{"id":11039,"date":"2026-08-13T18:08:28","date_gmt":"2026-08-13T10:08:28","guid":{"rendered":"https:\/\/infernews.com\/blog\/ncp-bench\/"},"modified":"2026-08-13T18:08:28","modified_gmt":"2026-08-13T10:08:28","slug":"ncp-bench","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/ncp-bench\/","title":{"rendered":"NCP-Bench \u628a\u96fb\u5f71\u6545\u4e8b\u8b8a\u6210\u53ef\u9010\u8f2a\u6aa2\u67e5\u7684\u4e92\u52d5\u74b0\u5883"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-cb97e42f92dc.jpg\" alt=\"NCP-Bench overview: data construction, interaction history, and evaluation framework\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">NCP-Bench \u662f\u4e00\u500b benchmark\uff0c\u52a0\u4e0a\u4e00\u5957\u56fa\u5b9a evaluator\uff0c\u7528\u4f86\u6e2c\u8a66 Interactive Narratives \u88e1\u7684 Narrative Commitment Preservation (NCP)\u3002\u5b83\u8655\u7406\u7684\u554f\u984c\u5f88\u76f4\u63a5\uff1a\u7576\u73a9\u5bb6\u81ea\u7531\u8f38\u5165\u52d5\u4f5c\u6642\uff0cNarrator agent \u6703\u4e0d\u6703\u5077\u6539\u5df2\u5efa\u7acb\u7684\u4e16\u754c\u4e8b\u5be6\uff0c\u6216\u8df3\u904e\u672c\u4f86\u61c9\u8a72\u51fa\u73fe\u7684\u5287\u60c5\u7bc0\u9ede\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f7f\u7528\u65b9\u5f0f\u504f\u5411\u7814\u7a76\u8207\u8a55\u6e2c\u6d41\u7a0b\uff1a\u6bcf\u500b\u96fb\u5f71 synopsis \u6703\u88ab\u8f49\u6210 stateful environment\uff0c\u5167\u6709 initial fact ledger\u3001narrative commitments \u540c ordered reference trajectory\uff1bagent \u6bcf\u56de\u5408\u56de\u61c9\u5f8c\uff0c\u56fa\u5b9a evaluator \u6703\u6aa2\u67e5 fact\u3001commitment \u540c player-input conflicts\uff0c\u518d\u66f4\u65b0\u72c0\u614b\u3002\u5132\u5b58\u5eab\u540c\u6642\u63d0\u4f9b 100 \u500b benchmark environments\u3001\u65b9\u6cd5\u7121\u95dc\u7684 episode runner\uff0c\u4ee5\u53ca Baseline \u548c HiAgent \u53c3\u8003\u65b9\u6cd5\uff0c\u65b9\u4fbf\u76f4\u63a5\u6bd4\u8f03\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u5957\u505a\u6cd5\u7684\u50f9\u503c\u5728\u65bc\uff0c\u5b83\u4e0d\u53ea\u770b\u6587\u5b57\u662f\u5426\u81ea\u7136\uff0c\u9084\u903c\u7cfb\u7d71\u5728\u9577 horizon \u5167\u7dad\u6301\u4e00\u81f4\u6027\u3002\u548c\u53ea\u770b\u55ae\u8f2a\u751f\u6210\u7684\u6e2c\u8a66\u6bd4\uff0cNCP-Bench \u66f4\u63a5\u8fd1\u4e92\u52d5\u5f0f\u6558\u4e8b\u771f\u6b63\u6703\u9047\u5230\u7684\u62c9\u626f\uff1a\u73a9\u5bb6\u53ef\u4ee5\u4e82\u5165\uff0c\u4f46\u7cfb\u7d71\u4ecd\u8981\u5b88\u4f4f\u5df2\u627f\u8afe\u7684\u5287\u60c5\u9aa8\u67b6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8f03\u9069\u5408\u95dc\u5fc3\u9577\u5c0d\u8a71\u6558\u4e8b\u3001\u4e92\u52d5\u5f0f\u6545\u4e8b\u751f\u6210\u3001\u89d2\u8272\u626e\u6f14\u578b agent \u7684\u7814\u7a76\u8005\u8207\u5718\u968a\u3002\u82e5\u8981\u90e8\u7f72\uff0c\u91cd\u9ede\u4e0d\u662f\u8a13\u7df4\u4e00\u500b\u65b0\u6a21\u578b\uff0c\u800c\u662f\u63a5\u4e0a\u65e2\u6709 narrator\uff0c\u518d\u7528 repo \u63d0\u4f9b\u7684\u74b0\u5883\u898f\u683c\u540c evaluator \u8dd1\u6574\u500b episode\uff0c\u89c0\u5bdf\u9055\u53cd\u4e8b\u5be6\u3001\u627f\u8afe\u548c\u5287\u60c5\u9806\u5e8f\u7684\u60c5\u6cc1\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>100 \u500b movie-level environments\uff0c\u6db5\u84cb 18 \u500b genres<\/li>\n<li>\u6bcf\u500b\u74b0\u5883\u90fd\u6709 YAML \u898f\u683c\uff0c\u5305\u542b metadata\u3001facts\u3001commitments \u540c trajectory nodes<\/li>\n<li>\u8a55\u6e2c\u6838\u5fc3\u662f\u9010\u56de\u5408\u6aa2\u67e5\u4e00\u81f4\u6027\uff0c\u800c\u975e\u53ea\u770b\u6700\u7d42\u8f38\u51fa<\/li>\n<li>\u63d0\u4f9b Baseline \u8207 HiAgent\uff0c\u65b9\u4fbf\u548c\u8ad6\u6587\u7d50\u679c\u5c0d\u7167<\/li>\n<li>\u9069\u5408\u505a\u9577\u7bc7\u4e92\u52d5\u6558\u4e8b\u3001Narrative agent \u8207\u4e00\u81f4\u6027\u8a55\u6e2c<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/github.com\/NLP2CT\/NCP-Bench\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9019\u500b\u57fa\u6e96\u628a\u96fb\u5f71\u6545\u4e8b\u8b8a\u6210\u53ef\u9010\u8f2a\u6aa2\u67e5\u7684\u4e92\u52d5\u74b0\u5883\uff0c\u91cd\u9ede\u4e0d\u662f\u8b1b\u5f97\u9806\uff0c\u800c\u662f\u80fd\u5426\u5b88\u4f4f\u65e2\u5b9a\u4e8b\u5be6\u8207\u5287\u60c5\u627f\u8afe\u3002\u4f60\u53ef\u4ee5\u628a\u5b83\u7406\u89e3\u6210\u7528\u4f86\u6e2c\u8a66 LLM agent \u6703\u5426\u5728\u9577\u5c0d\u8a71\u4e2d\u6539\u5beb\u6545\u4e8b\u7684\u5de5\u5177\u3002<\/p>\n","protected":false},"author":8,"featured_media":11038,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,116,197,199],"tags":[],"class_list":["post-11039","post","type-post","status-publish","format-standard","hentry","category-133","category-agentic","category-framework","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11039","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=11039"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11039\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11038"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11039"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11039"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11039"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}