الملخص
We changed the agent: did it actually get better? Every self-improving agent loop answers this hundreds of times, and every answer comes from a verifier. On open-ended tasks none exists, so the loop is handed a hand-written rubric or a bare LLM judge grading output from a model like itself, inviting reward hacking and shared blind spots. We make the verifier the evolving object: an inspectable expression over small, mostly deterministic drawback detectors, synthesized from clustered failures, gated at birth, and selected for agreement with a ten-item anchored reference set plus consensus over unlabeled outputs, never for the agent's score. On MBPP+ it gains +0.21 held-out agreement over the hand-authored seed composition, on every seed, and ends ahead of the bare LLM judge it contains. One finding should change how co-evolved verifiers are validated: removing the anchor guards collapses the verifier into a vacuous always-pass grader, yet that collapsed verifier trains skills just as well. Downstream task score cannot certify a self-evolved verifier. Score does answer sufficiency, and there an evolved verifier can substitute: Double Ratchet, pairing the verifier with a lifecycle-managed skill loop, retains 88-110% of the lift that ground truth or a rubric buys the same loop, across code generation, enterprise text-to-SQL, and reference-free report generation. When evolved skills gamed the report rubric, an outer judge caught it and one added detector repaired it; the judge itself was wrong until given the task contract.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Zhang, X., Wang, G., Cui, Y., Li, Z., Qiu, W., Zhu, B., & He, P. (2026). Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents. https://omanscience.com/ar/articles/who-verifies-the-verifier-co-evolving-inspectable-graders-with-self-improving-agents
MLA 9
Zhang, Xing, et al. "Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents." https://omanscience.com/ar/articles/who-verifies-the-verifier-co-evolving-inspectable-graders-with-self-improving-agents.
شيكاغو (المؤلف–التاريخ)
Zhang, Xing, Guanghui Wang, Yanwei Cui, Ziyuan Li, Wei Qiu, Bing Zhu, and Peiyang He. 2026. "Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents." https://omanscience.com/ar/articles/who-verifies-the-verifier-co-evolving-inspectable-graders-with-self-improving-agents.
هارفارد
Zhang, X., Wang, G., Cui, Y., Li, Z., Qiu, W., Zhu, B. and He, P. (2026) 'Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents', Available at: https://omanscience.com/ar/articles/who-verifies-the-verifier-co-evolving-inspectable-graders-with-self-improving-agents.
فانكوفر
Zhang X, Wang G, Cui Y, Li Z, Qiu W, Zhu B, et al. Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents. https://omanscience.com/ar/articles/who-verifies-the-verifier-co-evolving-inspectable-graders-with-self-improving-agents
IEEE
X. Zhang, G. Wang, Y. Cui, Z. Li, W. Qiu, B. Zhu, and P. He, "Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents," https://omanscience.com/ar/articles/who-verifies-the-verifier-co-evolving-inspectable-graders-with-self-improving-agents.