK;CAClC,MAAM,OAAO,eAAe;AAE5B,KAAI,CAAC,uBAAuB,KAAK,CAAE,QAAO;CAE1C,MAAM,SAAS,QAAQ,OAAO,SAAS;CACvC,MAAM,UAAU,QAAQ,OAAO,UAAU;AACzC,KAAI,CAAC,QAAQ;AACX,UAAQ,OAAO,MAAM,0CAA0C;AAC/D,SAAO;;AAET,KAAI,CAAC,SAAS;AACZ,UAAQ,OAAO,MAAM,2CAA2C;AAChE,SAAO;;CAGT,MAAM,SAAS,QAAQ,OAAO,SAAS,IAAI;CAC3C,MAAM,aAAa,QAAQ,OAAO,aAAa,IAAI;CACnD,MAAM,aAAa,SAAS,OAAO,aAAa;CAChD,MAAM,gBAAgB,SAAS,OAAO,kBAAkB;CACxD,MAAM,eAAe,QAAQ,OAAO,YAAY;CAChD,MAAM,WAAW,eAAe,OAAO,aAAa,GAAG;AACvD,KAAI,aAAa,SAAS,CAAC,OAAO,SAAS,SAAS,IAAI,YAAY,IAAI;AACtE,UAAQ,OAAO,MACb,uDACD;AACD,SAAO;;CAET,MAAM,mBACJ,QAAQ,OAAO,OAAO,IACtB,QAAQ,OAAO,eAAe,IAC9B,QAAQ,OAAO,gBAAgB,IAC/B;CACF,MAAM,cAAc,qBAAqB,iBAAiB;AAC1D,KAAI,oBAAoB,CAAC,aAAa;AACpC,UAAQ,OAAO,MACb,oFACD;AACD,SAAO;;CAGT,MAAM,SAAS,MAAM,cAAc;EACjC,WAAW;EACX,UAAU;EACV;EACA;EACA;EACA;EACA;EACA;EACA,eAAe,kBAAkB;EACjC;EACD,CAAC;CAGF,MAAM,QAAQ,OAAO,UAAU,OAAO,SAAS,OAAO,MAAM,QAAQ,EAAE;CACtE,MAAM,WAAW,OAAO,aAAa,OAAO,SAAS,OAAO,OAAO,SAAS;AAC5E,SAAQ,OAAO,MACb,SAAS,MAAM,UAAU,OAAO,MAAM,UAAU,OAAO,OAAO,YAAY,SAAS,IACpF;AACD,QAAO,OAAO;;AAGhB,SAAS,mBAA2B;AAClC,KAAI;EAGF,MAAM,UAAU,QAFH,QAAQ,cAAc,OAAO,KAAK,IAAI,CAEvB,EAAE,MAAM,MAAM,eAAe;AACzD,SACE,KAAK,MAAM,aAAa,SAAS,QAAQ,CAAC,CAAC,WAAW;SAElD;AACN,SAAO"}

package/dist/src/commands/studio.js

@@ -0,0 +1,39 @@
+import { packageRoot, resolveStateDir, workspaceRoot } from "../lib/paths.js";
+import { parseFlags, strFlag } from "../lib/parse-args.js";
+import { existsSync } from "node:fs";
+import { join } from "node:path";
+import { spawn } from "node:child_process";
+//#region src/commands/studio.ts
+async function runStudio(argv) {
+	const { flags } = parseFlags(argv);
+	const port = strFlag(flags, "port") ?? "6789";
+	const root = workspaceRoot();
+	const stateDir = resolveStateDir(root);
+	const serverPath = join(packageRoot(), "dist", "studio", "server.js");
+	if (!existsSync(serverPath)) {
+		process.stderr.write(`kaizen studio: bundled Studio server not found at ${serverPath}.\nBuild the package first with \`pnpm --filter @percepta/kaizen build\`.
+`);
+		return 1;
+	}
+	const env = {
+		...process.env,
+		PORT: port,
+		KAIZEN_WORKSPACE: root,
+		KAIZEN_STATE_DIR: stateDir
+	};
+	const studio = spawn(process.execPath, [serverPath], {
+		cwd: root,
+		env,
+		stdio: "inherit"
+	});
+	const shutdown = () => studio.kill("SIGTERM");
+	process.on("SIGINT", shutdown);
+	process.on("SIGTERM", shutdown);
+	return await new Promise((resolve) => {
+		studio.on("exit", (code) => resolve(code ?? 0));
+	});
+}
+//#endregion
+export { runStudio };
+
+//# sourceMappingURL=studio.js.map

package/dist/src/commands/studio.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"studio.js","names":[],"sources":["../../../src/commands/studio.ts"],"sourcesContent":["import { spawn } from \"node:child_process\";\nimport { existsSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { parseFlags, strFlag } from \"../lib/parse-args.js\";\nimport { packageRoot, resolveStateDir, workspaceRoot } from \"../lib/paths.js\";\n\nexport async function runStudio(argv: string[]): Promise<number> {\n  const { flags } = parseFlags(argv);\n  const port = strFlag(flags, \"port\") ?? \"6789\";\n  const root = workspaceRoot();\n  const stateDir = resolveStateDir(root);\n  const serverPath = join(packageRoot(), \"dist\", \"studio\", \"server.js\");\n\n  if (!existsSync(serverPath)) {\n    process.stderr.write(\n      `kaizen studio: bundled Studio server not found at ${serverPath}.\\n` +\n        \"Build the package first with `pnpm --filter @percepta/kaizen build`.\\n\",\n    );\n    return 1;\n  }\n\n  const env = {\n    ...process.env,\n    PORT: port,\n    KAIZEN_WORKSPACE: root,\n    KAIZEN_STATE_DIR: stateDir,\n  };\n\n  const studio = spawn(process.execPath, [serverPath], {\n    cwd: root,\n    env,\n    stdio: \"inherit\",\n  });\n\n  const shutdown = () => studio.kill(\"SIGTERM\");\n  process.on(\"SIGINT\", shutdown);\n  process.on(\"SIGTERM\", shutdown);\n\n  return await new Promise<number>((resolve) => {\n    studio.on(\"exit\", (code) => resolve(code ?? 0));\n  });\n}\n"],"mappings":";;;;;;AAMA,eAAsB,UAAU,MAAiC;CAC/D,MAAM,EAAE,UAAU,WAAW,KAAK;CAClC,MAAM,OAAO,QAAQ,OAAO,OAAO,IAAI;CACvC,MAAM,OAAO,eAAe;CAC5B,MAAM,WAAW,gBAAgB,KAAK;CACtC,MAAM,aAAa,KAAK,aAAa,EAAE,QAAQ,UAAU,YAAY;AAErE,KAAI,CAAC,WAAW,WAAW,EAAE;AAC3B,UAAQ,OAAO,MACb,qDAAqD,WAAW;EAEjE;AACD,SAAO;;CAGT,MAAM,MAAM;EACV,GAAG,QAAQ;EACX,MAAM;EACN,kBAAkB;EAClB,kBAAkB;EACnB;CAED,MAAM,SAAS,MAAM,QAAQ,UAAU,CAAC,WAAW,EAAE;EACnD,KAAK;EACL;EACA,OAAO;EACR,CAAC;CAEF,MAAM,iBAAiB,OAAO,KAAK,UAAU;AAC7C,SAAQ,GAAG,UAAU,SAAS;AAC9B,SAAQ,GAAG,WAAW,SAAS;AAE/B,QAAO,MAAM,IAAI,SAAiB,YAAY;AAC5C,SAAO,GAAG,SAAS,SAAS,QAAQ,QAAQ,EAAE,CAAC;GAC/C"}

package/dist/src/lib/bootstrap.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"bootstrap.js","names":[],"sources":["../../../src/lib/bootstrap.ts"],"sourcesContent":["// Paired bootstrap on per-item score deltas.\n// Returns a 95% CI on the mean delta. Auto-promotion rule: lower bound > 0.\n//\n// Why bootstrap and not a t-test:\n//   - No normality assumption (judge scores are bounded, often skewed)\n//   - Works on tiny n (we typically have 30 items)\n//   - Two-sided 95% CI is the right shape for the auto-promotion gate\n//   - Trivially implementable in vanilla JS — no scipy / numpy / jStat dep\n\nexport interface BootstrapResult {\n  mean_delta: number;\n  ci_low: number;\n  ci_high: number;\n  n: number;\n}\n\n/**\n * Bootstrap a 95% CI for the mean of `deltas`.\n * Uses 10000 resamples by default; deterministic for a given `seed`.\n */\nexport function bootstrapMeanCI(\n  deltas: number[],\n  opts: { resamples?: number; seed?: number; alpha?: number } = {},\n): BootstrapResult {\n  const resamples = opts.resamples ?? 10000;\n  const alpha = opts.alpha ?? 0.05;\n  const n = deltas.length;\n  if (n === 0) {\n    return { mean_delta: 0, ci_low: 0, ci_high: 0, n: 0 };\n  }\n  const observed = mean(deltas);\n  const rng = mulberry32(opts.seed ?? 0xc0ffee);\n  const means = new Float64Array(resamples);\n  for (let r = 0; r < resamples; r++) {\n    let sum = 0;\n    for (let i = 0; i < n; i++) {\n      sum += deltas[(rng() * n) | 0];\n    }\n    means[r] = sum / n;\n  }\n  const sorted = Array.from(means).sort((a, b) => a - b);\n  const loIdx = Math.floor((alpha / 2) * resamples);\n  const hiIdx = Math.min(\n    resamples - 1,\n    Math.ceil((1 - alpha / 2) * resamples) - 1,\n  );\n  return {\n    mean_delta: observed,\n    ci_low: sorted[loIdx],\n    ci_high: sorted[hiIdx],\n    n,\n  };\n}\n\n/** Pair item scores from new run with baseline scores by id. Returns deltas only for matched ids. */\nexport function pairDeltas(\n  newScores: Map<string, number>,\n  baselineScores: Map<string, number>,\n): { deltas: number[]; matchedIds: string[] } {\n  const deltas: number[] = [];\n  const matchedIds: string[] = [];\n  for (const [id, s] of newScores) {\n    const b = baselineScores.get(id);\n    if (b === undefined) continue;\n    deltas.push(s - b);\n    matchedIds.push(id);\n  }\n  return { deltas, matchedIds };\n}\n\nfunction mean(xs: number[]): number {\n  if (xs.length === 0) return 0;\n  let s = 0;\n  for (const x of xs) s += x;\n  return s / xs.length;\n}\n\n// Small deterministic PRNG (mulberry32). Public domain.\nfunction mulberry32(seed: number): () => number {\n  let a = seed >>> 0;\n  return function () {\n    a = (a + 0x6d2b79f5) >>> 0;\n    let t = a;\n    t = Math.imul(t ^ (t >>> 15), t | 1);\n    t ^= t + Math.imul(t ^ (t >>> 7), t | 61);\n    return ((t ^ (t >>> 14)) >>> 0) / 4294967296;\n  };\n}\n"],"mappings":";;;;;AAoBA,SAAgB,gBACd,QACA,OAA8D,EAAE,EAC/C;CACjB,MAAM,YAAY,KAAK,aAAa;CACpC,MAAM,QAAQ,KAAK,SAAS;CAC5B,MAAM,IAAI,OAAO;AACjB,KAAI,MAAM,EACR,QAAO;EAAE,YAAY;EAAG,QAAQ;EAAG,SAAS;EAAG,GAAG;EAAG;CAEvD,MAAM,WAAW,KAAK,OAAO;CAC7B,MAAM,MAAM,WAAW,KAAK,QAAQ,SAAS;CAC7C,MAAM,QAAQ,IAAI,aAAa,UAAU;AACzC,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,KAAK;EAClC,IAAI,MAAM;AACV,OAAK,IAAI,IAAI,GAAG,IAAI,GAAG,IACrB,QAAO,OAAQ,KAAK,GAAG,IAAK;AAE9B,QAAM,KAAK,MAAM;;CAEnB,MAAM,SAAS,MAAM,KAAK,MAAM,CAAC,MAAM,GAAG,MAAM,IAAI,EAAE;CACtD,MAAM,QAAQ,KAAK,MAAO,QAAQ,IAAK,UAAU;CACjD,MAAM,QAAQ,KAAK,IACjB,YAAY,GACZ,KAAK,MAAM,IAAI,QAAQ,KAAK,UAAU,GAAG,EAC1C;AACD,QAAO;EACL,YAAY;EACZ,QAAQ,OAAO;EACf,SAAS,OAAO;EAChB;EACD;;;AAIH,SAAgB,WACd,WACA,gBAC4C;CAC5C,MAAM,SAAmB,EAAE;CAC3B,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,CAAC,IAAI,MAAM,WAAW;EAC/B,MAAM,IAAI,eAAe,IAAI,GAAG;AAChC,MAAI,MAAM,KAAA,EAAW;AACrB,SAAO,KAAK,IAAI,EAAE;AAClB,aAAW,KAAK,GAAG;;AAErB,QAAO;EAAE;EAAQ;EAAY;;AAG/B,SAAS,KAAK,IAAsB;AAClC,KAAI,GAAG,WAAW,EAAG,QAAO;CAC5B,IAAI,IAAI;AACR,MAAK,MAAM,KAAK,GAAI,MAAK;AACzB,QAAO,IAAI,GAAG;;AAIhB,SAAS,WAAW,MAA4B;CAC9C,IAAI,IAAI,SAAS;AACjB,QAAO,WAAY;AACjB,MAAK,IAAI,eAAgB;EACzB,IAAI,IAAI;AACR,MAAI,KAAK,KAAK,IAAK,MAAM,IAAK,IAAI,EAAE;AACpC,OAAK,IAAI,KAAK,KAAK,IAAK,MAAM,GAAI,IAAI,GAAG;AACzC,WAAS,IAAK,MAAM,QAAS,KAAK"}

package/dist/src/lib/cli.js

@@ -0,0 +1,20 @@
+import { kaizenConfigPath } from "../../shared/workspace-paths.js";
+import "./paths.js";
+import { existsSync } from "node:fs";
+//#region src/lib/cli.ts
+function requireKaizenWorkspace(root) {
+	if (existsSync(kaizenConfigPath(root))) return true;
+	process.stderr.write(`no kaizen/config.ts in ${root}. run \`kaizen init\` first.\n`);
+	return false;
+}
+function isSystemId(value) {
+	return /^[a-z][a-z0-9-]*$/.test(value);
+}
+function pad(s, w) {
+	if (s.length >= w) return s.slice(0, w - 1) + " ";
+	return s + " ".repeat(w - s.length);
+}
+//#endregion
+export { isSystemId, pad, requireKaizenWorkspace };
+
+//# sourceMappingURL=cli.js.map

package/dist/src/lib/cli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.js","names":[],"sources":["../../../src/lib/cli.ts"],"sourcesContent":["import { existsSync } from \"node:fs\";\nimport { kaizenConfigPath } from \"./paths.js\";\n\nexport function requireKaizenWorkspace(root: string): boolean {\n  if (existsSync(kaizenConfigPath(root))) return true;\n  process.stderr.write(\n    `no kaizen/config.ts in ${root}. run \\`kaizen init\\` first.\\n`,\n  );\n  return false;\n}\n\nexport function isSystemId(value: string): boolean {\n  return /^[a-z][a-z0-9-]*$/.test(value);\n}\n\nexport function pad(s: string, w: number): string {\n  if (s.length >= w) return s.slice(0, w - 1) + \" \";\n  return s + \" \".repeat(w - s.length);\n}\n"],"mappings":";;;;AAGA,SAAgB,uBAAuB,MAAuB;AAC5D,KAAI,WAAW,iBAAiB,KAAK,CAAC,CAAE,QAAO;AAC/C,SAAQ,OAAO,MACb,0BAA0B,KAAK,gCAChC;AACD,QAAO;;AAGT,SAAgB,WAAW,OAAwB;AACjD,QAAO,oBAAoB,KAAK,MAAM;;AAGxC,SAAgB,IAAI,GAAW,GAAmB;AAChD,KAAI,EAAE,UAAU,EAAG,QAAO,EAAE,MAAM,GAAG,IAAI,EAAE,GAAG;AAC9C,QAAO,IAAI,IAAI,OAAO,IAAI,EAAE,OAAO"}

package/dist/src/lib/events.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"events.js","names":[],"sources":["../../../src/lib/events.ts"],"sourcesContent":["// Event schema for the NDJSON contract between eval scripts and the runner.\n\ninterface BaseEvent {\n  type: string;\n  ts?: number;\n}\n\ninterface StartEvent extends BaseEvent {\n  type: \"start\";\n  n: number;\n  eval_version: number;\n  dataset_version: string;\n  variant?: string;\n}\n\nexport interface ItemEvent extends BaseEvent {\n  type: \"item\";\n  id: string;\n  score: number;\n  breakdown?: Record<string, number>;\n  trace_id?: string | null;\n  why?: string;\n  subgroup?: Record<string, string>;\n}\n\ninterface ProgressEvent extends BaseEvent {\n  type: \"progress\";\n  done: number;\n  total: number;\n}\n\nexport interface CompleteEvent extends BaseEvent {\n  type: \"complete\";\n  score: number;\n  breakdown?: Record<string, number>;\n  n: number;\n  worst_traces?: Array<{ id: string; score: number; trace_id?: string | null }>;\n}\n\ninterface ErrorEvent extends BaseEvent {\n  type: \"error\";\n  message: string;\n}\n\n// Synthesized by the runner, not emitted by the eval.\nexport interface CrashedEvent extends BaseEvent {\n  type: \"crashed\";\n  exit_code: number | null;\n  signal?: string | null;\n  stderr_tail?: string;\n  reason: string;\n}\n\ninterface AbortedEvent extends BaseEvent {\n  type: \"aborted\";\n  reason: string;\n}\n\nexport interface PromotionEvent extends BaseEvent {\n  type: \"promotion\";\n  promoted: boolean;\n  rule:\n    | \"auto\"\n    | \"force\"\n    | \"no_baseline\"\n    | \"version_mismatch\"\n    | \"stat_insufficient\"\n    | \"subgroup_regression\";\n  ci_low?: number;\n  ci_high?: number;\n  mean_delta?: number;\n  n_compared?: number;\n  parent_run_id?: string;\n  details?: string;\n}\n\nexport type Event =\n  | StartEvent\n  | ItemEvent\n  | ProgressEvent\n  | CompleteEvent\n  | ErrorEvent\n  | CrashedEvent\n  | AbortedEvent\n  | PromotionEvent;\n\nexport function validateEvent(value: unknown): Event {\n  if (!value || typeof value !== \"object\") {\n    throw new Error(\"event must be an object\");\n  }\n  const e = value as Record<string, unknown>;\n  if (typeof e.type !== \"string\") {\n    throw new Error(\"event.type must be a string\");\n  }\n\n  switch (e.type) {\n    case \"start\":\n      requireNumber(e, \"n\");\n      requireNumber(e, \"eval_version\");\n      requireString(e, \"dataset_version\");\n      break;\n    case \"item\":\n      requireString(e, \"id\");\n      requireNumber(e, \"score\");\n      break;\n    case \"progress\":\n      requireNumber(e, \"done\");\n      requireNumber(e, \"total\");\n      break;\n    case \"complete\":\n      requireNumber(e, \"score\");\n      requireNumber(e, \"n\");\n      break;\n    case \"error\":\n      requireString(e, \"message\");\n      break;\n    case \"crashed\":\n      requireString(e, \"reason\");\n      break;\n    case \"aborted\":\n      requireString(e, \"reason\");\n      break;\n    case \"promotion\":\n      if (typeof e.promoted !== \"boolean\")\n        throw new Error(\"promotion.promoted must be boolean\");\n      requireString(e, \"rule\");\n      break;\n    default:\n      throw new Error(`unknown event type: ${e.type}`);\n  }\n\n  if (e.ts !== undefined && typeof e.ts !== \"number\") {\n    throw new Error(`${e.type}.ts must be a number when present`);\n  }\n  return e as unknown as Event;\n}\n\nfunction requireString(e: Record<string, unknown>, key: string): void {\n  if (typeof e[key] !== \"string\" || e[key] === \"\") {\n    throw new Error(`${String(e.type)}.${key} must be a non-empty string`);\n  }\n}\n\nfunction requireNumber(e: Record<string, unknown>, key: string): void {\n  if (typeof e[key] !== \"number\" || !Number.isFinite(e[key])) {\n    throw new Error(`${String(e.type)}.${key} must be a finite number`);\n  }\n}\n\nfunction isTerminal(\n  e: Event,\n): e is CompleteEvent | CrashedEvent | AbortedEvent {\n  return e.type === \"complete\" || e.type === \"crashed\" || e.type === \"aborted\";\n}\n\nexport function findTerminal(\n  events: Event[],\n): CompleteEvent | CrashedEvent | AbortedEvent | null {\n  for (let i = events.length - 1; i >= 0; i--) {\n    if (isTerminal(events[i]))\n      return events[i] as CompleteEvent | CrashedEvent | AbortedEvent;\n  }\n  return null;\n}\n\n/** Stream-mode NDJSON parser. Feed `push(chunk)` from a Readable; receive parsed events via `onEvent`. */\nexport class NdjsonReader {\n  private buffer = \"\";\n  public constructor(\n    private onEvent: (e: Event) => void,\n    private onParseError?: (line: string, err: unknown) => void,\n  ) {}\n\n  public push(chunk: string | Buffer): void {\n    this.buffer += typeof chunk === \"string\" ? chunk : chunk.toString(\"utf-8\");\n    let nl: number;\n    while ((nl = this.buffer.indexOf(\"\\n\")) !== -1) {\n      const line = this.buffer.slice(0, nl);\n      this.buffer = this.buffer.slice(nl + 1);\n      if (!line.trim()) continue;\n      try {\n        this.onEvent(validateEvent(JSON.parse(line)));\n      } catch (err) {\n        this.onParseError?.(line, err);\n      }\n    }\n  }\n\n  /** Flush any remaining unterminated content as a best-effort parse. */\n  public end(): void {\n    const tail = this.buffer.trim();\n    this.buffer = \"\";\n    if (!tail) return;\n    try {\n      this.onEvent(validateEvent(JSON.parse(tail)));\n    } catch (err) {\n      this.onParseError?.(tail, err);\n    }\n  }\n}\n"],"mappings":";AAsFA,SAAgB,cAAc,OAAuB;AACnD,KAAI,CAAC,SAAS,OAAO,UAAU,SAC7B,OAAM,IAAI,MAAM,0BAA0B;CAE5C,MAAM,IAAI;AACV,KAAI,OAAO,EAAE,SAAS,SACpB,OAAM,IAAI,MAAM,8BAA8B;AAGhD,SAAQ,EAAE,MAAV;EACE,KAAK;AACH,iBAAc,GAAG,IAAI;AACrB,iBAAc,GAAG,eAAe;AAChC,iBAAc,GAAG,kBAAkB;AACnC;EACF,KAAK;AACH,iBAAc,GAAG,KAAK;AACtB,iBAAc,GAAG,QAAQ;AACzB;EACF,KAAK;AACH,iBAAc,GAAG,OAAO;AACxB,iBAAc,GAAG,QAAQ;AACzB;EACF,KAAK;AACH,iBAAc,GAAG,QAAQ;AACzB,iBAAc,GAAG,IAAI;AACrB;EACF,KAAK;AACH,iBAAc,GAAG,UAAU;AAC3B;EACF,KAAK;AACH,iBAAc,GAAG,SAAS;AAC1B;EACF,KAAK;AACH,iBAAc,GAAG,SAAS;AAC1B;EACF,KAAK;AACH,OAAI,OAAO,EAAE,aAAa,UACxB,OAAM,IAAI,MAAM,qCAAqC;AACvD,iBAAc,GAAG,OAAO;AACxB;EACF,QACE,OAAM,IAAI,MAAM,uBAAuB,EAAE,OAAO;;AAGpD,KAAI,EAAE,OAAO,KAAA,KAAa,OAAO,EAAE,OAAO,SACxC,OAAM,IAAI,MAAM,GAAG,EAAE,KAAK,mCAAmC;AAE/D,QAAO;;AAGT,SAAS,cAAc,GAA4B,KAAmB;AACpE,KAAI,OAAO,EAAE,SAAS,YAAY,EAAE,SAAS,GAC3C,OAAM,IAAI,MAAM,GAAG,OAAO,EAAE,KAAK,CAAC,GAAG,IAAI,6BAA6B;;AAI1E,SAAS,cAAc,GAA4B,KAAmB;AACpE,KAAI,OAAO,EAAE,SAAS,YAAY,CAAC,OAAO,SAAS,EAAE,KAAK,CACxD,OAAM,IAAI,MAAM,GAAG,OAAO,EAAE,KAAK,CAAC,GAAG,IAAI,0BAA0B;;AAIvE,SAAS,WACP,GACkD;AAClD,QAAO,EAAE,SAAS,cAAc,EAAE,SAAS,aAAa,EAAE,SAAS;;AAGrE,SAAgB,aACd,QACoD;AACpD,MAAK,IAAI,IAAI,OAAO,SAAS,GAAG,KAAK,GAAG,IACtC,KAAI,WAAW,OAAO,GAAG,CACvB,QAAO,OAAO;AAElB,QAAO;;;AAIT,IAAa,eAAb,MAA0B;CACxB,SAAiB;CACjB,YACE,SACA,cACA;AAFQ,OAAA,UAAA;AACA,OAAA,eAAA;;CAGV,KAAY,OAA8B;AACxC,OAAK,UAAU,OAAO,UAAU,WAAW,QAAQ,MAAM,SAAS,QAAQ;EAC1E,IAAI;AACJ,UAAQ,KAAK,KAAK,OAAO,QAAQ,KAAK,MAAM,IAAI;GAC9C,MAAM,OAAO,KAAK,OAAO,MAAM,GAAG,GAAG;AACrC,QAAK,SAAS,KAAK,OAAO,MAAM,KAAK,EAAE;AACvC,OAAI,CAAC,KAAK,MAAM,CAAE;AAClB,OAAI;AACF,SAAK,QAAQ,cAAc,KAAK,MAAM,KAAK,CAAC,CAAC;YACtC,KAAK;AACZ,SAAK,eAAe,MAAM,IAAI;;;;;CAMpC,MAAmB;EACjB,MAAM,OAAO,KAAK,OAAO,MAAM;AAC/B,OAAK,SAAS;AACd,MAAI,CAAC,KAAM;AACX,MAAI;AACF,QAAK,QAAQ,cAAc,KAAK,MAAM,KAAK,CAAC,CAAC;WACtC,KAAK;AACZ,QAAK,eAAe,MAAM,IAAI"}

package/dist/src/lib/fs-utils.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fs-utils.js","names":[],"sources":["../../../src/lib/fs-utils.ts"],"sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { dirname } from \"node:path\";\n\nexport function ensureDir(p: string): void {\n  mkdirSync(p, { recursive: true });\n}\n\nexport function writeFileSafely(\n  p: string,\n  contents: string,\n  opts: { overwrite?: boolean } = {},\n): boolean {\n  if (existsSync(p) && !opts.overwrite) return false;\n  ensureDir(dirname(p));\n  writeFileSync(p, contents);\n  return true;\n}\n\nexport function applyVars(text: string, vars: Record<string, string>): string {\n  return text.replace(/\\{\\{\\s*(\\w+)\\s*\\}\\}/g, (m, key) =>\n    key in vars ? vars[key] : m,\n  );\n}\n\nexport function appendIfMissing(path: string, line: string): boolean {\n  const existing = existsSync(path) ? readFileSync(path, \"utf-8\") : \"\";\n  if (existing.split(\"\\n\").some((l) => l.trim() === line.trim())) return false;\n  ensureDir(dirname(path));\n  const sep = existing.length === 0 || existing.endsWith(\"\\n\") ? \"\" : \"\\n\";\n  writeFileSync(path, existing + sep + line + \"\\n\");\n  return true;\n}\n"],"mappings":";;;AAGA,SAAgB,UAAU,GAAiB;AACzC,WAAU,GAAG,EAAE,WAAW,MAAM,CAAC;;AAGnC,SAAgB,gBACd,GACA,UACA,OAAgC,EAAE,EACzB;AACT,KAAI,WAAW,EAAE,IAAI,CAAC,KAAK,UAAW,QAAO;AAC7C,WAAU,QAAQ,EAAE,CAAC;AACrB,eAAc,GAAG,SAAS;AAC1B,QAAO;;AAGT,SAAgB,UAAU,MAAc,MAAsC;AAC5E,QAAO,KAAK,QAAQ,yBAAyB,GAAG,QAC9C,OAAO,OAAO,KAAK,OAAO,EAC3B;;AAGH,SAAgB,gBAAgB,MAAc,MAAuB;CACnE,MAAM,WAAW,WAAW,KAAK,GAAG,aAAa,MAAM,QAAQ,GAAG;AAClE,KAAI,SAAS,MAAM,KAAK,CAAC,MAAM,MAAM,EAAE,MAAM,KAAK,KAAK,MAAM,CAAC,CAAE,QAAO;AACvE,WAAU,QAAQ,KAAK,CAAC;AAExB,eAAc,MAAM,YADR,SAAS,WAAW,KAAK,SAAS,SAAS,KAAK,GAAG,KAAK,QAC/B,OAAO,KAAK;AACjD,QAAO"}

package/dist/src/lib/leaderboard.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"leaderboard.js","names":[],"sources":["../../../src/lib/leaderboard.ts"],"sourcesContent":["import { existsSync, readdirSync, statSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { systemRunsDir, readJsonIfExists } from \"./run-dir.js\";\n\nexport interface RunSummary {\n  run_id: string;\n  system: string;\n  variant: string;\n  parent_id: string | null;\n  hypothesis: string;\n  status: \"complete\" | \"crashed\" | \"aborted\" | \"running\";\n  score: number | null;\n  n: number | null;\n  promoted: boolean | null;\n  started_at: string;\n  ended_at: string | null;\n  eval_version: number;\n  dataset_version: string;\n  events_path: string;\n  state_path: string;\n  manifest_path: string;\n  linear_issue_id?: string | null;\n  linear_issue_url?: string | null;\n}\n\nexport interface ManifestFile {\n  run_id: string;\n  system: string;\n  variant: string;\n  parent_id: string | null;\n  hypothesis: string;\n  git_sha?: string | null;\n  git_branch?: string | null;\n  worktree_root?: string | null;\n  git_common_dir?: string | null;\n  eval_version: number;\n  dataset_version: string;\n  started_at: string;\n  host?: string;\n  kaizen_version?: string;\n  state_dir?: string;\n  diagnostic?: boolean;\n  linear_issue_id?: string | null;\n  linear_issue_url?: string | null;\n}\n\nexport interface StateFile {\n  run_id: string;\n  status: RunSummary[\"status\"];\n  score: number | null;\n  n_total?: number | null;\n  n_done?: number;\n  promoted?: boolean | null;\n  ended_at?: string | null;\n}\n\nexport function listRuns(stateDir: string, systemId: string): RunSummary[] {\n  const dir = systemRunsDir(stateDir, systemId);\n  if (!existsSync(dir)) return [];\n  const out: RunSummary[] = [];\n  for (const entry of readdirSync(dir)) {\n    const runDir = join(dir, entry);\n    if (!statSync(runDir).isDirectory()) continue;\n    const manifest_path = join(runDir, \"manifest.json\");\n    const state_path = join(runDir, \"state.json\");\n    const events_path = join(runDir, \"events.jsonl\");\n    const manifest = readJsonIfExists<ManifestFile>(manifest_path);\n    if (!manifest) continue;\n    const state = readJsonIfExists<StateFile>(state_path);\n    out.push({\n      run_id: manifest.run_id,\n      system: manifest.system,\n      variant: manifest.variant,\n      parent_id: manifest.parent_id,\n      hypothesis: manifest.hypothesis,\n      status: state?.status ?? \"running\",\n      score: state?.score ?? null,\n      n: state?.n_total ?? null,\n      promoted: state?.promoted ?? null,\n      started_at: manifest.started_at,\n      ended_at: state?.ended_at ?? null,\n      eval_version: manifest.eval_version,\n      dataset_version: manifest.dataset_version,\n      events_path,\n      state_path,\n      manifest_path,\n      linear_issue_id: manifest.linear_issue_id ?? null,\n      linear_issue_url: manifest.linear_issue_url ?? null,\n    });\n  }\n  // Newest first\n  out.sort((a, b) => (b.started_at < a.started_at ? -1 : 1));\n  return out;\n}\n\n/**\n * Promoted baseline = the most recent run that was promoted, under matching versions.\n * Falls back to the highest-scored complete run if nothing has been promoted yet\n * (e.g. the very first run before promotion semantics kick in).\n */\nexport function currentBaseline(\n  runs: RunSummary[],\n  evalVersion: number,\n  datasetVersion: string,\n): RunSummary | null {\n  const eligible = runs.filter(\n    (r) =>\n      r.status === \"complete\" &&\n      r.score !== null &&\n      r.eval_version === evalVersion &&\n      r.dataset_version === datasetVersion,\n  );\n  if (eligible.length === 0) return null;\n  // listRuns sorts newest-first; find the newest with promoted: true.\n  const promoted = eligible.find((r) => r.promoted === true);\n  if (promoted) return promoted;\n  // Fallback: highest-scored eligible run when promotion data is unavailable.\n  let best = eligible[0];\n  for (const r of eligible) {\n    if (r.score! > best.score!) best = r;\n  }\n  return best;\n}\n"],"mappings":";;;;AAwDA,SAAgB,SAAS,UAAkB,UAAgC;CACzE,MAAM,MAAM,cAAc,UAAU,SAAS;AAC7C,KAAI,CAAC,WAAW,IAAI,CAAE,QAAO,EAAE;CAC/B,MAAM,MAAoB,EAAE;AAC5B,MAAK,MAAM,SAAS,YAAY,IAAI,EAAE;EACpC,MAAM,SAAS,KAAK,KAAK,MAAM;AAC/B,MAAI,CAAC,SAAS,OAAO,CAAC,aAAa,CAAE;EACrC,MAAM,gBAAgB,KAAK,QAAQ,gBAAgB;EACnD,MAAM,aAAa,KAAK,QAAQ,aAAa;EAC7C,MAAM,cAAc,KAAK,QAAQ,eAAe;EAChD,MAAM,WAAW,iBAA+B,cAAc;AAC9D,MAAI,CAAC,SAAU;EACf,MAAM,QAAQ,iBAA4B,WAAW;AACrD,MAAI,KAAK;GACP,QAAQ,SAAS;GACjB,QAAQ,SAAS;GACjB,SAAS,SAAS;GAClB,WAAW,SAAS;GACpB,YAAY,SAAS;GACrB,QAAQ,OAAO,UAAU;GACzB,OAAO,OAAO,SAAS;GACvB,GAAG,OAAO,WAAW;GACrB,UAAU,OAAO,YAAY;GAC7B,YAAY,SAAS;GACrB,UAAU,OAAO,YAAY;GAC7B,cAAc,SAAS;GACvB,iBAAiB,SAAS;GAC1B;GACA;GACA;GACA,iBAAiB,SAAS,mBAAmB;GAC7C,kBAAkB,SAAS,oBAAoB;GAChD,CAAC;;AAGJ,KAAI,MAAM,GAAG,MAAO,EAAE,aAAa,EAAE,aAAa,KAAK,EAAG;AAC1D,QAAO;;;;;;;AAQT,SAAgB,gBACd,MACA,aACA,gBACmB;CACnB,MAAM,WAAW,KAAK,QACnB,MACC,EAAE,WAAW,cACb,EAAE,UAAU,QACZ,EAAE,iBAAiB,eACnB,EAAE,oBAAoB,eACzB;AACD,KAAI,SAAS,WAAW,EAAG,QAAO;CAElC,MAAM,WAAW,SAAS,MAAM,MAAM,EAAE,aAAa,KAAK;AAC1D,KAAI,SAAU,QAAO;CAErB,IAAI,OAAO,SAAS;AACpB,MAAK,MAAM,KAAK,SACd,KAAI,EAAE,QAAS,KAAK,MAAQ,QAAO;AAErC,QAAO"}

package/dist/src/lib/parse-args.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"parse-args.js","names":[],"sources":["../../../src/lib/parse-args.ts"],"sourcesContent":["export type Flags = Record<string, string | boolean>;\n\nexport function parseFlags(argv: string[]): {\n  positional: string[];\n  flags: Flags;\n} {\n  const positional: string[] = [];\n  const flags: Flags = {};\n  for (let i = 0; i < argv.length; i++) {\n    const a = argv[i];\n    if (a.startsWith(\"--\")) {\n      const eq = a.indexOf(\"=\");\n      let key: string, valFromEq: string | undefined;\n      if (eq >= 0) {\n        key = a.slice(2, eq);\n        valFromEq = a.slice(eq + 1);\n      } else {\n        key = a.slice(2);\n      }\n      if (valFromEq !== undefined) {\n        flags[key] = valFromEq;\n        continue;\n      }\n      const next = argv[i + 1];\n      // A flag's value is the next token unless it looks like another flag (starts with `-`).\n      if (next === undefined || next.startsWith(\"-\")) {\n        flags[key] = true;\n      } else {\n        flags[key] = next;\n        i++;\n      }\n    } else if (a.startsWith(\"-\") && a.length > 1) {\n      // Short flag like `-n 3`. Treat similarly to long flags.\n      const key = a.slice(1);\n      const next = argv[i + 1];\n      if (next === undefined || next.startsWith(\"-\")) {\n        flags[key] = true;\n      } else {\n        flags[key] = next;\n        i++;\n      }\n    } else {\n      positional.push(a);\n    }\n  }\n  return { positional, flags };\n}\n\nexport function strFlag(flags: Flags, key: string): string | undefined {\n  const v = flags[key];\n  return typeof v === \"string\" ? v : undefined;\n}\n\nexport function boolFlag(flags: Flags, key: string): boolean {\n  return flags[key] === true || flags[key] === \"true\";\n}\n"],"mappings":";AAEA,SAAgB,WAAW,MAGzB;CACA,MAAM,aAAuB,EAAE;CAC/B,MAAM,QAAe,EAAE;AACvB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,IAAI,KAAK;AACf,MAAI,EAAE,WAAW,KAAK,EAAE;GACtB,MAAM,KAAK,EAAE,QAAQ,IAAI;GACzB,IAAI,KAAa;AACjB,OAAI,MAAM,GAAG;AACX,UAAM,EAAE,MAAM,GAAG,GAAG;AACpB,gBAAY,EAAE,MAAM,KAAK,EAAE;SAE3B,OAAM,EAAE,MAAM,EAAE;AAElB,OAAI,cAAc,KAAA,GAAW;AAC3B,UAAM,OAAO;AACb;;GAEF,MAAM,OAAO,KAAK,IAAI;AAEtB,OAAI,SAAS,KAAA,KAAa,KAAK,WAAW,IAAI,CAC5C,OAAM,OAAO;QACR;AACL,UAAM,OAAO;AACb;;aAEO,EAAE,WAAW,IAAI,IAAI,EAAE,SAAS,GAAG;GAE5C,MAAM,MAAM,EAAE,MAAM,EAAE;GACtB,MAAM,OAAO,KAAK,IAAI;AACtB,OAAI,SAAS,KAAA,KAAa,KAAK,WAAW,IAAI,CAC5C,OAAM,OAAO;QACR;AACL,UAAM,OAAO;AACb;;QAGF,YAAW,KAAK,EAAE;;AAGtB,QAAO;EAAE;EAAY;EAAO;;AAG9B,SAAgB,QAAQ,OAAc,KAAiC;CACrE,MAAM,IAAI,MAAM;AAChB,QAAO,OAAO,MAAM,WAAW,IAAI,KAAA;;AAGrC,SAAgB,SAAS,OAAc,KAAsB;AAC3D,QAAO,MAAM,SAAS,QAAQ,MAAM,SAAS"}