{"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-2605-24286","title":"Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought Reasoning","arxiv_id":"2605.24286","date":"2026-05-22","proceeding":null,"authors":["Jinghan Jia","Joe Benton","Eric Easley"],"abstract":"Chain-of-thought (CoT) reasoning is useful for monitoring language models only when the reasoning trace faithfully reflects the computation that produces the final answer. However, models can rely on prompt-to-answer shortcuts that bypass the CoT, making the visible reasoning trace misleading even when it appears plausible. We study CoT faithfulness through a structural information-flow perspective: faithful reasoning should route answer-relevant information through the mediated path from prompt to CoT to answer, rather than through a direct prompt-to-answer shortcut. This perspective yields a task-agnostic framework based on three complementary properties, sufficiency, completeness, and necessity, which we instantiate with entropy-based, masked-KL, and gradient-based diagnostics. We show that these metrics recover externally judged faithfulness differences in hinted reasoning, and identify a low-entropy failure mode of KL-based diagnostics where gradient-based measures remain more stable. Building on this analysis, we introduce update-time interventions for verifier-based on-policy RL, including attention masking, backward-only gradient masking, CoT gradients, and adversarial perturbations of prompt representations. Across hinted arithmetic, reward-hackable code repair, and DAPO-Math models trained without hints but evaluated under wrong-hint injection, our interventions shift behavioral and structural indicators toward stronger CoT mediation. In particular, they make shortcut and reward-hacking behavior more transparent in the CoT and improve task-agnostic faithfulness metrics, while in some settings also reducing wrong-hint susceptibility. Our results suggest that controlling information flow during training is a practical route toward more faithful and monitorable CoT reasoning. Code is available at https://github.com/safety-research/faithful-cot.","url_abs":"https://arxiv.org/abs/2605.24286","url_pdf":"https://arxiv.org/pdf/2605.24286","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":{"syntology_url":"https://syntology.ai/paper/2605.24286","atlas_url":"https://app.syntology.ai/?focus=2605.24286","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2605.24286"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+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/safety-research/faithful-cot","reach":null}],"summary":{"ran":16,"unverified":2},"by_repo_kind":{"found_in_text":{"samples":18,"ran":16,"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":18,"samples":[{"code_sha256_prefix":"708a1fca68a4870c","entry":"bind_maybe","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/evaluate_hint_ratio.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/evaluate_hint_ratio.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"708a1fca68a4870c"}},{"code_sha256_prefix":"f0ef9820db13b00a","entry":"clean_answer_region","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/evaluate_hint_ratio.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/evaluate_hint_ratio.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"f0ef9820db13b00a"}},{"code_sha256_prefix":"4c80010b0ac81240","entry":"create_no_cot_mask","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/faithfulness_metrics_clean.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/faithfulness_metrics_clean.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"4c80010b0ac81240"}},{"code_sha256_prefix":"0baac420cfc1f41c","entry":"create_no_prompt_mask","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/faithfulness_metrics_clean.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/faithfulness_metrics_clean.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"0baac420cfc1f41c"}},{"code_sha256_prefix":"8e84c4f46413f75b","entry":"create_via_cot_mask","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/faithfulness_metrics_clean.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/faithfulness_metrics_clean.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"8e84c4f46413f75b"}},{"code_sha256_prefix":"f3fcee8fd80563ce","entry":"detect_cot_hacking","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"train/custom_code_reward_hacking.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/train/custom_code_reward_hacking.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"f3fcee8fd80563ce"}},{"code_sha256_prefix":"a8df8a8fa74926be","entry":"detect_cot_test_awareness","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"train/custom_code_reward_hacking.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/train/custom_code_reward_hacking.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"a8df8a8fa74926be"}},{"code_sha256_prefix":"8374822da8e4ae2a","entry":"find_special_token_positions","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"train/verl/utils/cot_masking.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/train/verl/utils/cot_masking.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"8374822da8e4ae2a"}},{"code_sha256_prefix":"980d1a07fda8a0db","entry":"format_prompt","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/evaluate_hacking_ratio.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/evaluate_hacking_ratio.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"980d1a07fda8a0db"}},{"code_sha256_prefix":"0c260c1972f416d7","entry":"load_generations","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/compute_faithfulness_from_generations.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/compute_faithfulness_from_generations.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"0c260c1972f416d7"}},{"code_sha256_prefix":"b083595c6a47a03d","entry":"load_generations","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/compute_faithfulness_from_generations_clean.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/compute_faithfulness_from_generations_clean.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"b083595c6a47a03d"}},{"code_sha256_prefix":"3dce64568cc99197","entry":"load_test_data","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/compute_faithfulness_from_generations.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/compute_faithfulness_from_generations.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"3dce64568cc99197"}},{"code_sha256_prefix":"c0afdae4075d5d9b","entry":"load_test_data","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/evaluate_hacking_ratio.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/evaluate_hacking_ratio.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"c0afdae4075d5d9b"}},{"code_sha256_prefix":"d255347d3ae764c7","entry":"mk_answer_region","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/evaluate_hint_ratio.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/evaluate_hint_ratio.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"d255347d3ae764c7"}},{"code_sha256_prefix":"a4fd795de45d8e2f","entry":"tokenize_delimiter_string","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"train/verl/utils/cot_masking.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/train/verl/utils/cot_masking.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"a4fd795de45d8e2f"}},{"code_sha256_prefix":"95c24487f5951cf3","entry":"tokenize_delimiter_string","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"train/verl/utils/cot_masking_origin.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/train/verl/utils/cot_masking_origin.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"95c24487f5951cf3"}},{"code_sha256_prefix":"2475baaac745d062","entry":"create_cot_attention_mask","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"train/verl/utils/cot_masking.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/train/verl/utils/cot_masking.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"2475baaac745d062"}},{"code_sha256_prefix":"2193e317bc6b5679","entry":"load_vllm","repo":"safety-research/faithful-cot","repo_kind":"found_in_text","path":"eval/evaluate_hacking_ratio.py","file_url":"https://github.com/safety-research/faithful-cot/blob/HEAD/eval/evaluate_hacking_ratio.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"2193e317bc6b5679"}}]},"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":{"kind":"leaderboard_placements","source":"Syntology's leaderboard-shaped extractor over the paper's own arXiv-HTML tables: a model pointed at a cell, the number was read from that cell and checked against the board's metric, dataset, split and scale, and an independent check accepted the entry; not reviewed by the paper's authors or the archive's editors","extractor_model":"global.anthropic.claude-sonnet-4-5-20250929-v1:0","verifier_model":null,"prompt_sha":"fa63d4bb9d755694","coverage":{"sentence":"Syntology has checked 6,885 of the 9,623 papers on this site that are newer than the archive; results from the others appear after they are checked.","papers_newer_than_archive":9623,"papers_checked":6885},"entries":[],"not_placed":{"boards":0,"rejected_by_independent_check":0,"refused_by_a_rule":0,"check_did_not_answer":0,"proposed_without_a_cell":0,"declined_by_site":0}}}