{"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-2606-16920","title":"Demystifying Variance in Circuit Discovery of LLMs","arxiv_id":"2606.16920","date":"2026-06-15","proceeding":null,"authors":["Frank Zhengqing Wu","Francesco Tonin","Volkan Cevher"],"abstract":"Circuit discovery is a key technique in mechanistic interpretability to pinpoint the model components that are crucial for performing a given task. Although the current state-of-the-art method (EAP-IG) performs well on the metric of (un)faithfulness, it suffers from substantial variability. This includes resampling variance, where the circuit changes when we probe with a new batch of data from the same distribution; rephrasing variance, where the discovered circuit shifts when the prompts are rephrased; and sample-wise variance, where a circuit with low population unfaithfulness exhibits large fluctuations in unfaithfulness across individual samples. This paper studies the roots of these variances. We demonstrate that CEAP, our new circuit discovery method that improves upon EAP-IG with a theoretical guarantee, can substantially lessen resampling variance. We further show that rephrasing variance arises because prompts with different templates tend to activate different circuits in the model. This leads us to argue that it may be challenging to find a comprehensive circuit that explains and controls the model's behavior on a task, which can be expressed in countless templates, suggesting that LLMs may be inherently hard to steer. We show that sparsity, which has been claimed to form more compact and interpretable task circuits, fails to solve this problem. Regarding sample-wise variance, we argue that it is largely benign: extremely poor unfaithfulness scores often stem from how unfaithfulness is defined, rather than from defects in the measured circuits. We show that the magnitude of unfaithfulness is affected by selective contribution scaling, a neural mechanism that accounts for the extremely poor scores sometimes observed.","url_abs":"https://arxiv.org/abs/2606.16920","url_pdf":"https://arxiv.org/pdf/2606.16920","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=2606.16920","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2606.16920"}},"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/LIONS-EPFL/Circuit-Variance","reach":null}],"summary":{"ran":13,"unverified":1},"by_repo_kind":{"found_in_text":{"samples":14,"ran":13,"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":"a163722c56d36ba6","entry":"collate_EAP","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"dataset.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/dataset.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":"a163722c56d36ba6"}},{"code_sha256_prefix":"a386c66aefa26449","entry":"format_model_name","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"visualization/plot_rephrasing_samplewise_variance.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/visualization/plot_rephrasing_samplewise_variance.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":"a386c66aefa26449"}},{"code_sha256_prefix":"ebeddb42ba66079b","entry":"format_model_name","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"visualization/plot_resampling_variance.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/visualization/plot_resampling_variance.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":"ebeddb42ba66079b"}},{"code_sha256_prefix":"3e1fbc61482f7d19","entry":"format_task_name","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"visualization/plot_rephrasing_samplewise_variance.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/visualization/plot_rephrasing_samplewise_variance.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":"3e1fbc61482f7d19"}},{"code_sha256_prefix":"6eb050397dd71ef2","entry":"get_logit_positions","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"metrics.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/metrics.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":"6eb050397dd71ef2"}},{"code_sha256_prefix":"f948655140bd7b45","entry":"get_n_edges_target_list","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"pareto_dev_utils.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/pareto_dev_utils.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":"f948655140bd7b45"}},{"code_sha256_prefix":"e9375a8ddbd853a1","entry":"is_circuit_sparsity_model_identifier","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"circuit_sparse_adapter.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/circuit_sparse_adapter.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":"e9375a8ddbd853a1"}},{"code_sha256_prefix":"495039ab141aa6fa","entry":"js_div","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"metrics.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/metrics.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":"495039ab141aa6fa"}},{"code_sha256_prefix":"8412b8161dd09236","entry":"make_train_test_dataframe","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"dataset.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/dataset.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":"8412b8161dd09236"}},{"code_sha256_prefix":"8cff6cdc74875cc7","entry":"natural_key","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"visualization/plot_rephrasing_samplewise_variance.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/visualization/plot_rephrasing_samplewise_variance.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":"8cff6cdc74875cc7"}},{"code_sha256_prefix":"6f859f08878f0484","entry":"natural_key","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"visualization/plot_resampling_variance.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/visualization/plot_resampling_variance.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":"6f859f08878f0484"}},{"code_sha256_prefix":"75eee1fc5eaccebe","entry":"normalize_sparse_config","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"circuit_sparse_adapter.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/circuit_sparse_adapter.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":"75eee1fc5eaccebe"}},{"code_sha256_prefix":"0964215b60e518f9","entry":"setup_label","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"visualization/plot_resampling_variance.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/visualization/plot_resampling_variance.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":"0964215b60e518f9"}},{"code_sha256_prefix":"90d82934607439cb","entry":"build_result_dir","repo":"LIONS-EPFL/Circuit-Variance","repo_kind":"found_in_text","path":"pareto_dev_utils.py","file_url":"https://github.com/LIONS-EPFL/Circuit-Variance/blob/HEAD/pareto_dev_utils.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"90d82934607439cb"}}]},"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}