{"id":105068,"date":"2026-10-28T00:19:37","date_gmt":"2026-10-27T22:19:37","guid":{"rendered":"https:\/\/twodots.gr\/?p=105068"},"modified":"2026-10-28T00:19:39","modified_gmt":"2026-10-27T22:19:39","slug":"llm-judges-ai-agents-megalyteros-kritis-den-arkei","status":"publish","type":"post","link":"https:\/\/twodots.gr\/de\/llm-judges-ai-agents-megalyteros-kritis-den-arkei\/","title":{"rendered":"LLM judges \u03c3\u03c4\u03b1 AI agents: \u03b3\u03b9\u03b1\u03c4\u03af \u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2 \u03ba\u03c1\u03b9\u03c4\u03ae\u03c2 \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af"},"content":{"rendered":"<div class=\"td-article-lede\">\n<p><strong>\u0397 \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b7 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7:<\/strong> \u03ad\u03bd\u03b1\u03c2 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2 LLM judge \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03b1\u03c0\u03bf\u03b4\u03b5\u03af\u03be\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03ad\u03bd\u03b1\u03c2 AI agent \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b5 \u03c3\u03c9\u03c3\u03c4\u03ac \u03c4\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03c4\u03bf\u03c5. \u03a4\u03bf AgentJudgeBench \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9, \u03c3\u03c4\u03b1 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b1 tool-calling tasks \u03c7\u03c9\u03c1\u03af\u03c2 ground truth, \u03ad\u03be\u03b9 judges \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ae\u03c2 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1\u03c2 \u03c3\u03c5\u03b3\u03ba\u03bb\u03af\u03bd\u03bf\u03c5\u03bd \u03c3\u03b5 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 77%\u201382% \u03bc\u03b5 \u03ad\u03bd\u03b1\u03bd \u03c0\u03c1\u03bf\u03b3\u03c1\u03b1\u03bc\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf. \u0397 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c3\u03c4\u03b5\u03c1\u03b7 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03bd\u03b4\u03c5\u03b1\u03c3\u03bc\u03cc\u03c2 deterministic checks, \u03b4\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b7\u03c2 rubric, \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2 \u03b1\u03bd\u03ac failure type \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7\u03c2 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7\u03c2.<\/p>\n<\/div>\n<p>\u03a4\u03bf \u03b5\u03cd\u03c1\u03b7\u03bc\u03b1 \u03b1\u03c6\u03bf\u03c1\u03ac \u03ac\u03bc\u03b5\u03c3\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03c5 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd AI agents \u03c3\u03b5 customer support, e-commerce, \u03c0\u03c9\u03bb\u03ae\u03c3\u03b5\u03b9\u03c2, \u03ad\u03c1\u03b5\u03c5\u03bd\u03b1 \u03ae back-office \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2. \u0391\u03bd \u03ad\u03bd\u03b1\u03c2 agent \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af \u03b1\u03bb\u03c5\u03c3\u03af\u03b4\u03b1 \u03b5\u03bd\u03b5\u03c1\u03b3\u03b5\u03b9\u03ce\u03bd \u03ba\u03b1\u03b9 \u03c4\u03bf quality control \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf LLM, \u03c4\u03cc\u03c4\u03b5 \u03b7 \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1 \u03c4\u03bf\u03c5 \u00ab\u03ba\u03c1\u03b9\u03c4\u03ae\u00bb \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1\u03c2 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7\u03c2 \u03c4\u03b7\u03c2 \u03c5\u03c0\u03b7\u03c1\u03b5\u03c3\u03af\u03b1\u03c2. \u0388\u03bd\u03b1 \u03c0\u03b5\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc verdict \u03ae \u03ad\u03bd\u03b1\u03c2 \u03c5\u03c8\u03b7\u03bb\u03cc\u03c2 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03cc\u03c2 \u03b2\u03b1\u03b8\u03bc\u03cc\u03c2 \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b5\u03c0\u03b9\u03bb\u03ad\u03c7\u03b8\u03b7\u03ba\u03b1\u03bd \u03c4\u03b1 \u03c3\u03c9\u03c3\u03c4\u03ac \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1, \u03bc\u03b5 \u03c4\u03b1 \u03c3\u03c9\u03c3\u03c4\u03ac arguments \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c3\u03b5\u03b9\u03c1\u03ac.<\/p>\n<div class=\"td-article-toc\">\n<div class=\"td-toc-title\">Inhaltsverzeichnis<\/div>\n<ul>\n<li><a href=\"#ti-deichnei-agentjudgebench\">\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c4\u03bf AgentJudgeBench<\/a><\/li>\n<li><a href=\"#giati-tool-calling-diaforetiko\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf tool calling \u03b4\u03b5\u03bd \u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c0\u03c9\u03c2 \u03ad\u03bd\u03b1 \u03ba\u03b5\u03af\u03bc\u03b5\u03bd\u03bf<\/a><\/li>\n<li><a href=\"#ti-perilamvanei-benchmark\">\u03a4\u03b9 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03c4\u03bf benchmark<\/a><\/li>\n<li><a href=\"#programmatic-judge\">\u039f programmatic judge \u03c9\u03c2 \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ac\u03c2<\/a><\/li>\n<li><a href=\"#dyskolia-kai-ceiling\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03af\u03b1 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03ba\u03bf\u03b9\u03bd\u03cc ceiling<\/a><\/li>\n<li><a href=\"#ground-truth-anchoring\">\u03a0\u03cc\u03c4\u03b5 \u03c4\u03bf ground truth \u03b2\u03bf\u03b7\u03b8\u03ac \u03ae \u03b1\u03b3\u03ba\u03c5\u03c1\u03ce\u03bd\u03b5\u03b9<\/a><\/li>\n<li><a href=\"#rubrics-kai-prompts\">\u0393\u03b9\u03b1\u03c4\u03af \u03bf\u03b9 structured rubrics \u03b5\u03af\u03bd\u03b1\u03b9 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf\u03c2 \u03bc\u03bf\u03c7\u03bb\u03cc\u03c2<\/a><\/li>\n<li><a href=\"#dag-topologies\">\u03a4\u03b9 \u03b1\u03c0\u03bf\u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03bf\u03c5\u03bd \u03bf\u03b9 DAG topologies<\/a><\/li>\n<li><a href=\"#production-evaluation-pipeline\">\u03a0\u03ce\u03c2 \u03c3\u03c4\u03ae\u03bd\u03b5\u03c4\u03b1\u03b9 production evaluation pipeline<\/a><\/li>\n<li><a href=\"#oria-meletis\">\u03a4\u03b1 \u03cc\u03c1\u03b9\u03b1 \u03c4\u03b7\u03c2 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7\u03c2<\/a><\/li>\n<li><a href=\"#symperasma\">\u03a3\u03c5\u03bc\u03c0\u03ad\u03c1\u03b1\u03c3\u03bc\u03b1: \u03bf judge \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03b9\u03c3\u03b8\u03b7\u03c4\u03ae\u03c1\u03b1\u03c2<\/a><\/li>\n<\/ul>\n<\/div>\n<h2 id=\"ti-deichnei-agentjudgebench\">\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c4\u03bf AgentJudgeBench<\/h2>\n<p>\u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u00abAgentJudgeBench: A Multi-Difficulty Benchmark for Evaluating LLM Judges on Agentic Tool-Calling\u00bb \u03b5\u03be\u03b5\u03c4\u03ac\u03b6\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b7 \u03c3\u03c5\u03bd\u03b7\u03b8\u03b9\u03c3\u03bc\u03ad\u03bd\u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03ba\u03b5\u03b9\u03bc\u03ad\u03bd\u03bf\u03c5. \u0391\u03bd\u03c4\u03af \u03bd\u03b1 \u03c1\u03c9\u03c4\u03ac \u03c0\u03bf\u03b9\u03b1 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03b9\u03bf \u03ba\u03b1\u03bb\u03bf\u03b3\u03c1\u03b1\u03bc\u03bc\u03ad\u03bd\u03b7, \u03b6\u03b7\u03c4\u03ac \u03b1\u03c0\u03cc LLM judges \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03be\u03bf\u03c5\u03bd \u03b4\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03c1\u03bf\u03ad\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd: \u03c0\u03bf\u03b9\u03bf tool \u03b5\u03c0\u03b9\u03bb\u03ad\u03c7\u03b8\u03b7\u03ba\u03b5, \u03b1\u03bd \u03c4\u03b1 parameters \u03ad\u03c7\u03bf\u03c5\u03bd \u03c3\u03c9\u03c3\u03c4\u03ae \u03b4\u03bf\u03bc\u03ae, \u03b1\u03bd \u03b7 \u03c3\u03b5\u03b9\u03c1\u03ac \u03c3\u03ad\u03b2\u03b5\u03c4\u03b1\u03b9 \u03c4\u03b9\u03c2 \u03b5\u03be\u03b1\u03c1\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03bd \u03ba\u03b1\u03bb\u03cd\u03c6\u03b8\u03b7\u03ba\u03b5 \u03cc\u03bb\u03bf \u03c4\u03bf user intent.<\/p>\n<p>\u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03bf\u03c5\u03bd \u03ba\u03ac\u03b8\u03b5 verdict \u03bc\u03b5 \u03ad\u03bd\u03b1\u03bd deterministic programmatic scorer. \u03a4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 \u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c3\u03bf \u03b1\u03c5\u03be\u03ac\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b7 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03af\u03b1, \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 1,5 \u03c6\u03bf\u03c1\u03ad\u03c2 \u03c4\u03b1\u03c7\u03cd\u03c4\u03b5\u03c1\u03b1 \u03cc\u03c4\u03b1\u03bd \u03bf judge \u03b4\u03b5\u03bd \u03b2\u03bb\u03ad\u03c0\u03b5\u03b9 ground truth. \u03a3\u03c4\u03b1 hard queries \u03c7\u03c9\u03c1\u03af\u03c2 reference trace, \u03cc\u03bb\u03bf\u03b9 \u03bf\u03b9 \u03ad\u03be\u03b9 \u03b2\u03b1\u03c3\u03b9\u03ba\u03bf\u03af judges \u03b2\u03c1\u03af\u03c3\u03ba\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03c3\u03c4\u03b5\u03bd\u03ae \u03b6\u03ce\u03bd\u03b7 77%\u201382%, \u03c0\u03b1\u03c1\u03cc\u03c4\u03b9 \u03b4\u03b9\u03b1\u03c6\u03ad\u03c1\u03bf\u03c5\u03bd \u03c3\u03b7\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03ac \u03c3\u03b5 \u03bc\u03ad\u03b3\u03b5\u03b8\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03c0\u03c1\u03bf\u03ad\u03bb\u03b5\u03c5\u03c3\u03b7.<\/p>\n<p>\u0391\u03c5\u03c4\u03cc \u03b4\u03b5\u03bd \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03bf\u03b9 LLM judges \u03b5\u03af\u03bd\u03b1\u03b9 \u03ac\u03c7\u03c1\u03b7\u03c3\u03c4\u03bf\u03b9. \u03a3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03c4\u03bf\u03c5 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5 \u03b4\u03b5\u03bd \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af \u03ad\u03bd\u03b1\u03bd \u03b4\u03bf\u03bc\u03b9\u03ba\u03cc \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03cc: \u03c7\u03c9\u03c1\u03af\u03c2 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7 \u03ae \u03b3\u03bd\u03c9\u03c3\u03c4\u03ae \u03c3\u03c9\u03c3\u03c4\u03ae \u03c1\u03bf\u03ae, \u03bf judge \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03b1\u03c4\u03b1\u03c3\u03ba\u03b5\u03c5\u03ac\u03c3\u03b5\u03b9 \u03c4\u03b7\u03bd \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bc\u03cc\u03bd\u03bf \u03b1\u03c0\u03cc \u03c4\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03b1 tool schemas. \u03a3\u03b5 \u03ad\u03bd\u03b1 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03cc \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1, \u03b7 \u03b1\u03bd\u03b1\u03b2\u03ac\u03b8\u03bc\u03b9\u03c3\u03b7 \u03c3\u03b5 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf judge \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03b5\u03c0\u03bf\u03bc\u03ad\u03bd\u03c9\u03c2 \u03bd\u03b1 \u03b1\u03c0\u03bf\u03b4\u03ce\u03c3\u03b5\u03b9 \u03bb\u03b9\u03b3\u03cc\u03c4\u03b5\u03c1\u03bf \u03b1\u03c0\u03cc \u03bc\u03b9\u03b1 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03b5\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5.<\/p>\n<div class=\"td-article-note\"><strong>Lesepraxis:<\/strong> \u03c4\u03bf 77%\u201382% \u03b5\u03af\u03bd\u03b1\u03b9 alignment \u03bc\u03b5 \u03c4\u03bf\u03bd \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf programmatic scorer \u03c3\u03c4\u03b1 hard, no-ground-truth queries \u03c4\u03bf\u03c5 benchmark. \u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b3\u03b5\u03bd\u03b9\u03ba\u03cc\u03c2 \u03b4\u03b5\u03af\u03ba\u03c4\u03b7\u03c2 \u03b1\u03ba\u03c1\u03af\u03b2\u03b5\u03b9\u03b1\u03c2 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 agent, \u03ba\u03ac\u03b8\u03b5 \u03ba\u03bb\u03ac\u03b4\u03bf \u03ae \u03ba\u03ac\u03b8\u03b5 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03cc \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd.<\/div>\n<h2 id=\"giati-tool-calling-diaforetiko\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf tool calling \u03b4\u03b5\u03bd \u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c0\u03c9\u03c2 \u03ad\u03bd\u03b1 \u03ba\u03b5\u03af\u03bc\u03b5\u03bd\u03bf<\/h2>\n<p>\u03a3\u03b5 \u03bc\u03b9\u03b1 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03c6\u03c5\u03c3\u03b9\u03ba\u03ae\u03c2 \u03b3\u03bb\u03ce\u03c3\u03c3\u03b1\u03c2, \u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03c6\u03bf\u03c1\u03ac \u03c3\u03b1\u03c6\u03ae\u03bd\u03b5\u03b9\u03b1, \u03c3\u03c5\u03bd\u03ac\u03c6\u03b5\u03b9\u03b1 \u03ae \u03c0\u03c1\u03bf\u03c4\u03af\u03bc\u03b7\u03c3\u03b7. \u03a3\u03b5 \u03bc\u03b9\u03b1 agentic \u03c1\u03bf\u03ae, \u03b7 \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b9\u03ba\u03ae. \u039f agent \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03be\u03b5\u03b9 functions \u03b1\u03c0\u03cc typed schemas, \u03bd\u03b1 \u03c3\u03c7\u03b7\u03bc\u03b1\u03c4\u03af\u03c3\u03b5\u03b9 \u03ad\u03b3\u03ba\u03c5\u03c1\u03b1 arguments, \u03bd\u03b1 \u03c4\u03b7\u03c1\u03ae\u03c3\u03b5\u03b9 execution dependencies \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03c3\u03b5\u03b9 \u03cc\u03bb\u03b1 \u03c4\u03b1 \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bc\u03b5\u03bd\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1. \u039c\u03b9\u03b1 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03ba\u03bf\u03cd\u03b3\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ae \u03ba\u03b1\u03b9 \u03c4\u03b1\u03c5\u03c4\u03cc\u03c7\u03c1\u03bf\u03bd\u03b1 \u03bd\u03b1 \u03ad\u03c7\u03b5\u03b9 \u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b5\u03b9 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03af\u03b1.<\/p>\n<p>\u03a4\u03bf benchmark \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b9\u03c2 \u03b4\u03b9\u03b1\u03c3\u03c4\u03ac\u03c3\u03b5\u03b9\u03c2: tool selection, parameter structure, sequence accuracy \u03ba\u03b1\u03b9 query coverage. \u0391\u03c5\u03c4\u03ad\u03c2 \u03b4\u03b5\u03bd \u03ba\u03b9\u03bd\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03bc\u03b1\u03b6\u03af. \u0388\u03bd\u03b1\u03c2 judge \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03bd\u03b1\u03b3\u03bd\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03b1 \u03c3\u03c9\u03c3\u03c4\u03ac \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03b1\u03bb\u03bb\u03ac \u03bd\u03b1 \u03c7\u03ac\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b4\u03cd\u03bf branches \u03ad\u03c0\u03c1\u03b5\u03c0\u03b5 \u03bd\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03c9\u03b8\u03bf\u03cd\u03bd \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03b2\u03ae\u03bc\u03b1. \u0388\u03bd\u03b1\u03c2 \u03ac\u03bb\u03bb\u03bf\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ac \u03c4\u03b7 \u03c3\u03b5\u03b9\u03c1\u03ac, \u03b1\u03bb\u03bb\u03ac \u03bd\u03b1 \u03bc\u03b7\u03bd \u03b5\u03bd\u03c4\u03bf\u03c0\u03af\u03b6\u03b5\u03b9 missing argument \u03ae \u03c4\u03bc\u03ae\u03bc\u03b1 \u03c4\u03bf\u03c5 \u03b1\u03b9\u03c4\u03ae\u03bc\u03b1\u03c4\u03bf\u03c2 \u03c0\u03bf\u03c5 \u03ad\u03bc\u03b5\u03b9\u03bd\u03b5 \u03b1\u03ba\u03ac\u03bb\u03c5\u03c0\u03c4\u03bf.<\/p>\n<p>\u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7, \u03b1\u03c5\u03c4\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03bb\u03cc\u03b3\u03bf\u03c2 \u03bd\u03b1 \u03b1\u03c0\u03bf\u03c6\u03b5\u03cd\u03b3\u03b5\u03b9 \u03ad\u03bd\u03b1\u03bd \u03bc\u03bf\u03bd\u03b1\u03b4\u03b9\u03ba\u03cc \u00abAI score\u00bb. \u0388\u03bd\u03b1 aggregate pass rate \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03ba\u03c1\u03cd\u03b2\u03b5\u03b9 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03c4\u03bf failure type \u03c0\u03bf\u03c5 \u03c0\u03c1\u03bf\u03ba\u03b1\u03bb\u03b5\u03af \u03c4\u03b7 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b6\u03b7\u03bc\u03b9\u03ac, \u03cc\u03c0\u03c9\u03c2 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ae \u03c7\u03c1\u03b7\u03bc\u03ac\u03c4\u03c9\u03bd, \u03b1\u03c0\u03bf\u03c3\u03c4\u03bf\u03bb\u03ae \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b1\u03c2 \u03c7\u03c9\u03c1\u03af\u03c2 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03b1\u03c0\u03bf\u03b8\u03ad\u03bc\u03b1\u03c4\u03bf\u03c2 \u03ae \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03bb\u03bf\u03b3\u03b1\u03c1\u03b9\u03b1\u03c3\u03bc\u03bf\u03cd \u03c7\u03c9\u03c1\u03af\u03c2 \u03c4\u03b7\u03bd \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bc\u03b5\u03bd\u03b7 \u03ad\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7.<\/p>\n<h2 id=\"ti-perilamvanei-benchmark\">\u03a4\u03b9 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03c4\u03bf benchmark<\/h2>\n<p>\u03a4\u03bf AgentJudgeBench \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 3.808 \u03bc\u03bf\u03bd\u03b1\u03b4\u03b9\u03ba\u03ac records \u03b1\u03c0\u03cc 15 enterprise domains. \u039a\u03ac\u03b8\u03b5 record \u03b4\u03b9\u03b1\u03b8\u03ad\u03c4\u03b5\u03b9 8 \u03ad\u03c9\u03c2 19 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1 programmatically verified ground-truth trace. \u039f\u03b9 \u03c1\u03bf\u03ad\u03c2 \u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03bf\u03c5\u03bd \u03ad\u03be\u03b9 DAG topologies \u03ba\u03b1\u03b9 \u03c4\u03c1\u03af\u03b1 difficulty tiers, easy, medium \u03ba\u03b1\u03b9 hard, \u03bc\u03b5 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b5\u03c2 \u03b1\u03bd\u03b1\u03b4\u03b9\u03b1\u03c4\u03c5\u03c0\u03ce\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03c5 \u03b4\u03b9\u03b1\u03c4\u03b7\u03c1\u03bf\u03cd\u03bd \u03c4\u03bf \u03b1\u03bd\u03b1\u03bc\u03b5\u03bd\u03cc\u03bc\u03b5\u03bd\u03bf trace.<\/p>\n<p>\u03a0\u03ad\u03bd\u03c4\u03b5 generator models \u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03c5\u03bd tool-call sequences \u03ba\u03b1\u03b9 \u03ad\u03be\u03b9 \u03b2\u03b1\u03c3\u03b9\u03ba\u03bf\u03af LLM judges \u03c4\u03b9\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03bf\u03cd\u03bd \u03c4\u03cc\u03c3\u03bf \u03bc\u03b5 \u03cc\u03c3\u03bf \u03ba\u03b1\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 \u03c0\u03c1\u03cc\u03c3\u03b2\u03b1\u03c3\u03b7 \u03c3\u03c4\u03bf ground truth. \u039c\u03b5\u03c4\u03ac \u03c4\u03bf\u03bd \u03ba\u03b1\u03b8\u03b1\u03c1\u03b9\u03c3\u03bc\u03cc \u03c4\u03c9\u03bd \u03ad\u03b3\u03ba\u03c5\u03c1\u03c9\u03bd \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03b5\u03c3\u03bc\u03ac\u03c4\u03c9\u03bd, \u03b7 \u03c0\u03bb\u03ae\u03c1\u03c9\u03c2 \u03b4\u03b9\u03b1\u03c3\u03c4\u03b1\u03c5\u03c1\u03c9\u03bc\u03ad\u03bd\u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 321.648 paired verdicts. \u03a4\u03bf dataset \u03b4\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03cd\u03b5\u03b9 \u03c4\u03b1 records, \u03c4\u03b1 generator outputs, \u03c4\u03bf\u03c5\u03c2 programmatic scores \u03ba\u03b1\u03b9 \u03c4\u03b9\u03c2 verdicts, \u03ce\u03c3\u03c4\u03b5 \u03c4\u03b1 \u03b5\u03c5\u03c1\u03ae\u03bc\u03b1\u03c4\u03b1 \u03bd\u03b1 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03b8\u03bf\u03cd\u03bd \u03c0\u03ad\u03c1\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf paper.<\/p>\n<div class=\"td-chart td-chart--metrics\">\n<div class=\"td-chart-head\">\n<p class=\"td-chart-title\">\u0397 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03c4\u03bf\u03c5 AgentJudgeBench<\/p>\n<p class=\"td-chart-subtitle\">\u0394\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03c5\u03bc\u03ad\u03bd\u03b1 \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03b1 \u03c4\u03b7\u03c2 \u03c0\u03c1\u03c9\u03c4\u03bf\u03b3\u03b5\u03bd\u03bf\u03cd\u03c2 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03bf\u03c5 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf\u03c5 dataset.<\/p>\n<\/div>\n<div class=\"td-chart-body\">\n<div class=\"td-metric-grid td-metric-grid--three-columns\">\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">3.808<\/span><span class=\"td-metric-label\">\u03bc\u03bf\u03bd\u03b1\u03b4\u03b9\u03ba\u03ac records<\/span><span class=\"td-metric-note\">\u03a3\u03b5 15 enterprise domains \u03ba\u03b1\u03b9 6 DAG topologies<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">321.648<\/span><span class=\"td-metric-label\">paired \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03ae\u03c3\u03b5\u03b9\u03c2<\/span><span class=\"td-metric-note\">\u039c\u03b5 \u03ba\u03b1\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 ground-truth exposure<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">77%\u201382%<\/span><span class=\"td-metric-label\">hard no-GT alignment band<\/span><span class=\"td-metric-note\">\u039a\u03bf\u03b9\u03bd\u03cc ceiling \u03b3\u03b9\u03b1 \u03c4\u03bf\u03c5\u03c2 6 \u03b2\u03b1\u03c3\u03b9\u03ba\u03bf\u03cd\u03c2 judges<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<p>\u03a4\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03bd\u03b8\u03b5\u03c4\u03b9\u03ba\u03ac, \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b7 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b3\u03bd\u03c9\u03c3\u03c4\u03ae \u03c3\u03c9\u03c3\u03c4\u03ae \u03c1\u03bf\u03ae, \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b7 topology \u03ba\u03b1\u03b9 \u03bc\u03b5\u03b3\u03ac\u03bb\u03b7 \u03c0\u03b5\u03b9\u03c1\u03b1\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1. \u0391\u03c5\u03c4\u03cc \u03c0\u03c1\u03bf\u03c3\u03c6\u03ad\u03c1\u03b5\u03b9 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03bc\u03cc\u03c4\u03b7\u03c4\u03b1, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03b1\u03c0\u03cc\u03c3\u03c4\u03b1\u03c3\u03b7 \u03b1\u03c0\u03cc \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac production logs, \u03c0\u03bf\u03bb\u03c5\u03b3\u03c5\u03c1\u03b9\u03ba\u03ad\u03c2 \u03c3\u03c5\u03bd\u03bf\u03bc\u03b9\u03bb\u03af\u03b5\u03c2, failures \u03b4\u03b9\u03ba\u03c4\u03cd\u03bf\u03c5 \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03b1\u03b2\u03b1\u03bb\u03bb\u03cc\u03bc\u03b5\u03bd\u03b1 permissions. \u03a4\u03bf benchmark \u03b5\u03af\u03bd\u03b1\u03b9 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc \u03b5\u03c1\u03b3\u03b1\u03c3\u03c4\u03ae\u03c1\u03b9\u03bf \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7\u03c2, \u03cc\u03c7\u03b9 \u03c0\u03b9\u03c3\u03c4\u03bf\u03c0\u03bf\u03b9\u03b7\u03c4\u03b9\u03ba\u03cc \u03b3\u03b9\u03b1 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03c7\u03c1\u03ae\u03c3\u03b7.<\/p>\n<h2 id=\"programmatic-judge\">\u039f programmatic judge \u03c9\u03c2 \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ac\u03c2<\/h2>\n<p>\u039a\u03ac\u03b8\u03b5 \u03c0\u03c1\u03cc\u03b2\u03bb\u03b5\u03c8\u03b7 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bb\u03bf\u03b3\u03b5\u03af\u03c4\u03b1\u03b9 \u03c0\u03c1\u03ce\u03c4\u03b1 \u03b1\u03c0\u03cc deterministic scorer. \u0391\u03bd \u03b7 \u00ab\u03b1\u03bb\u03ae\u03b8\u03b5\u03b9\u03b1\u00bb \u03c0\u03c1\u03bf\u03b5\u03c1\u03c7\u03cc\u03c4\u03b1\u03bd \u03b1\u03c0\u03cc \u03ac\u03bb\u03bb\u03bf LLM, \u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03b5\u03c1\u03ce\u03c4\u03b7\u03bc\u03b1 \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1\u03c2 \u03b8\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c6\u03b5\u03c1\u03cc\u03c4\u03b1\u03bd \u03b1\u03c0\u03bb\u03ce\u03c2 \u03ad\u03bd\u03b1 \u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03bf \u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b1. \u0391\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b1, \u03c0\u03bb\u03ae\u03c1\u03b7\u03c2 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03b4\u03b5\u03bd \u03ba\u03bb\u03b9\u03bc\u03b1\u03ba\u03ce\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b5\u03cd\u03ba\u03bf\u03bb\u03b1 \u03c3\u03b5 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b1\u03c0\u03cc 300 \u03c7\u03b9\u03bb\u03b9\u03ac\u03b4\u03b5\u03c2 verdicts. \u039f programmatic judge \u03b4\u03af\u03bd\u03b5\u03b9 \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03cc \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 cell \u03c4\u03bf\u03c5 \u03c0\u03b5\u03b9\u03c1\u03ac\u03bc\u03b1\u03c4\u03bf\u03c2.<\/p>\n<p>\u039f \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03b4\u03b5\u03bd \u03b8\u03b5\u03c9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9 \u03b1\u03bb\u03ac\u03bd\u03b8\u03b1\u03c3\u03c4\u03bf\u03c2. \u03a3\u03b5 validation subset 120 hard records, \u03bf scorer \u03b5\u03af\u03c7\u03b5 92,7% \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 \u03b1\u03bd\u03ac metric \u03bc\u03b5 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03bf\u03c5\u03c2 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03bf\u03c5\u03c2, \u03bc\u03b5 \u03c4\u03b7 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7 \u03c3\u03c4\u03b7 \u03b4\u03bf\u03bc\u03ae \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03c9\u03bd. \u0391\u03c5\u03c4\u03ae \u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03b3\u03b9\u03b1\u03c4\u03af \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c0\u03bf\u03c5 \u03b5\u03c5\u03b8\u03c5\u03b3\u03c1\u03b1\u03bc\u03bc\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b1 \u03bc\u03b5 \u03c4\u03bf\u03bd programmatic reference \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03b1\u03c1\u03b1\u03af\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c0\u03bf\u03c5 \u03b5\u03c5\u03b8\u03c5\u03b3\u03c1\u03b1\u03bc\u03bc\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b1 \u03bc\u03b5 \u03c4\u03b7\u03bd \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03ba\u03c1\u03af\u03c3\u03b7.<\/p>\n<div class=\"td-comparison-cards\">\n<article class=\"td-platform-card\">\n<h3>Deterministic checks<\/h3>\n<p>\u0395\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5\u03bd \u03bc\u03b5 \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1 schema validity, required arguments, \u03b5\u03c0\u03b9\u03c4\u03c1\u03b5\u03c0\u03c4\u03ad\u03c2 \u03c4\u03b9\u03bc\u03ad\u03c2, dependencies \u03ba\u03b1\u03b9 \u03c5\u03c0\u03bf\u03c7\u03c1\u03b5\u03c9\u03c4\u03b9\u03ba\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1. \u0395\u03af\u03bd\u03b1\u03b9 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b1\u03b3\u03ce\u03b3\u03b9\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03ba\u03b1\u03c4\u03ac\u03bb\u03bb\u03b7\u03bb\u03b1 \u03cc\u03c4\u03b1\u03bd \u03bf \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03ba\u03c6\u03c1\u03b1\u03c3\u03c4\u03b5\u03af \u03c1\u03b7\u03c4\u03ac.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">\u03a3\u03c4\u03b1\u03b8\u03b5\u03c1\u03cc\u03c2 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2<\/span><span class=\"td-badge\">\u0386\u03bc\u03b5\u03c3\u03bf failure signal<\/span><\/div>\n<\/article>\n<article class=\"td-platform-card td-platform-card--navy\">\n<h3>LLM judge<\/h3>\n<p>\u0395\u03be\u03b5\u03c4\u03ac\u03b6\u03b5\u03b9 \u03b1\u03bc\u03c6\u03af\u03c3\u03b7\u03bc\u03b7 \u03ba\u03ac\u03bb\u03c5\u03c8\u03b7 intent, \u03c0\u03bf\u03b9\u03bf\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b1\u03c3\u03c5\u03bd\u03ad\u03c0\u03b5\u03b9\u03b5\u03c2 \u03ba\u03b1\u03b9 cases \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03c7\u03c9\u03c1\u03bf\u03cd\u03bd \u03c0\u03bb\u03ae\u03c1\u03c9\u03c2 \u03c3\u03b5 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b5\u03c2. \u03a7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 calibration, versioning \u03ba\u03b1\u03b9 \u03c3\u03b1\u03c6\u03ae \u03cc\u03c1\u03b9\u03b1 \u03b5\u03be\u03bf\u03c5\u03c3\u03af\u03b1\u03c2.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">\u03a3\u03b7\u03bc\u03b1\u03c3\u03b9\u03bf\u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03ba\u03c1\u03af\u03c3\u03b7<\/span><span class=\"td-badge\">\u039c\u03b7 deterministic<\/span><\/div>\n<\/article>\n<\/div>\n<p>\u0397 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03b4\u03b5\u03bd \u03b4\u03b9\u03b1\u03bb\u03ad\u03b3\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b1 \u03b4\u03cd\u03bf. \u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af deterministic checks \u03c9\u03c2 \u03c0\u03c1\u03ce\u03c4\u03bf \u03c6\u03af\u03bb\u03c4\u03c1\u03bf, LLM judge \u03b3\u03b9\u03b1 \u03c4\u03b9\u03c2 \u03c3\u03b7\u03bc\u03b1\u03c3\u03b9\u03bf\u03bb\u03bf\u03b3\u03b9\u03ba\u03ad\u03c2 \u03b3\u03ba\u03c1\u03af\u03b6\u03b5\u03c2 \u03b6\u03ce\u03bd\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03ac\u03bd\u03b8\u03c1\u03c9\u03c0\u03bf \u03b3\u03b9\u03b1 disagreements \u03ae \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03c5\u03c8\u03b7\u03bb\u03bf\u03cd \u03c1\u03af\u03c3\u03ba\u03bf\u03c5. \u03a4\u03bf \u03af\u03b4\u03b9\u03bf \u03bc\u03ac\u03b8\u03b7\u03bc\u03b1 \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b7\u03bd <a href=\"https:\/\/twodots.gr\/ai-agents-tool-calling-proto-lathos-alysida\/\">\u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1 \u03c4\u03c9\u03bd AI agents \u03c3\u03c4\u03bf tool calling<\/a>: \u03b7 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03b7\u03c2 \u03b1\u03bb\u03c5\u03c3\u03af\u03b4\u03b1\u03c2 \u03ba\u03b1\u03b8\u03bf\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf\u03bd \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03ba\u03ac\u03b8\u03b5 invocation, \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03b1\u03c0\u03cc \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03ba\u03b5\u03af\u03bc\u03b5\u03bd\u03bf.<\/p>\n<h2 id=\"dyskolia-kai-ceiling\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03af\u03b1 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03ba\u03bf\u03b9\u03bd\u03cc ceiling<\/h2>\n<p>\u0397 alignment \u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03bf\u03bd\u03bf\u03c4\u03bf\u03bd\u03b9\u03ba\u03ac \u03b1\u03c0\u03cc easy \u03c3\u03b5 hard. \u03a7\u03c9\u03c1\u03af\u03c2 ground truth, \u03b7 \u03c0\u03c4\u03ce\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 1,5 \u03c6\u03bf\u03c1\u03ad\u03c2 \u03c4\u03b1\u03c7\u03cd\u03c4\u03b5\u03c1\u03b7. \u03a3\u03c4\u03b1 hard queries, \u03c4\u03bf frontier model \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03ba\u03c4\u03ac \u03b1\u03c0\u03b5\u03c1\u03b9\u03cc\u03c1\u03b9\u03c3\u03c4\u03bf \u03c0\u03bb\u03b5\u03bf\u03bd\u03ad\u03ba\u03c4\u03b7\u03bc\u03b1, \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03b1\u03c4\u03b1\u03c3\u03ba\u03b5\u03c5\u03ac\u03c3\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03b5\u03be\u03b1\u03c1\u03c4\u03ce\u03bc\u03b5\u03bd\u03b7 \u03c1\u03bf\u03ae \u03bc\u03cc\u03bd\u03bf \u03b1\u03c0\u03cc \u03c4\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03b1 tool definitions.<\/p>\n<p>\u03a4\u03bf \u03b5\u03cd\u03c1\u03b7\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03b7\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03cc \u03b3\u03b9\u03b1 budgeting. \u0391\u03bd \u03b7 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03b4\u03b5\u03bd \u03ba\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 execution traces \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03ad\u03c7\u03b5\u03b9 deterministic validators, \u03ad\u03bd\u03b1 \u03b1\u03ba\u03c1\u03b9\u03b2\u03cc\u03c4\u03b5\u03c1\u03bf judge \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03c3\u03b5\u03b9 \u03bb\u03af\u03b3\u03bf \u03c4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b1\u03c6\u03b1\u03b9\u03c1\u03ad\u03c3\u03b5\u03b9 \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc blind spot. \u03a3\u03c5\u03c7\u03bd\u03ac \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b1\u03be\u03af\u03b1 \u03b4\u03af\u03bd\u03b5\u03b9 \u03b7 \u03b5\u03c0\u03ad\u03bd\u03b4\u03c5\u03c3\u03b7 \u03c3\u03b5 observability, trace storage, failure taxonomy \u03ba\u03b1\u03b9 test cases \u03c0\u03bf\u03c5 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03c5\u03bd \u03c4\u03b9\u03c2 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b5\u03be\u03b1\u03c1\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c4\u03b7\u03c2 \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ae\u03c2.<\/p>\n<p>\u0397 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03c3\u03b7\u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03bf \u03b1\u03ba\u03c1\u03b9\u03b2\u03ad\u03c2 \u03cd\u03c8\u03bf\u03c2 \u03c4\u03bf\u03c5 ceiling \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03bd \u03bc\u03ad\u03c1\u03b5\u03b9 prompt-dependent \u03b3\u03b9\u03b1 \u03b1\u03c3\u03b8\u03b5\u03bd\u03ad\u03c3\u03c4\u03b5\u03c1\u03bf\u03c5\u03c2 generators. \u0391\u03c5\u03c4\u03cc \u03b1\u03c0\u03bf\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03c5\u03c0\u03b5\u03c1\u03b2\u03bf\u03bb\u03b9\u03ba\u03ac \u03b1\u03c0\u03cc\u03bb\u03c5\u03c4\u03b7 \u03b3\u03b5\u03bd\u03af\u03ba\u03b5\u03c5\u03c3\u03b7. \u03a4\u03bf \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03cc \u03c3\u03c5\u03bc\u03c0\u03ad\u03c1\u03b1\u03c3\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03af\u03b1 \u03ba\u03b1\u03b9 \u03b7 \u03b4\u03bf\u03bc\u03ae \u03c4\u03bf\u03c5 task \u03b5\u03c0\u03b7\u03c1\u03b5\u03ac\u03b6\u03bf\u03c5\u03bd \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1, \u03cc\u03c7\u03b9 \u03cc\u03c4\u03b9 \u03ba\u03ac\u03b8\u03b5 future judge \u03b8\u03b1 \u03c3\u03c4\u03b1\u03bc\u03b1\u03c4\u03ac \u03c5\u03c0\u03bf\u03c7\u03c1\u03b5\u03c9\u03c4\u03b9\u03ba\u03ac \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03cc.<\/p>\n<h2 id=\"ground-truth-anchoring\">\u03a0\u03cc\u03c4\u03b5 \u03c4\u03bf ground truth \u03b2\u03bf\u03b7\u03b8\u03ac \u03ae \u03b1\u03b3\u03ba\u03c5\u03c1\u03ce\u03bd\u03b5\u03b9<\/h2>\n<p>\u0397 \u03c0\u03b1\u03c1\u03bf\u03c7\u03ae ground truth \u03b4\u03b5\u03bd \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b5 \u03cc\u03bb\u03bf\u03c5\u03c2 \u03c4\u03bf\u03c5\u03c2 judges. \u03a3\u03c4\u03bf aggregate \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1, \u03bc\u03b5\u03af\u03c9\u03c3\u03b5 \u03c4\u03b7\u03bd alignment \u03c4\u03bf\u03c5 GPT-5.4 \u03ba\u03b1\u03c4\u03ac 1,5 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03bf\u03c5 Gemini-2.5-Pro \u03ba\u03b1\u03c4\u03ac 3,9 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2. \u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03b8\u03b5\u03c9\u03c1\u03bf\u03cd\u03bd \u03c4\u03bf \u03bc\u03bf\u03c4\u03af\u03b2\u03bf \u03c3\u03c5\u03bc\u03b2\u03b1\u03c4\u03cc \u03bc\u03b5 over-anchoring: \u03bf judge \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c0\u03c1\u03bf\u03c3\u03ba\u03bf\u03bb\u03bb\u03b7\u03b8\u03b5\u03af \u03c3\u03c4\u03bf reference \u03b1\u03bd\u03c4\u03af \u03bd\u03b1 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03ae\u03c3\u03b5\u03b9 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03b1 \u03c4\u03b7\u03bd \u03c0\u03c1\u03cc\u03b2\u03bb\u03b5\u03c8\u03b7.<\/p>\n<p>\u03a4\u03bf \u03b5\u03cd\u03c1\u03b7\u03bc\u03b1 \u03b4\u03b5\u03bd \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03c1\u03cd\u03b2\u03b5\u03c4\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03c4\u03bf reference. \u039c\u03b5 ground truth, \u03c4\u03bf QwQ-32B \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03bf\u03cd\u03c3\u03b5 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03b5 \u03c4\u03bf\u03bd programmatic scorer. \u03a3\u03c4\u03bf \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf validation subset, \u03cc\u03bc\u03c9\u03c2, \u03c4\u03bf GPT-OSS-120B \u03ae\u03c4\u03b1\u03bd \u03c4\u03bf \u03c0\u03b9\u03bf human-aligned. \u03a0\u03c1\u03b9\u03bd \u03bc\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c7\u03b1\u03c1\u03b1\u03ba\u03c4\u03b7\u03c1\u03af\u03c3\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u00ab\u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf judge\u00bb, \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bf\u03c1\u03af\u03c3\u03b5\u03b9 \u03c0\u03bf\u03b9\u03b1 \u03ad\u03bd\u03bd\u03bf\u03b9\u03b1 \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03bc\u03b5\u03c4\u03c1\u03ac \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03be\u03b5\u03b9 \u03b1\u03bd \u03c4\u03bf reference \u03ad\u03c7\u03b5\u03b9 \u03b4\u03b9\u03ba\u03cc \u03c4\u03bf\u03c5 bias \u03ae \u03b1\u03bc\u03c6\u03b9\u03c3\u03b7\u03bc\u03af\u03b1.<\/p>\n<div class=\"td-decision-band\">\n<p class=\"td-decision-kicker\">\u039a\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2<\/p>\n<p class=\"td-decision-title\"><strong>\u03a4\u03bf reference \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7\u03c2, \u03cc\u03c7\u03b9 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b7 \u03b1\u03bb\u03ae\u03b8\u03b5\u03b9\u03b1<\/strong><\/p>\n<p>\u0394\u03bf\u03ba\u03b9\u03bc\u03ac\u03c3\u03c4\u03b5 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf judge \u03bc\u03b5 \u03ba\u03b1\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 reference, \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b5 \u03c4\u03b9\u03c2 \u03b4\u03b9\u03b1\u03c6\u03c9\u03bd\u03af\u03b5\u03c2 \u03bc\u03b5 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf subset \u03ba\u03b1\u03b9 \u03ba\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 \u03c0\u03cc\u03c4\u03b5 \u03b7 \u03c0\u03c1\u03cc\u03c3\u03b8\u03b5\u03c4\u03b7 \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03b5\u03b9 \u03ae \u03c0\u03b1\u03c1\u03b1\u03bc\u03bf\u03c1\u03c6\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03ba\u03c1\u03af\u03c3\u03b7. \u03a4\u03bf calibration \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf \u03b4\u03b9\u03ba\u03cc \u03c3\u03b1\u03c2 use case.<\/p>\n<\/div>\n<h2 id=\"rubrics-kai-prompts\">\u0393\u03b9\u03b1\u03c4\u03af \u03bf\u03b9 structured rubrics \u03b5\u03af\u03bd\u03b1\u03b9 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf\u03c2 \u03bc\u03bf\u03c7\u03bb\u03cc\u03c2<\/h2>\n<p>\u0397 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae temperature \u03ba\u03b1\u03b9 \u03b7 \u03c5\u03c0\u03bf\u03c7\u03c1\u03b5\u03c9\u03c4\u03b9\u03ba\u03ae chain-of-thought \u03b1\u03b9\u03c4\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03b5\u03af\u03c7\u03b1\u03bd \u03b1\u03bc\u03b5\u03bb\u03b7\u03c4\u03ad\u03b1 \u03b5\u03c0\u03af\u03b4\u03c1\u03b1\u03c3\u03b7 \u03c3\u03c4\u03b1 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ac ablations. \u03a3\u03c4\u03bf QwQ-32B, \u03c4\u03bf chain-of-thought \u03ad\u03b4\u03c9\u03c3\u03b5 \u03bc\u03ad\u03c3\u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03bc\u03cc\u03bb\u03b9\u03c2 +0,11 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2 \u03bc\u03b5 ground truth, \u03b5\u03bd\u03ce \u03b7 judge temperature \u03c0\u03b1\u03c1\u03ae\u03b3\u03b1\u03b3\u03b5 \u03c3\u03c7\u03b5\u03b4\u03cc\u03bd \u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03b5\u03c2 \u03ba\u03b1\u03bc\u03c0\u03cd\u03bb\u03b5\u03c2.<\/p>\n<p>\u0397 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7 \u03c0\u03b1\u03c1\u03ad\u03bc\u03b2\u03b1\u03c3\u03b7 \u03ae\u03c4\u03b1\u03bd \u03b7 \u03b4\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b7 rubric. \u038c\u03c4\u03b1\u03bd \u03c4\u03bf prompt \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03c3\u03b5 \u03be\u03b5\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ae \u03ba\u03c1\u03af\u03c3\u03b7 \u03b3\u03b9\u03b1 tool selection, parameter structure, sequence accuracy \u03ba\u03b1\u03b9 query coverage, \u03bc\u03af\u03b1 judge-generator \u03c3\u03cd\u03b6\u03b5\u03c5\u03be\u03b7 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03b8\u03b7\u03ba\u03b5 \u03ba\u03b1\u03c4\u03ac 4,8 \u03ad\u03c9\u03c2 6,5 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2 \u03bc\u03b5 ground truth. \u0397 rubric \u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf\u03bd \u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03bf \u03bd\u03b1 \u03c3\u03c5\u03bc\u03c0\u03c5\u03ba\u03bd\u03c9\u03b8\u03b5\u03af \u03cc\u03bb\u03b7 \u03b7 \u03c1\u03bf\u03ae \u03c3\u03b5 \u03bc\u03af\u03b1 \u03b3\u03b5\u03bd\u03b9\u03ba\u03ae \u03b5\u03bd\u03c4\u03cd\u03c0\u03c9\u03c3\u03b7.<\/p>\n<p>\u0397 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7 \u03b4\u03b5\u03bd \u03ae\u03c4\u03b1\u03bd \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03ae. \u03a3\u03b5 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03b7 \u03c3\u03cd\u03b6\u03b5\u03c5\u03be\u03b7 \u03ae\u03c4\u03b1\u03bd \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b1 hard queries \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03c3\u03b5 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03c3\u03c4\u03c1\u03b1\u03c6\u03b5\u03af. \u0395\u03c0\u03bf\u03bc\u03ad\u03bd\u03c9\u03c2, \u03c4\u03bf evaluation prompt \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03bf\u03c5 \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 version control, regression tests \u03ba\u03b1\u03b9 change approval. \u0397 \u03af\u03b4\u03b9\u03b1 \u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03b9\u03c3\u03c7\u03cd\u03b5\u03b9 \u03ba\u03b1\u03b9 \u03b3\u03b9\u03b1 \u03c4\u03bf agent harness: \u03cc\u03c0\u03c9\u03c2 \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03b7 \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b3\u03b9\u03b1 <a href=\"https:\/\/twodots.gr\/coding-agents-idio-ai-diaforetiko-harness\/\">coding agents \u03bc\u03b5 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b1\u03bb\u03bb\u03ac \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc harness<\/a>, \u03b7 \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03c5\u03c3\u03b1 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03c4\u03b7\u03bd \u03b1\u03c0\u03cc\u03b4\u03bf\u03c3\u03b7.<\/p>\n<h2 id=\"dag-topologies\">\u03a4\u03b9 \u03b1\u03c0\u03bf\u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03bf\u03c5\u03bd \u03bf\u03b9 DAG topologies<\/h2>\n<p>\u039f\u03b9 \u03ad\u03be\u03b9 topologies \u03c4\u03bf\u03c5 benchmark \u03b5\u03af\u03bd\u03b1\u03b9 linear, fan-out, fan-in, diamond, optional enrichment \u03ba\u03b1\u03b9 loop-like. \u03a3\u03b5 linear \u03c1\u03bf\u03ae \u03c4\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03bf\u03cd\u03bd \u03b1\u03c0\u03bb\u03ae \u03c3\u03b5\u03b9\u03c1\u03ac. \u03a3\u03b5 fan-out, \u03bc\u03af\u03b1 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1 \u03c4\u03c1\u03bf\u03c6\u03bf\u03b4\u03bf\u03c4\u03b5\u03af \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03b1 branches. \u03a3\u03b5 fan-in, \u03c0\u03bf\u03bb\u03bb\u03ac \u03c0\u03c1\u03bf\u03b7\u03b3\u03bf\u03cd\u03bc\u03b5\u03bd\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c3\u03c5\u03b3\u03ba\u03b5\u03bd\u03c4\u03c1\u03c9\u03b8\u03bf\u03cd\u03bd. \u0397 diamond topology \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03b5\u03b9 \u03b4\u03b9\u03b1\u03ba\u03bb\u03ac\u03b4\u03c9\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bd\u03ad\u03bd\u03c9\u03c3\u03b7, \u03b5\u03bd\u03ce \u03bf\u03b9 \u03b4\u03cd\u03bf \u03c4\u03b5\u03bb\u03b5\u03c5\u03c4\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03c1\u03c6\u03ad\u03c2 \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03bf\u03c5\u03bd \u03c0\u03c1\u03bf\u03b1\u03b9\u03c1\u03b5\u03c4\u03b9\u03ba\u03ae \u03ae \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b7\u03c0\u03c4\u03b9\u03ba\u03ae \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac.<\/p>\n<p>\u0393\u03b9\u03b1 production testing, \u03b7 topology \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03c0\u03b9\u03bf \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03b7 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1\u03bd \u03b3\u03b5\u03bd\u03b9\u03ba\u03cc \u03c7\u03b1\u03c1\u03b1\u03ba\u03c4\u03b7\u03c1\u03b9\u03c3\u03bc\u03cc \u00ab\u03b5\u03cd\u03ba\u03bf\u03bb\u03bf\u00bb \u03ae \u00ab\u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03bf\u00bb. \u0388\u03bd\u03b1\u03c2 commerce agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03b9 \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd, \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03b1\u03c0\u03cc\u03b8\u03b5\u03bc\u03b1, \u03bd\u03b1 \u03c5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03b6\u03b5\u03b9 \u03cc\u03c1\u03bf\u03c5\u03c2 \u03c0\u03c1\u03bf\u03c3\u03c6\u03bf\u03c1\u03ac\u03c2 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b1. \u0388\u03bd\u03b1\u03c2 support agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03c4\u03ac \u03bb\u03bf\u03b3\u03b1\u03c1\u03b9\u03b1\u03c3\u03bc\u03cc, \u03bd\u03b1 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03b9 \u03b9\u03c3\u03c4\u03bf\u03c1\u03b9\u03ba\u03cc, \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 policy \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03ba\u03ac\u03bd\u03b5\u03b9 escalation \u03c5\u03c0\u03cc \u03c0\u03c1\u03bf\u03cb\u03c0\u03bf\u03b8\u03ad\u03c3\u03b5\u03b9\u03c2. \u03a4\u03b1 tests \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b9\u03c3\u03c4\u03bf\u03cd\u03bd \u03b1\u03c5\u03c4\u03ad\u03c2 \u03c4\u03b9\u03c2 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b5\u03be\u03b1\u03c1\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2.<\/p>\n<p>\u0397 \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b1\u03bd\u03ac topology \u03b2\u03bf\u03b7\u03b8\u03ac \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03c3\u03c4\u03b7\u03bd \u03b1\u03c0\u03bf\u03bc\u03cc\u03bd\u03c9\u03c3\u03b7 \u03c4\u03bf\u03c5 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c5\u03c2. \u0391\u03bd \u03bf\u03b9 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b5\u03c2 \u03b1\u03c3\u03c4\u03bf\u03c7\u03af\u03b5\u03c2 \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03b5 fan-in, \u03b7 \u03bb\u03cd\u03c3\u03b7 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf state handling \u03ae dependency validation, \u03cc\u03c7\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc\u03c2 judge. \u0391\u03bd \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03b5 optional enrichment, \u03af\u03c3\u03c9\u03c2 \u03c4\u03bf user intent \u03ae \u03bf\u03b9 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b5\u03c2 \u03bf\u03bb\u03bf\u03ba\u03bb\u03ae\u03c1\u03c9\u03c3\u03b7\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c3\u03b1\u03c6\u03b5\u03af\u03c2.<\/p>\n<h2 id=\"production-evaluation-pipeline\">\u03a0\u03ce\u03c2 \u03c3\u03c4\u03ae\u03bd\u03b5\u03c4\u03b1\u03b9 production evaluation pipeline<\/h2>\n<p>\u0397 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03b1\u03c0\u03cc \u03c4\u03b1 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac workflows \u03c4\u03b7\u03c2 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 \u03b3\u03b5\u03bd\u03b9\u03ba\u03cc benchmark. \u0397 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 inventory \u03c4\u03c9\u03bd tools, typed schemas, permission boundaries, \u03b1\u03bd\u03b1\u03bc\u03b5\u03bd\u03cc\u03bc\u03b5\u03bd\u03b5\u03c2 dependencies \u03ba\u03b1\u03b9 \u03ba\u03b1\u03c4\u03ac\u03bb\u03bf\u03b3\u03bf \u03b5\u03bd\u03b5\u03c1\u03b3\u03b5\u03b9\u03ce\u03bd \u03c0\u03bf\u03c5 \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bd \u03ac\u03bd\u03b8\u03c1\u03c9\u03c0\u03bf. \u039a\u03ac\u03b8\u03b5 test case \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03c1\u03b1\u03c4\u03ac \u03c4\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1, \u03c4\u03bf trace, \u03c4\u03b1 intermediate results, \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc output \u03ba\u03b1\u03b9 \u03c4\u03bf verdict \u03ba\u03ac\u03b8\u03b5 layer.<\/p>\n<p>\u03a3\u03c4\u03b7 \u03c3\u03c5\u03bd\u03ad\u03c7\u03b5\u03b9\u03b1 \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03bf\u03b9 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03c3\u03c4\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc \u03c4\u03b9\u03c2 \u03ba\u03c1\u03af\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03c5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03b7\u03bc\u03b1\u03c3\u03b9\u03bf\u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7. Required arguments, \u03c4\u03cd\u03c0\u03bf\u03b9, \u03c3\u03b5\u03b9\u03c1\u03ac \u03ba\u03b1\u03b9 policy gates \u03b1\u03bd\u03ae\u03ba\u03bf\u03c5\u03bd \u03c3\u03b5 deterministic validators. \u0397 \u03c0\u03bb\u03b7\u03c1\u03cc\u03c4\u03b7\u03c4\u03b1 intent, \u03b7 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bc\u03b9\u03b1\u03c2 \u03b5\u03be\u03ae\u03b3\u03b7\u03c3\u03b7\u03c2 \u03ae \u03bc\u03b9\u03b1 \u03b1\u03bc\u03c6\u03af\u03c3\u03b7\u03bc\u03b7 \u03b5\u03be\u03b1\u03af\u03c1\u03b5\u03c3\u03b7 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03c0\u03b5\u03c1\u03ac\u03c3\u03bf\u03c5\u03bd \u03b1\u03c0\u03cc LLM rubric. \u03a4\u03b1 disagreements \u03ba\u03b1\u03b9 \u03bf\u03b9 high-impact \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03c3\u03b5 reviewer.<\/p>\n<p>\u0388\u03bd\u03b1 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf <a href=\"https:\/\/twodots.gr\/data-visualisation-ux-dashboards-apofaseis\/\">dashboard \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2<\/a> \u03b4\u03b5\u03bd \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03bc\u03cc\u03bd\u03bf \u03ad\u03bd\u03b1\u03bd \u03bc\u03ad\u03c3\u03bf \u03cc\u03c1\u03bf. \u0394\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 pass rate \u03b1\u03bd\u03ac metric, topology, difficulty, tool, model version \u03ba\u03b1\u03b9 risk class, \u03bc\u03b1\u03b6\u03af \u03bc\u03b5 trend \u03ba\u03b1\u03b9 \u03c0\u03b1\u03c1\u03b1\u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1\u03c2. \u0388\u03c4\u03c3\u03b9 \u03b7 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03b3\u03bd\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03b1\u03bd \u03bc\u03b9\u03b1 \u03bd\u03ad\u03b1 \u03ad\u03ba\u03b4\u03bf\u03c3\u03b7 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf \u03c3\u03cd\u03bd\u03bf\u03bb\u03bf \u03b1\u03bb\u03bb\u03ac \u03c5\u03c0\u03bf\u03b2\u03b1\u03b8\u03bc\u03af\u03b6\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03b7 \u03ba\u03b1\u03c4\u03b7\u03b3\u03bf\u03c1\u03af\u03b1.<\/p>\n<div class=\"td-step-list\">\n<p class=\"td-step-list-title\">\u0395\u03c0\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b3\u03b9\u03b1 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 AI agents<\/p>\n<ol>\n<li><span class=\"td-step-kicker\">\u03a7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03ac\u03c6\u03b7\u03c3\u03b7<\/span><strong>\u039a\u03b1\u03c4\u03ad\u03b3\u03c1\u03b1\u03c8\u03b5 \u03c4\u03b1 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac tool workflows<\/strong>\n<p>\u038c\u03c1\u03b9\u03c3\u03b5 tools, parameters, permissions, dependencies \u03ba\u03b1\u03b9 \u03c4\u03b1 \u03c3\u03b7\u03bc\u03b5\u03af\u03b1 \u03cc\u03c0\u03bf\u03c5 \u03bc\u03b9\u03b1 \u03b1\u03c3\u03c4\u03bf\u03c7\u03af\u03b1 \u03b5\u03c0\u03b7\u03c1\u03b5\u03ac\u03b6\u03b5\u03b9 \u03c0\u03b5\u03bb\u03ac\u03c4\u03b7, \u03c7\u03c1\u03ae\u03bc\u03b1\u03c4\u03b1 \u03ae \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1. \u03a4\u03bf test set \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af \u03c3\u03c4\u03b7 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03af\u03b1 \u03c4\u03b7\u03c2 \u03c5\u03c0\u03b7\u03c1\u03b5\u03c3\u03af\u03b1\u03c2.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Traces<\/span><strong>\u0391\u03c0\u03bf\u03b8\u03ae\u03ba\u03b5\u03c5\u03c3\u03b5 \u03c0\u03bb\u03ae\u03c1\u03b7 execution evidence<\/strong>\n<p>\u039a\u03c1\u03ac\u03c4\u03b7\u03c3\u03b5 request, tool calls, arguments, \u03c3\u03b5\u03b9\u03c1\u03ac, intermediate outputs \u03ba\u03b1\u03b9 final response. \u03a7\u03c9\u03c1\u03af\u03c2 trace, \u03bf judge \u03c0\u03c1\u03bf\u03c3\u03c0\u03b1\u03b8\u03b5\u03af \u03bd\u03b1 \u03bc\u03b1\u03bd\u03c4\u03ad\u03c8\u03b5\u03b9 \u03c4\u03b9 \u03c3\u03c5\u03bd\u03ad\u03b2\u03b7.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u039a\u03b1\u03bd\u03cc\u03bd\u03b5\u03c2<\/span><strong>\u039a\u03ac\u03bd\u03b5 deterministic \u03cc,\u03c4\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b9\u03bc\u03bf<\/strong>\n<p>\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b5 validators \u03b3\u03b9\u03b1 schemas, required fields, allowed values, dependencies \u03ba\u03b1\u03b9 policy checks. \u039c\u03b7\u03bd \u03b1\u03bd\u03b1\u03b8\u03ad\u03c4\u03b5\u03b9\u03c2 \u03c3\u03b5 LLM \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03c0\u03bf\u03c5 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b3\u03af\u03bd\u03b5\u03b9 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b1 \u03bc\u03b5 \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Rubric<\/span><strong>\u03a7\u03ce\u03c1\u03b9\u03c3\u03b5 \u03c4\u03b7\u03bd \u03ba\u03c1\u03af\u03c3\u03b7 \u03b1\u03bd\u03ac failure type<\/strong>\n<p>\u0396\u03ae\u03c4\u03b7\u03c3\u03b5 \u03be\u03b5\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03cc verdict \u03b3\u03b9\u03b1 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf\u03c5, \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03bf\u03c5\u03c2, \u03c3\u03b5\u03b9\u03c1\u03ac \u03ba\u03b1\u03b9 \u03ba\u03ac\u03bb\u03c5\u03c8\u03b7 intent. \u0391\u03c0\u03ad\u03c6\u03c5\u03b3\u03b5 \u03ad\u03bd\u03b1 \u03b1\u03b4\u03b9\u03b1\u03c6\u03b1\u03bd\u03ad\u03c2 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03cc score \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03b9\u03c4\u03b9\u03bf\u03bb\u03bf\u03b3\u03af\u03b1 \u03ba\u03b1\u03b9 evidence.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Calibration<\/span><strong>\u0394\u03b7\u03bc\u03b9\u03bf\u03cd\u03c1\u03b3\u03b7\u03c3\u03b5 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf benchmark<\/strong>\n<p>\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b5 \u03b1\u03bd\u03c4\u03b9\u03c0\u03c1\u03bf\u03c3\u03c9\u03c0\u03b5\u03c5\u03c4\u03b9\u03ba\u03cc subset \u03bc\u03b5 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03bf\u03c5\u03c2 reviewers \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1 \u03b5\u03c0\u03af\u03bb\u03c5\u03c3\u03b7\u03c2 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ce\u03bd. \u039c\u03ad\u03c4\u03c1\u03b7\u03c3\u03b5 \u03ba\u03b1\u03b9 \u03c4\u03b7 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 \u03c4\u03c9\u03bd \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03c9\u03bd, \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03bf\u03c5 AI.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Slices<\/span><strong>\u0391\u03bd\u03ac\u03bb\u03c5\u03c3\u03b5 topology, difficulty \u03ba\u03b1\u03b9 risk<\/strong>\n<p>\u03a3\u03cd\u03b3\u03ba\u03c1\u03b9\u03bd\u03b5 linear, fan-in, diamond \u03ba\u03b1\u03b9 optional flows, \u03bc\u03b1\u03b6\u03af \u03bc\u03b5 easy \u03ba\u03b1\u03b9 hard requests. \u0388\u03bd\u03b1\u03c2 \u03ba\u03b1\u03bb\u03cc\u03c2 \u03bc\u03ad\u03c3\u03bf\u03c2 \u03cc\u03c1\u03bf\u03c2 \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03b1\u03bd \u03b1\u03c0\u03bf\u03c4\u03c5\u03b3\u03c7\u03ac\u03bd\u03b5\u03b9 \u03c4\u03bf \u03c0\u03b9\u03bf \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf slice.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u039b\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03af\u03b1<\/span><strong>\u0388\u03bb\u03b5\u03b3\u03be\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc rollout<\/strong>\n<p>\u039a\u03c1\u03ac\u03c4\u03b7\u03c3\u03b5 versioned prompts, models, validators \u03ba\u03b1\u03b9 regression suite. \u039f\u03c0\u03bf\u03b9\u03b1\u03b4\u03ae\u03c0\u03bf\u03c4\u03b5 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03c3\u03c4\u03bf harness \u03ae \u03c3\u03c4\u03b7 rubric \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03bc\u03b5 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03cc test set.<\/p>\n<\/li>\n<\/ol>\n<\/div>\n<p>Die <a href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03af \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03c9\u03bd \u03bc\u03b5 AI<\/a> \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03b9\u03bf \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf\u03b9 \u03cc\u03c4\u03b1\u03bd \u03c4\u03bf evaluation layer \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b1\u03b6\u03af \u03bc\u03b5 \u03c4\u03b7 \u03c1\u03bf\u03ae \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c9\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc\u03c2 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03c2 \u03bc\u03b5\u03c4\u03ac \u03c4\u03bf deployment. Observability, fallback \u03ba\u03b1\u03b9 ownership \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c5\u03c0\u03ac\u03c1\u03c7\u03bf\u03c5\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf pilot.<\/p>\n<h2 id=\"oria-meletis\">\u03a4\u03b1 \u03cc\u03c1\u03b9\u03b1 \u03c4\u03b7\u03c2 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7\u03c2<\/h2>\n<p>\u03a4\u03b1 records \u03b5\u03af\u03bd\u03b1\u03b9 synthetic \u03ba\u03b1\u03b9 single-turn. \u0394\u03b5\u03bd \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03bf\u03c5\u03bd \u03cc\u03bb\u03b5\u03c2 \u03c4\u03b9\u03c2 \u03c3\u03c5\u03bd\u03b8\u03ae\u03ba\u03b5\u03c2 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae\u03c2, \u03cc\u03c0\u03c9\u03c2 network timeouts, stale state, \u03bc\u03b5\u03c4\u03b1\u03b2\u03b1\u03bb\u03bb\u03cc\u03bc\u03b5\u03bd\u03b1 permissions, retries \u03ae \u03b1\u03bb\u03bb\u03b7\u03bb\u03b5\u03c0\u03b9\u03b4\u03c1\u03ac\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03bb\u03bb\u03ce\u03bd \u03b3\u03cd\u03c1\u03c9\u03bd. \u0397 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03b5\u03c0\u03b9\u03ba\u03cd\u03c1\u03c9\u03c3\u03b7 \u03b1\u03c6\u03bf\u03c1\u03ac 120 hard records \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03c5\u03c0\u03bf\u03ba\u03b1\u03b8\u03b9\u03c3\u03c4\u03ac \u03c0\u03bb\u03ae\u03c1\u03b7 human evaluation \u03cc\u03bb\u03bf\u03c5 \u03c4\u03bf\u03c5 corpus.<\/p>\n<p>\u0397 \u03b4\u03b9\u03b1\u03b2\u03ac\u03b8\u03bc\u03b9\u03c3\u03b7 medium \u03b4\u03b5\u03bd \u03ae\u03c4\u03b1\u03bd \u03c0\u03ac\u03bd\u03c4\u03bf\u03c4\u03b5 \u03ba\u03b1\u03b8\u03b1\u03c1\u03ae. \u0397 \u03bc\u03b5\u03c4\u03ac\u03b2\u03b1\u03c3\u03b7 easy \u03c0\u03c1\u03bf\u03c2 medium \u03ba\u03c1\u03af\u03b8\u03b7\u03ba\u03b5 \u03bf\u03bc\u03cc\u03c6\u03c9\u03bd\u03b1 \u03c9\u03c2 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03cc\u03c4\u03b5\u03c1\u03b7 \u03bc\u03cc\u03bd\u03bf \u03c3\u03c4\u03bf 58,1% \u03c4\u03bf\u03c5 stratified validation sample, \u03b5\u03bd\u03ce \u03b7 \u03bc\u03b5\u03c4\u03ac\u03b2\u03b1\u03c3\u03b7 medium \u03c0\u03c1\u03bf\u03c2 hard \u03c3\u03c4\u03bf 93,9%. \u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03bf\u03c5\u03bd \u03c4\u03bf medium \u03c9\u03c2 robustness check \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c9\u03c2 \u03b1\u03c0\u03bf\u03bb\u03cd\u03c4\u03c9\u03c2 \u03ba\u03b1\u03b8\u03b1\u03c1\u03cc \u03c3\u03ba\u03b1\u03bb\u03bf\u03c0\u03ac\u03c4\u03b9 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03af\u03b1\u03c2.<\/p>\n<p>\u03a5\u03c0\u03ac\u03c1\u03c7\u03bf\u03c5\u03bd \u03ba\u03b1\u03b9 model-specific \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03bf\u03af. \u03a4\u03bf GPT-5.4 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af\u03c4\u03b1\u03b9 \u03c9\u03c2 generator, judge \u03ba\u03b1\u03b9 rewrite meta-judge. \u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03b4\u03b5\u03bd \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03b9 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03ae self-preference, \u03b1\u03bb\u03bb\u03ac \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c5\u03c0\u03b5\u03c1\u03c0\u03af\u03c3\u03c4\u03c9\u03c3\u03b7 \u03c3\u03c4\u03b7 sequence accuracy \u03b3\u03b9\u03b1 \u03b4\u03b9\u03ba\u03ad\u03c2 \u03c4\u03bf\u03c5 \u03b3\u03b5\u03bd\u03b9\u03ad\u03c2. \u0395\u03c0\u03b9\u03c0\u03bb\u03ad\u03bf\u03bd, \u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf Azure snapshot \u03b4\u03b5\u03bd \u03b4\u03b9\u03b1\u03b8\u03ad\u03c4\u03b5\u03b9 \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03cc \u03b4\u03b7\u03bc\u03bf\u03c3\u03b9\u03b5\u03c5\u03bc\u03ad\u03bd\u03bf version string, \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03c2 \u03c4\u03b7\u03bd \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03b7 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae.<\/p>\n<p>\u03a4\u03ad\u03bb\u03bf\u03c2, \u03bf NIST AI RMF Generative AI Profile \u03c5\u03c0\u03b5\u03bd\u03b8\u03c5\u03bc\u03af\u03b6\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03bd\u03c4\u03ac\u03c3\u03c3\u03b5\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03ba\u03cd\u03ba\u03bb\u03bf governance, measurement \u03ba\u03b1\u03b9 risk management. \u0388\u03bd\u03b1 benchmark \u03ae \u03ad\u03bd\u03b1\u03c2 judge \u03b4\u03b5\u03bd \u03b1\u03bd\u03c4\u03b9\u03ba\u03b1\u03b8\u03b9\u03c3\u03c4\u03ac \u03c4\u03b7\u03bd \u03c0\u03b1\u03c1\u03b1\u03ba\u03bf\u03bb\u03bf\u03cd\u03b8\u03b7\u03c3\u03b7 \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03c7\u03c1\u03ae\u03c3\u03b7, \u03c4\u03b7\u03bd \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03c9\u03c3\u03b7 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ce\u03bd \u03ba\u03b1\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b1\u03bd\u03b5\u03ba\u03c4\u03af\u03bc\u03b7\u03c3\u03b7 \u03cc\u03c4\u03b1\u03bd \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03ae \u03c4\u03bf \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03cc \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd.<\/p>\n<h2 id=\"symperasma\">\u03a3\u03c5\u03bc\u03c0\u03ad\u03c1\u03b1\u03c3\u03bc\u03b1: \u03bf judge \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03b9\u03c3\u03b8\u03b7\u03c4\u03ae\u03c1\u03b1\u03c2<\/h2>\n<p>\u03a4\u03bf AgentJudgeBench \u03b4\u03b5\u03bd \u03b1\u03ba\u03c5\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf LLM-as-a-judge. \u0394\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c0\u03bf\u03cd \u03c0\u03c1\u03bf\u03c3\u03c6\u03ad\u03c1\u03b5\u03b9 \u03b1\u03be\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03bf\u03cd \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b5\u03c0\u03b9\u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03bf \u03bd\u03b1 \u03b8\u03b5\u03c9\u03c1\u03b7\u03b8\u03b5\u03af \u03bc\u03bf\u03bd\u03b1\u03b4\u03b9\u03ba\u03cc\u03c2 \u03b5\u03bb\u03b5\u03b3\u03ba\u03c4\u03ae\u03c2. \u0388\u03bd\u03b1\u03c2 judge \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03c0\u03b9\u03b8\u03b5\u03c9\u03c1\u03b5\u03af \u03b3\u03c1\u03ae\u03b3\u03bf\u03c1\u03b1 \u03c0\u03bf\u03bb\u03bb\u03ad\u03c2 \u03c1\u03bf\u03ad\u03c2 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03bd\u03c4\u03bf\u03c0\u03af\u03b6\u03b5\u03b9 patterns \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03c7\u03c9\u03c1\u03bf\u03cd\u03bd \u03b5\u03cd\u03ba\u03bf\u03bb\u03b1 \u03c3\u03b5 \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1. \u0397 \u03b1\u03c0\u03cc\u03b4\u03bf\u03c3\u03ae \u03c4\u03bf\u03c5, \u03cc\u03bc\u03c9\u03c2, \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc difficulty, generator, prompt, reference, metric \u03ba\u03b1\u03b9 topology.<\/p>\n<p>\u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7, \u03b7 \u03c0\u03b9\u03bf \u03ce\u03c1\u03b9\u03bc\u03b7 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf\u03bd judge \u03c9\u03c2 \u03b1\u03b9\u03c3\u03b8\u03b7\u03c4\u03ae\u03c1\u03b1 \u03bc\u03ad\u03c3\u03b1 \u03c3\u03b5 \u03c0\u03bf\u03bb\u03c5\u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1. \u039f\u03b9 deterministic validators \u03b5\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5\u03bd \u03cc,\u03c4\u03b9 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bf\u03c1\u03b9\u03c3\u03c4\u03b5\u03af \u03c1\u03b7\u03c4\u03ac, \u03b7 rubric \u03b1\u03c0\u03bf\u03bc\u03bf\u03bd\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b9\u03c2 \u03c3\u03b7\u03bc\u03b1\u03c3\u03b9\u03bf\u03bb\u03bf\u03b3\u03b9\u03ba\u03ad\u03c2 \u03b4\u03b9\u03b1\u03c3\u03c4\u03ac\u03c3\u03b5\u03b9\u03c2, \u03bf\u03b9 \u03ac\u03bd\u03b8\u03c1\u03c9\u03c0\u03bf\u03b9 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03bf\u03cd\u03bd \u03c4\u03b1 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b1 cases \u03ba\u03b1\u03b9 \u03c4\u03bf monitoring \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c0\u03cc\u03c4\u03b5 \u03b7 \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac \u03bc\u03b5\u03c4\u03b1\u03b2\u03ac\u03bb\u03bb\u03b5\u03c4\u03b1\u03b9.<\/p>\n<p>\u0397 \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1 \u03b5\u03bd\u03cc\u03c2 AI agent \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 demo, \u03c4\u03bf \u03bc\u03ad\u03b3\u03b5\u03b8\u03bf\u03c2 \u03b5\u03bd\u03cc\u03c2 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5 \u03ae \u03ad\u03bd\u03b1\u03bd \u03b5\u03bd\u03b9\u03b1\u03af\u03bf AI \u03b2\u03b1\u03b8\u03bc\u03cc. \u0391\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc observable workflows, \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ae\u03c8\u03b9\u03bc\u03b1 tests, \u03c3\u03b1\u03c6\u03ae failure taxonomy \u03ba\u03b1\u03b9 \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03c4\u03c5\u03b3\u03c7\u03ac\u03bd\u03bf\u03c5\u03bd \u03bc\u03b5 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc \u03c4\u03c1\u03cc\u03c0\u03bf. \u0391\u03c5\u03c4\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03cc\u03c0\u03bf\u03c5 \u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03b1\u03cd\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 prompt \u03ba\u03b1\u03b9 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03ba\u03ae \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03bf\u03c2.<\/p>\n<section class=\"td-service-cta\">\n<div class=\"td-service-cta-content\">\n<p class=\"td-service-cta-eyebrow\">\u0391\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03af \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03c9\u03bd &amp; AI<\/p>\n<p class=\"td-service-cta-title\"><strong>\u03a3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 AI agents \u03bc\u03b5 \u03b5\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5\u03c2 \u03c0\u03bf\u03c5 \u03bc\u03b5\u03c4\u03c1\u03b9\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9<\/strong><\/p>\n<p>\u0397 TWO DOTS \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae\u03c3\u03b5\u03b9 \u03c4\u03b1 tool workflows, \u03bd\u03b1 \u03bf\u03c1\u03b3\u03b1\u03bd\u03ce\u03c3\u03b5\u03b9 validation \u03ba\u03b1\u03b9 observability, \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03ad\u03c3\u03b5\u03b9 \u03c4\u03bf\u03bd \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03cc \u03bc\u03b5 \u03c4\u03b1 \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c4\u03b7\u03c2 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7\u03c2 \u03bc\u03b5 \u03c3\u03b1\u03c6\u03ae human fallback.<\/p>\n<div class=\"td-service-cta-actions\"><a class=\"td-service-cta-button\" href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u0394\u03b5\u03af\u03c4\u03b5 \u03c4\u03b9\u03c2 \u03bb\u03cd\u03c3\u03b5\u03b9\u03c2 \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd &amp; AI<\/a><\/div>\n<\/div>\n<\/section>\n<section id=\"sychnes-erotiseis\" class=\"td-faq-section\">\n<div class=\"td-faq\">\n<p class=\"td-faq-heading\">H\u00e4ufig gestellte Fragen<\/p>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 \u03ad\u03bd\u03b1 LLM judge;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03af\u03bd\u03b1\u03b9 \u03ad\u03bd\u03b1 \u03b3\u03bb\u03c9\u03c3\u03c3\u03b9\u03ba\u03cc \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c0\u03bf\u03c5 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03b5\u03af \u03c4\u03b7\u03bd \u03ad\u03be\u03bf\u03b4\u03bf \u03ac\u03bb\u03bb\u03bf\u03c5 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5 \u03ae AI agent. \u03a3\u03c4\u03bf AgentJudgeBench \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 tool selection, parameter structure, sequence accuracy \u03ba\u03b1\u03b9 query coverage.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03cc\u03c3\u03bf \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf AgentJudgeBench;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 3.808 \u03bc\u03bf\u03bd\u03b1\u03b4\u03b9\u03ba\u03ac records, \u03ad\u03be\u03b9 DAG topologies, \u03c4\u03c1\u03af\u03b1 difficulty tiers, \u03c0\u03ad\u03bd\u03c4\u03b5 generators \u03ba\u03b1\u03b9 \u03ad\u03be\u03b9 \u03b2\u03b1\u03c3\u03b9\u03ba\u03bf\u03cd\u03c2 judges, \u03bc\u03b5 321.648 paired \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03ae\u03c3\u03b5\u03b9\u03c2 \u03bc\u03b5 \u03ba\u03b1\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 ground truth.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03ba\u03c1\u03af\u03c3\u03b7 \u03c7\u03c9\u03c1\u03af\u03c2 ground truth \u03b5\u03af\u03bd\u03b1\u03b9 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03cc\u03c4\u03b5\u03c1\u03b7;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039f judge \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03b1\u03c4\u03b1\u03c3\u03ba\u03b5\u03c5\u03ac\u03c3\u03b5\u03b9 \u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03af\u03b1 \u03bc\u03cc\u03bd\u03bf \u03b1\u03c0\u03cc \u03c4\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03b1 tool schemas. \u03a3\u03c4\u03b1 hard no-GT queries \u03bf\u03b9 \u03ad\u03be\u03b9 judges \u03c3\u03c5\u03b3\u03ba\u03bb\u03af\u03bd\u03bf\u03c5\u03bd \u03c3\u03b5 alignment \u03b1\u03c0\u03cc 77 \u03ad\u03c9\u03c2 82 \u03c4\u03bf\u03b9\u03c2 \u03b5\u03ba\u03b1\u03c4\u03cc \u03bc\u03b5 \u03c4\u03bf\u03bd programmatic reference.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0392\u03bf\u03b7\u03b8\u03ac \u03c0\u03ac\u03bd\u03c4\u03b1 \u03b7 \u03c0\u03b1\u03c1\u03bf\u03c7\u03ae ground truth;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u03a3\u03c4\u03b7 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03bc\u03b5\u03af\u03c9\u03c3\u03b5 \u03c4\u03b7\u03bd aggregate alignment \u03b3\u03b9\u03b1 \u03b4\u03cd\u03bf frontier judges, \u03b5\u03cd\u03c1\u03b7\u03bc\u03b1 \u03c3\u03c5\u03bc\u03b2\u03b1\u03c4\u03cc \u03bc\u03b5 over-anchoring. \u0397 \u03b5\u03c0\u03af\u03b4\u03c1\u03b1\u03c3\u03b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03b9\u03ad\u03c4\u03b1\u03b9 \u03b1\u03bd\u03ac \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf, prompt \u03ba\u03b1\u03b9 use case.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03bf chain-of-thought \u03ad\u03ba\u03b1\u03bd\u03b5 \u03c4\u03bf\u03c5\u03c2 judges \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c5\u03c2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a3\u03c4\u03bf \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03cc ablation \u03b7 \u03bc\u03ad\u03c3\u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03bc\u03b5 ground truth \u03ae\u03c4\u03b1\u03bd \u03bc\u03cc\u03bb\u03b9\u03c2 +0,11 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03b7 \u03bc\u03ad\u03b3\u03b9\u03c3\u03c4\u03b7 \u03b1\u03bd\u03ac cell 0,3 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2. \u0397 \u03b5\u03c0\u03af\u03b4\u03c1\u03b1\u03c3\u03b7 \u03b8\u03b5\u03c9\u03c1\u03ae\u03b8\u03b7\u03ba\u03b5 \u03b1\u03bc\u03b5\u03bb\u03b7\u03c4\u03ad\u03b1.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03b1 \u03c0\u03b1\u03c1\u03ad\u03bc\u03b2\u03b1\u03c3\u03b7 \u03b2\u03bf\u03ae\u03b8\u03b7\u03c3\u03b5 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03bf;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0397 structured rubric \u03b1\u03bd\u03ac metric \u03ad\u03b4\u03c9\u03c3\u03b5 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7 4,8 \u03ad\u03c9\u03c2 6,5 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2 \u03c3\u03b5 \u03bc\u03af\u03b1 judge-generator \u03c3\u03cd\u03b6\u03b5\u03c5\u03be\u03b7, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03b3\u03b5\u03bd\u03b9\u03ba\u03b5\u03cd\u03c4\u03b7\u03ba\u03b5 \u03bf\u03bc\u03bf\u03b9\u03cc\u03bc\u03bf\u03c1\u03c6\u03b1 \u03ba\u03b1\u03b9 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03b9\u03ba\u03cc \u03c4\u03b7\u03c2 validation.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03bf\u03c2 judge \u03ae\u03c4\u03b1\u03bd \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039c\u03b5 programmatic reference \u03ba\u03b1\u03b9 ground truth, \u03c4\u03bf QwQ-32B \u03b5\u03af\u03c7\u03b5 \u03c4\u03b7\u03bd \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b7 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1. \u03a3\u03c4\u03bf human validation subset, \u03c4\u03bf GPT-OSS-120B \u03ae\u03c4\u03b1\u03bd \u03c0\u03b9\u03bf human-aligned. \u0397 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf reference \u03ba\u03b1\u03b9 \u03c4\u03bf deployment scenario.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03ce\u03c2 \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 AI agents;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039e\u03b5\u03ba\u03b9\u03bd\u03ac \u03bc\u03b5 \u03c0\u03bb\u03ae\u03c1\u03b7 traces, deterministic validators, rubric \u03b1\u03bd\u03ac failure type, \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf benchmark \u03ba\u03b1\u03b9 metrics \u03b1\u03bd\u03ac topology, difficulty \u03ba\u03b1\u03b9 risk. \u039f LLM judge \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c9\u03c2 \u03ad\u03bd\u03b1 \u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03bf \u03b5\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5, \u03cc\u03c7\u03b9 \u03c9\u03c2 \u03bc\u03bf\u03bd\u03b1\u03b4\u03b9\u03ba\u03ae \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03b1\u03c1\u03c7\u03ae.<\/p>\n<\/div>\n<\/details>\n<\/div>\n<\/section>\n<div class=\"td-source-list\">\n<p id=\"piges\" class=\"td-source-list-title\">Quellen<\/p>\n<ul>\n<li><a href=\"https:\/\/arxiv.org\/html\/2608.26623\" target=\"_blank\" rel=\"noopener\">Verma et al., AgentJudgeBench: A Multi-Difficulty Benchmark for Evaluating LLM Judges on Agentic Tool-Calling<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/datasets\/ServiceNow-AI\/AgentJudgeBench\" target=\"_blank\" rel=\"noopener\">ServiceNow AI, \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf AgentJudgeBench dataset<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/ServiceNow\/SyGra\/tree\/scratch\/agent_judge_bench\/tasks\/agentic_bfcl_judge_eval\" target=\"_blank\" rel=\"noopener\">ServiceNow SyGra, \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1\u03c2 \u03ba\u03b1\u03b9 prompts \u03c4\u03bf\u03c5 AgentJudgeBench<\/a><\/li>\n<li><a href=\"https:\/\/nvlpubs.nist.gov\/nistpubs\/ai\/NIST.AI.600-1.pdf\" target=\"_blank\" rel=\"noopener\">NIST AI 600-1, Generative AI Profile \u03c4\u03bf\u03c5 AI Risk Management Framework<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c4\u03bf AgentJudgeBench \u03b3\u03b9\u03b1 \u03c4\u03bf\u03c5\u03c2 LLM judges \u03ba\u03b1\u03b9 \u03c0\u03ce\u03c2 \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 deterministic checks, rubrics \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf\u03c2 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03c2 \u03c3\u03b5 AI agents.<\/p>","protected":false},"author":1,"featured_media":113640,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_gspb_post_css":"","footnotes":""},"categories":[199],"tags":[6506,18888,18207,19154,20891],"class_list":["post-105068","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-techniti-noimosyni","tag-ai-agents","tag-ai-evaluation","tag-llm-judge","tag-responsible-ai","tag-tool-calling"],"_links":{"self":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts\/105068","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/comments?post=105068"}],"version-history":[{"count":0,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/posts\/105068\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/media\/113640"}],"wp:attachment":[{"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/media?parent=105068"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/categories?post=105068"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/twodots.gr\/de\/wp-json\/wp\/v2\/tags?post=105068"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}