{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:CMLU2UTAQ2DJ6XU265IJR3HLCF","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":"5fc549e455c477fa9e0de55bb264b961c852ed41497e0edc8e038b015deba7fd","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-09-30T17:55:47Z","title_canon_sha256":"3eaacefe7ae5dbdc157d21d0b60b7578559504beb12791b68f4a4dc74ee294bf"},"schema_version":"1.0","source":{"id":"2509.26619","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2509.26619","created_at":"2026-05-27T01:04:51Z"},{"alias_kind":"arxiv_version","alias_value":"2509.26619v3","created_at":"2026-05-27T01:04:51Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.26619","created_at":"2026-05-27T01:04:51Z"},{"alias_kind":"pith_short_12","alias_value":"CMLU2UTAQ2DJ","created_at":"2026-05-27T01:04:51Z"},{"alias_kind":"pith_short_16","alias_value":"CMLU2UTAQ2DJ6XU2","created_at":"2026-05-27T01:04:51Z"},{"alias_kind":"pith_short_8","alias_value":"CMLU2UTA","created_at":"2026-05-27T01:04:51Z"}],"graph_snapshots":[{"event_id":"sha256:9f6ad4d035212b995769c3d149f9017d86725127aea1e364d8193253fdf1b8df","target":"graph","created_at":"2026-05-27T01:04:51Z","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 epsilon-greedy strategy identifies the most challenging topics while exploring only 6% of the search space -- a 100 times cost reduction over exhaustive evaluation."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"Topic difficulty revealed through sample-and-evaluate queries is robust across independent metrics (GEMBA-SQA and MetricX), languages, and models."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"An epsilon-greedy multi-armed bandit framework automatically discovers challenging internet topics for benchmarks in machine translation and QA, exploring only 6% of the space for 100x cost reduction."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"An epsilon-greedy bandit search over web topics finds the hardest benchmarks after exploring only 6 percent of the space."}],"snapshot_sha256":"26e246d523aab861d310327c3c8fa92ccc31c1b198552413dd891191d174f335"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"af67a604f53c259cee19b61100a6fa4a5cfe13adfa1af17ab010d432ad5aac28"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2509.26619/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Many static benchmarks are beginning to saturate: as models rapidly improve, they achieve near-perfect scores on fixed test sets, leaving little headroom to expose genuine model weaknesses -- and even expert-curated challenge sets quickly saturate after hillclimbing. We present a fully automatic framework that searches the Internet at scale to construct challenging benchmarks without human curation. The key insight is to model the Internet as a vast space of topics and formalize the search as a multi-armed bandit problem, where each topic's difficulty is revealed only through expensive sample-","authors_text":"Daniel Deutsch, Mara Finkelstein, Markus Freitag, Parker Riley, Vil\\'em Zouhar, Wenda Xu","cross_cats":["cs.AI"],"headline":"An epsilon-greedy bandit search over web topics finds the hardest benchmarks after exploring only 6 percent of the space.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-09-30T17:55:47Z","title":"Searching the Internet for Challenging Benchmarks at Scale"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.26619","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-18T11:37:45.823519Z","id":"3153e4dc-123b-49ee-8fc5-f3cca2f80538","model_set":{"reader":"grok-4.3"},"one_line_summary":"An epsilon-greedy multi-armed bandit framework automatically discovers challenging internet topics for benchmarks in machine translation and QA, exploring only 6% of the space for 100x cost reduction.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"An epsilon-greedy bandit search over web topics finds the hardest benchmarks after exploring only 6 percent of the space.","strongest_claim":"Our epsilon-greedy strategy identifies the most challenging topics while exploring only 6% of the search space -- a 100 times cost reduction over exhaustive evaluation.","weakest_assumption":"Topic difficulty revealed through sample-and-evaluate queries is robust across independent metrics (GEMBA-SQA and MetricX), languages, and models."}},"verdict_id":"3153e4dc-123b-49ee-8fc5-f3cca2f80538"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:2d4ac82db54d787e7971c9bda666c4c0ea9b9b304ba35755a0a9f8f1343f09ef","target":"record","created_at":"2026-05-27T01:04:51Z","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":"5fc549e455c477fa9e0de55bb264b961c852ed41497e0edc8e038b015deba7fd","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-09-30T17:55:47Z","title_canon_sha256":"3eaacefe7ae5dbdc157d21d0b60b7578559504beb12791b68f4a4dc74ee294bf"},"schema_version":"1.0","source":{"id":"2509.26619","kind":"arxiv","version":3}},"canonical_sha256":"13174d526086869f5e9af75098eceb116510e657d9834072881b6d36d2eee454","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"13174d526086869f5e9af75098eceb116510e657d9834072881b6d36d2eee454","first_computed_at":"2026-05-27T01:04:51.448585Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-27T01:04:51.448585Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"G687P2YjGcr8DWEHw/cuU/iknoZiL40nhgjR+zgOTK394EJz+NqDHQ+0SuM4B7Yr2HRMmJdEGfI2SliMQQKkBQ==","signature_status":"signed_v1","signed_at":"2026-05-27T01:04:51.449400Z","signed_message":"canonical_sha256_bytes"},"source_id":"2509.26619","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:2d4ac82db54d787e7971c9bda666c4c0ea9b9b304ba35755a0a9f8f1343f09ef","sha256:9f6ad4d035212b995769c3d149f9017d86725127aea1e364d8193253fdf1b8df"],"state_sha256":"94e6dc20b440639e99207bf6f79168772d0f9dafbfa744e6972f3349782367a5"}