{"paper":{"title":"HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench","license":"http://creativecommons.org/licenses/by/4.0/","headline":"HE-SNR measures how models structure uncertainty into low-order states to guide mid-training better than perplexity for software engineering tasks.","cross_cats":["cs.CL","cs.SE"],"primary_cat":"cs.LG","authors_text":"Baolong Bi, Jiawei Fu, Xiaoqing Liu, Xili Wang, Yueyang Wang","submitted_at":"2026-01-28T05:03:24Z","abstract_excerpt":"SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the \"Long-Context Tax\" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucia"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders (reasonable hesitation). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the Entropy Compression Hypothesis is valid and that HE-SNR exhibits stronger correlation with downstream SWE-bench performance than perplexity while avoiding the long-context tax, without post-hoc fitting or selection effects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HE-SNR is a high-entropy signal-to-noise ratio metric derived from the Entropy Compression Hypothesis to better guide LLM mid-training on complex software engineering benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"HE-SNR measures how models structure uncertainty into low-order states to guide mid-training better than perplexity for software engineering tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"300ce2cc0c6106d4b752a0175ba2f9ab40d43c423b5eff4a492300c0bd9dec63"},"source":{"id":"2601.20255","kind":"arxiv","version":3},"verdict":{"id":"34c9e9df-1f2b-4ee4-94d2-72899a88b740","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T10:59:28.388616Z","strongest_claim":"We propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders (reasonable hesitation). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR.","one_line_summary":"HE-SNR is a high-entropy signal-to-noise ratio metric derived from the Entropy Compression Hypothesis to better guide LLM mid-training on complex software engineering benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the Entropy Compression Hypothesis is valid and that HE-SNR exhibits stronger correlation with downstream SWE-bench performance than perplexity while avoiding the long-context tax, without post-hoc fitting or selection effects.","pith_extraction_headline":"HE-SNR measures how models structure uncertainty into low-order states to guide mid-training better than perplexity for software engineering tasks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.20255/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"71535000f829d58d5419211554c3c034a8fa97a301135cabf1e0b3385d56dd56"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}