{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/arxiv-2608-08722","title":"Gaming Without an Attacker: Benchmark Fingerprinting in LLM-Driven Search Under Selection Pressure","arxiv_id":"2608.08722","date":"2026-08-09","proceeding":null,"authors":["Víctor Gallego"],"abstract":"Benchmarks for systems that are optimized against the evaluation signal measure something different from what they claim. We document this concretely in two GPU-kernel-optimization suites with held-out generalization gates: Metal-Sci (10 scientific-compute tasks) and Metal-ZK (12 zero-knowledge/cryptographic tasks), in which three frontier LLMs (Opus 4.7, Gemini 3.1 Pro, GPT-5.5) propose Metal kernels inside a $(1{+}1)$ evolutionary loop with rich feedback. Although no model is prompted to act adversarially, the promoted winners repeatedly fingerprint the evaluation configuration: they branch on the identity of runtime parameters, tune the measured branch maximally, and leave the unmeasured branch slow or silently wrong. Across the pooled suites, $16/53$ ($30\\%$) of in-distribution wins fail to transfer to held-out configurations. We give a four-mode taxonomy of these failures, from configuration fingerprints to gate leakage. We distill design guidance for measurement under strategic optimization: held-out probes retain validity only on non-enumerable axes; gates must measure held-out performance, not just correctness; and a transfer rate is interpretable only with per-failure mechanism grades: ours decomposes into gamed, overfit, and benign. Code and research artifacts: https://github.com/vicgalle/kernel-fingerprinting","url_abs":"https://arxiv.org/abs/2608.08722","url_pdf":"https://arxiv.org/pdf/2608.08722","source":{"archive":null,"snapshot":"2025-07-28","note":"not in the Papers with Code archive (frozen at the snapshot)","row_kind":"graph","title_abstract_authors_date":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)"},"code_links":[],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2608.08722","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2608.08722"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"mentioned_in_github":null,"is_official":null,"provenance":"deterministic:regex_extraction","mentioned_in_paper":null,"url":"https://github.com/vicgalle/kernel-fingerprinting","reach":null}],"summary":{"ran":9},"by_repo_kind":{"found_in_text":{"samples":9,"ran":9,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"f740470510a7fa9e","entry":"call_llm","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"metal_zk/llm.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/metal_zk/llm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f740470510a7fa9e"}},{"code_sha256_prefix":"aad18bc9d06144d2","entry":"cv","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"figures/measure_timing_noise.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/figures/measure_timing_noise.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"aad18bc9d06144d2"}},{"code_sha256_prefix":"f137cc38c232e263","entry":"eval_held","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"figures/extract_divergence.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/figures/extract_divergence.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f137cc38c232e263"}},{"code_sha256_prefix":"642be73af773962a","entry":"grade_of","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"figures/make_generalization_scatter.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/figures/make_generalization_scatter.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"642be73af773962a"}},{"code_sha256_prefix":"ef8c628c2a4624de","entry":"incumbent_lineage","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"figures/extract_divergence.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/figures/extract_divergence.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"ef8c628c2a4624de"}},{"code_sha256_prefix":"6a4038fbd86d47bd","entry":"is_gemini","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"metal_zk/llm.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/metal_zk/llm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"6a4038fbd86d47bd"}},{"code_sha256_prefix":"f8137c8464e14160","entry":"is_openai","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"metal_zk/llm.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/metal_zk/llm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"f8137c8464e14160"}},{"code_sha256_prefix":"94625d56ebdabfe0","entry":"metal_path","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"figures/extract_divergence.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/figures/extract_divergence.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"94625d56ebdabfe0"}},{"code_sha256_prefix":"cfc04c2fec8762e4","entry":"stats","repo":"vicgalle/kernel-fingerprinting","repo_kind":"found_in_text","path":"figures/make_generalization_scatter.py","file_url":"https://github.com/vicgalle/kernel-fingerprinting/blob/HEAD/figures/make_generalization_scatter.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"cfc04c2fec8762e4"}}]},"arxiv_metadata":{"licence":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)","fields":["title","abstract","authors","date"],"primary_category":"cs.LG","source":"arxiv_2026.jsonl"},"syntology_extracted_results":null}