{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:IOAUKH622IANBKDQEPR6F55X5X","short_pith_number":"pith:IOAUKH62","canonical_record":{"source":{"id":"2601.03969","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-07T14:31:07Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"dbdd4b9df4587477ef3565737c066a1bbc88c4d7321146d29860b177aea0af81","abstract_canon_sha256":"d957aed52f06d32163ff066e42dbe1c0311372ef4c98a1318662217d0246d440"},"schema_version":"1.0"},"canonical_sha256":"4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb","source":{"kind":"arxiv","id":"2601.03969","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.03969","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"arxiv_version","alias_value":"2601.03969v2","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.03969","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"pith_short_12","alias_value":"IOAUKH622IAN","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"IOAUKH622IANBKDQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"IOAUKH62","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:IOAUKH622IANBKDQEPR6F55X5X","target":"record","payload":{"canonical_record":{"source":{"id":"2601.03969","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-07T14:31:07Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"dbdd4b9df4587477ef3565737c066a1bbc88c4d7321146d29860b177aea0af81","abstract_canon_sha256":"d957aed52f06d32163ff066e42dbe1c0311372ef4c98a1318662217d0246d440"},"schema_version":"1.0"},"canonical_sha256":"4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:16.700498Z","signature_b64":"h/ICzIZJAOVOOQKT0si+TnyW9kOGuHtNoEIxx3+93THwfiaouPhMH39nkV+nlXxQ8Bx+vXhIbl1PImdO/lTEAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb","last_reissued_at":"2026-05-17T23:39:16.699768Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:16.699768Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2601.03969","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RHgWkkYypxqYMMT5s68wbRjPfkcT2o3h8VhLiexKQHNswD5Ld72w4iD5hgsQfLS6RYErXMPf/XT2HVeBFggLCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T20:54:02.444411Z"},"content_sha256":"74618cc4911ac5e1f4eee06587b597c750cca07cf5a4f6fe7dadc3109f77c62c","schema_version":"1.0","event_id":"sha256:74618cc4911ac5e1f4eee06587b597c750cca07cf5a4f6fe7dadc3109f77c62c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:IOAUKH622IANBKDQEPR6F55X5X","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy.","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Chengqiang Lu, Congxi Xiao, Hui Xiong, Liyi Chen, Qimeng Wang, Tianfu Wang, Wei Wu, Yan Gao, Yao Hu, Yi Wu","submitted_at":"2026-01-07T14:31:07Z","abstract_excerpt":"Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That selectively truncating only the extreme tail of response lengths within correct rollout groups will not degrade the model's ability to learn long-horizon reasoning on complex problems.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Dynamic Outlier Truncation during training reduces token usage by 78% on AIME-24 while increasing accuracy by suppressing extreme-length correct rollouts and adding KL regularization.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f66206928033fa06cff0c990d368df26e106e933d2eacb877dc2a34e331d92d1"},"source":{"id":"2601.03969","kind":"arxiv","version":2},"verdict":{"id":"3688b728-0aba-4c3f-b0e2-f55f95a08c7e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T16:23:21.082758Z","strongest_claim":"our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.","one_line_summary":"Dynamic Outlier Truncation during training reduces token usage by 78% on AIME-24 while increasing accuracy by suppressing extreme-length correct rollouts and adding KL regularization.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That selectively truncating only the extreme tail of response lengths within correct rollout groups will not degrade the model's ability to learn long-horizon reasoning on complex problems.","pith_extraction_headline":"Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2997ff378f28a07e84a11bf6ca19528898003586d92eb0c8c20d5e901d62c722"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"3688b728-0aba-4c3f-b0e2-f55f95a08c7e"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"7ulACWm/z876ZuNviBe+VCqDoL87q8GJak1e+GdIrRQz7nGLRMUkilKigxCVU/C5YeER6fjcyQML9MMYqw2ICA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T20:54:02.445261Z"},"content_sha256":"d3c9f717e0c0af31c3c5c9b6a58314ac94476c80ee0b1588312f457155ca0e87","schema_version":"1.0","event_id":"sha256:d3c9f717e0c0af31c3c5c9b6a58314ac94476c80ee0b1588312f457155ca0e87"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/IOAUKH622IANBKDQEPR6F55X5X/bundle.json","state_url":"https://pith.science/pith/IOAUKH622IANBKDQEPR6F55X5X/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/IOAUKH622IANBKDQEPR6F55X5X/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-27T20:54:02Z","links":{"resolver":"https://pith.science/pith/IOAUKH622IANBKDQEPR6F55X5X","bundle":"https://pith.science/pith/IOAUKH622IANBKDQEPR6F55X5X/bundle.json","state":"https://pith.science/pith/IOAUKH622IANBKDQEPR6F55X5X/state.json","well_known_bundle":"https://pith.science/.well-known/pith/IOAUKH622IANBKDQEPR6F55X5X/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:IOAUKH622IANBKDQEPR6F55X5X","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"d957aed52f06d32163ff066e42dbe1c0311372ef4c98a1318662217d0246d440","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-07T14:31:07Z","title_canon_sha256":"dbdd4b9df4587477ef3565737c066a1bbc88c4d7321146d29860b177aea0af81"},"schema_version":"1.0","source":{"id":"2601.03969","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2601.03969","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"arxiv_version","alias_value":"2601.03969v2","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.03969","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"pith_short_12","alias_value":"IOAUKH622IAN","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"IOAUKH622IANBKDQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"IOAUKH62","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:d3c9f717e0c0af31c3c5c9b6a58314ac94476c80ee0b1588312f457155ca0e87","target":"graph","created_at":"2026-05-17T23:39:16Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That selectively truncating only the extreme tail of response lengths within correct rollout groups will not degrade the model's ability to learn long-horizon reasoning on complex problems."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Dynamic Outlier Truncation during training reduces token usage by 78% on AIME-24 while increasing accuracy by suppressing extreme-length correct rollouts and adding KL regularization."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy."}],"snapshot_sha256":"f66206928033fa06cff0c990d368df26e106e933d2eacb877dc2a34e331d92d1"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2997ff378f28a07e84a11bf6ca19528898003586d92eb0c8c20d5e901d62c722"},"paper":{"abstract_excerpt":"Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivi","authors_text":"Chengqiang Lu, Congxi Xiao, Hui Xiong, Liyi Chen, Qimeng Wang, Tianfu Wang, Wei Wu, Yan Gao, Yao Hu, Yi Wu","cross_cats":["cs.CL"],"headline":"Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-07T14:31:07Z","title":"Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.03969","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-16T16:23:21.082758Z","id":"3688b728-0aba-4c3f-b0e2-f55f95a08c7e","model_set":{"reader":"grok-4.3"},"one_line_summary":"Dynamic Outlier Truncation during training reduces token usage by 78% on AIME-24 while increasing accuracy by suppressing extreme-length correct rollouts and adding KL regularization.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Dynamic outlier truncation during RL training counters length shift in reasoning models, cutting tokens by 78% while boosting accuracy.","strongest_claim":"our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.","weakest_assumption":"That selectively truncating only the extreme tail of response lengths within correct rollout groups will not degrade the model's ability to learn long-horizon reasoning on complex problems."}},"verdict_id":"3688b728-0aba-4c3f-b0e2-f55f95a08c7e"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:74618cc4911ac5e1f4eee06587b597c750cca07cf5a4f6fe7dadc3109f77c62c","target":"record","created_at":"2026-05-17T23:39:16Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"d957aed52f06d32163ff066e42dbe1c0311372ef4c98a1318662217d0246d440","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-01-07T14:31:07Z","title_canon_sha256":"dbdd4b9df4587477ef3565737c066a1bbc88c4d7321146d29860b177aea0af81"},"schema_version":"1.0","source":{"id":"2601.03969","kind":"arxiv","version":2}},"canonical_sha256":"4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4381451fdad200d0a87023e3e2f7b7edd854a0fbd6910255222fc5941321bffb","first_computed_at":"2026-05-17T23:39:16.699768Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:16.699768Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"h/ICzIZJAOVOOQKT0si+TnyW9kOGuHtNoEIxx3+93THwfiaouPhMH39nkV+nlXxQ8Bx+vXhIbl1PImdO/lTEAA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:16.700498Z","signed_message":"canonical_sha256_bytes"},"source_id":"2601.03969","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:74618cc4911ac5e1f4eee06587b597c750cca07cf5a4f6fe7dadc3109f77c62c","sha256:d3c9f717e0c0af31c3c5c9b6a58314ac94476c80ee0b1588312f457155ca0e87"],"state_sha256":"ab92f7a0f175e6fe78b0ada167f6c863570d1928b3cb14275d9898da0681fd49"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"nMz4DaYWYqjPcUKpFihZ4MgyQ/46lNyopwUs51wAdShHsmK+uHSj7CjlNHwyuH8L5/0clBd1u+oOoP+Qs152DQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-27T20:54:02.449275Z","bundle_sha256":"9ea10fec365e1c9972aa9d3472e07f6bf062a321e9a66e36be421fec41fc0d39"}}