{"id":"5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030","title":"A lattice-guided transformer takes on deduction in puzzle benchmarks","argument":"Liam Davis, Leopold Haller, Alberto Alfarano and Mark Santolucito introduce the Lattice Deduction Transformer (LDT), a recurrent transformer designed to approximate logically sound deduction. Between forward passes, it projects its latent state through a lattice; training follows a process modeled on deduction in a search-based constraint solver. The authors report that an 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of the estimated training FLOPs of prior small recurrent reasoners. They describe it as “nearly always sound”: across 1.27M puzzle evaluations, it gave a wrong answer four times, and otherwise returned a correct answer or abstained. A 1.8M-parameter variant reached 99.8% accuracy on Maze-Hard. These results make the paper’s case for a specific approach: combining recurrent transformers with lattice projections and a deduction-oriented training process can perform strongly on the reported puzzle tasks with a small model. The abstract reports results on Sudoku-Extreme and Maze-Hard; it does not establish performance on other kinds of problems.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports benchmark results for Sudoku-Extreme and Maze-Hard; these results alone do not establish performance on other kinds of problems. The authors characterize the model as “nearly always sound,” not always sound, and report four wrong answers among 1.27M puzzle evaluations.","speaker":{"name":"Davis, Liam, Haller, Leopold, Alfarano, Alberto, Santolucito, Mark","roleAtSource":"Article author"},"source":{"title":"Lattice Deduction Transformers","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2605.08605","publishedAt":"2026-05-09","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors introduce LDT, a recurrent transformer that projects its latent state through a lattice between forward passes to approximate logically sound deduction.","excerpt":"a recurrent transformer that approximates logically sound deduction by projecting its latent state through a lattice between forward passes","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"The authors train LDT on-policy using a process modeled on deduction in a search-based constraint solver, with supervision based on an approximation of the set of solution candidates.","excerpt":"We train on-policy in a process that mirrors deduction in a search-based constraint solver","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"An 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of prior small recurrent reasoners’ estimated training FLOPs.","excerpt":"solving $99.999\\%$ of the $422{,}786$ test puzzles","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"The authors report four wrong answers in 1.27M puzzle evaluations; otherwise, the model returned a correct answer or abstained.","excerpt":"it gives a wrong answer for $4$ of $1.27$M puzzle evaluations, and otherwise returns a correct answer or abstains.","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"A 1.8M-parameter LDT variant reached 99.8% accuracy on Maze-Hard.","excerpt":"A $1.8$M-parameter LDT variant reaches $99.8\\%$ accuracy on Maze-Hard.","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2605.08605","checkedAt":"2026-11-10T01:38:07.536Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How does LDT perform on problems beyond Sudoku-Extreme and Maze-Hard?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"A lattice-guided transformer takes on deduction in puzzle benchmarks","summary":"Liam Davis, Leopold Haller, Alberto Alfarano and Mark Santolucito introduce the Lattice Deduction Transformer (LDT), a recurrent transformer designed to approximate logically sound deduction. Between forward passes, it projects its latent state through a lattice; training follows a process modeled on deduction in a search-based constraint solver. The authors report that an 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of the estimated training FLOPs of prior small recurrent reasoners. They describe it as “nearly always sound”: across 1.27M puzzle evaluations, it gave a wrong answer four times, and otherwise returned a correct answer or abstained. A 1.8M-parameter variant reached 99.8% accuracy on Maze-Hard. These results make the paper’s case for a specific approach: combining recurrent transformers with lattice projections and a deduction-oriented training process can perform strongly on the reported puzzle tasks with a small model. The abstract reports results on Sudoku-Extreme and Maze-Hard; it does not establish performance on other kinds of problems.","question":"How does LDT perform on problems beyond Sudoku-Extreme and Maze-Hard?","claims":[{"text":"The authors introduce LDT, a recurrent transformer that projects its latent state through a lattice between forward passes to approximate logically sound deduction.","excerpt":"a recurrent transformer that approximates logically sound deduction by projecting its latent state through a lattice between forward passes","support":"source-stated"},{"text":"The authors train LDT on-policy using a process modeled on deduction in a search-based constraint solver, with supervision based on an approximation of the set of solution candidates.","excerpt":"We train on-policy in a process that mirrors deduction in a search-based constraint solver","support":"source-stated"},{"text":"An 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of prior small recurrent reasoners’ estimated training FLOPs.","excerpt":"solving $99.999\\%$ of the $422{,}786$ test puzzles","support":"source-stated"},{"text":"The authors report four wrong answers in 1.27M puzzle evaluations; otherwise, the model returned a correct answer or abstained.","excerpt":"it gives a wrong answer for $4$ of $1.27$M puzzle evaluations, and otherwise returns a correct answer or abstains.","support":"source-stated"},{"text":"A 1.8M-parameter LDT variant reached 99.8% accuracy on Maze-Hard.","excerpt":"A $1.8$M-parameter LDT variant reaches $99.8\\%$ accuracy on Maze-Hard.","support":"source-stated"}],"limitations":["The abstract reports benchmark results for Sudoku-Extreme and Maze-Hard; these results alone do not establish performance on other kinds of problems.","The authors characterize the model as “nearly always sound,” not always sound, and report four wrong answers among 1.27M puzzle evaluations."],"mediaId":"d7644790be2aecd7000ada43301d0a58810e1b4b988a603a2a4cb5e77812f643","correctionReason":null,"id":"5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030","publicationId":"eeb9f142-324c-465d-8174-1b696e3acf8d","version":1,"sourceUrl":"https://arxiv.org/abs/2605.08605","sourceTitle":"Lattice Deduction Transformers","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-05-09","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:06:33.842Z","createdAt":"2026-11-10T01:07:49.667Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"d7644790be2aecd7000ada43301d0a58810e1b4b988a603a2a4cb5e77812f643","kind":"article","url":"https://arxiv.org/abs/2605.08605","sourceUrl":"https://arxiv.org/abs/2605.08605","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-10T01:03:29.455Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2605.08605","checkedAt":"2026-11-10T01:38:07.536Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"d7644790be2aecd7000ada43301d0a58810e1b4b988a603a2a4cb5e77812f643","confidence":0.93,"provider":"typesafe","model":"jev-1.13.0","inputHash":"15edd136317b1808b1d37915de3164c2bcee2514017c10f33d47fd5b85ca6b1a","decidedAt":"2026-11-10T01:03:30.446Z"},"provenance":{"textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observationId":"a41e4ea50d556f435039c477928fd2b75613f0730b6381d48c355ebf2aad52fb","retrievedAt":"2026-11-10T01:03:29.314Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"61b77478-a361-48cb-8fe3-606a9811c1fe"},"reviewer":{"model":"gpt-6-luna","runId":"1ae29ba0-9907-4be6-8036-ea36311df051"}},"audit":{"status":"passed","reviewedAt":"2026-11-10T01:07:49.667Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Davis, Liam","url":null,"kind":"unknown"},{"name":"Haller, Leopold","url":null,"kind":"unknown"},{"name":"Alfarano, Alberto","url":null,"kind":"unknown"},{"name":"Santolucito, Mark","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Davis, Liam","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Davis, Liam","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}},{"actor":{"name":"Haller, Leopold","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Haller, Leopold","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}},{"actor":{"name":"Alfarano, Alberto","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alfarano, Alberto","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}},{"actor":{"name":"Santolucito, Mark","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Santolucito, Mark","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2605.08605","snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z","basis":"publisher-page-metadata","sourceLinks":48,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030/source"}},"canonicalUrl":"https://escapemeridian.com/readings/a-lattice-guided-transformer-takes-on-deduction-in-puzzle-5a4e12e1","formats":{"html":"/readings/a-lattice-guided-transformer-takes-on-deduction-in-puzzle-5a4e12e1","markdown":"/readings/a-lattice-guided-transformer-takes-on-deduction-in-puzzle-5a4e12e1.md","json":"/readings/a-lattice-guided-transformer-takes-on-deduction-in-puzzle-5a4e12e1.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"},{"id":"term-flops","name":"Floating-point operations","explanation":"A count of basic arithmetic steps; used to measure how much computation went into training a model.","referenceUrl":"https://en.wikipedia.org/wiki/FLOPS"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030","revision":"eeb9f142-324c-465d-8174-1b696e3acf8d","revisionKind":"publication-id","title":"A lattice-guided transformer takes on deduction in puzzle benchmarks","url":"https://escapemeridian.com/readings/a-lattice-guided-transformer-takes-on-deduction-in-puzzle-5a4e12e1"},"source":{"type":"source","id":"https://arxiv.org/abs/2605.08605","revision":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","revisionKind":"sha256-source-text","title":"Lattice Deduction Transformers","url":"https://arxiv.org/abs/2605.08605"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030","title":"A lattice-guided transformer takes on deduction in puzzle benchmarks","argument":"Liam Davis, Leopold Haller, Alberto Alfarano and Mark Santolucito introduce the Lattice Deduction Transformer (LDT), a recurrent transformer designed to approximate logically sound deduction. Between forward passes, it projects its latent state through a lattice; training follows a process modeled on deduction in a search-based constraint solver. The authors report that an 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of the estimated training FLOPs of prior small recurrent reasoners. They describe it as “nearly always sound”: across 1.27M puzzle evaluations, it gave a wrong answer four times, and otherwise returned a correct answer or abstained. A 1.8M-parameter variant reached 99.8% accuracy on Maze-Hard. These results make the paper’s case for a specific approach: combining recurrent transformers with lattice projections and a deduction-oriented training process can perform strongly on the reported puzzle tasks with a small model. The abstract reports results on Sudoku-Extreme and Maze-Hard; it does not establish performance on other kinds of problems.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports benchmark results for Sudoku-Extreme and Maze-Hard; these results alone do not establish performance on other kinds of problems. The authors characterize the model as “nearly always sound,” not always sound, and report four wrong answers among 1.27M puzzle evaluations.","speaker":{"name":"Davis, Liam, Haller, Leopold, Alfarano, Alberto, Santolucito, Mark","roleAtSource":"Article author"},"source":{"title":"Lattice Deduction Transformers","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2605.08605","publishedAt":"2026-05-09","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors introduce LDT, a recurrent transformer that projects its latent state through a lattice between forward passes to approximate logically sound deduction.","excerpt":"a recurrent transformer that approximates logically sound deduction by projecting its latent state through a lattice between forward passes","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"The authors train LDT on-policy using a process modeled on deduction in a search-based constraint solver, with supervision based on an approximation of the set of solution candidates.","excerpt":"We train on-policy in a process that mirrors deduction in a search-based constraint solver","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"An 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of prior small recurrent reasoners’ estimated training FLOPs.","excerpt":"solving $99.999\\%$ of the $422{,}786$ test puzzles","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"The authors report four wrong answers in 1.27M puzzle evaluations; otherwise, the model returned a correct answer or abstained.","excerpt":"it gives a wrong answer for $4$ of $1.27$M puzzle evaluations, and otherwise returns a correct answer or abstains.","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"},{"text":"A 1.8M-parameter LDT variant reached 99.8% accuracy on Maze-Hard.","excerpt":"A $1.8$M-parameter LDT variant reaches $99.8\\%$ accuracy on Maze-Hard.","url":"https://arxiv.org/abs/2605.08605","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2605.08605","checkedAt":"2026-11-10T01:38:07.536Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How does LDT perform on problems beyond Sudoku-Extreme and Maze-Hard?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"A lattice-guided transformer takes on deduction in puzzle benchmarks","summary":"Liam Davis, Leopold Haller, Alberto Alfarano and Mark Santolucito introduce the Lattice Deduction Transformer (LDT), a recurrent transformer designed to approximate logically sound deduction. Between forward passes, it projects its latent state through a lattice; training follows a process modeled on deduction in a search-based constraint solver. The authors report that an 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of the estimated training FLOPs of prior small recurrent reasoners. They describe it as “nearly always sound”: across 1.27M puzzle evaluations, it gave a wrong answer four times, and otherwise returned a correct answer or abstained. A 1.8M-parameter variant reached 99.8% accuracy on Maze-Hard. These results make the paper’s case for a specific approach: combining recurrent transformers with lattice projections and a deduction-oriented training process can perform strongly on the reported puzzle tasks with a small model. The abstract reports results on Sudoku-Extreme and Maze-Hard; it does not establish performance on other kinds of problems.","question":"How does LDT perform on problems beyond Sudoku-Extreme and Maze-Hard?","claims":[{"text":"The authors introduce LDT, a recurrent transformer that projects its latent state through a lattice between forward passes to approximate logically sound deduction.","excerpt":"a recurrent transformer that approximates logically sound deduction by projecting its latent state through a lattice between forward passes","support":"source-stated"},{"text":"The authors train LDT on-policy using a process modeled on deduction in a search-based constraint solver, with supervision based on an approximation of the set of solution candidates.","excerpt":"We train on-policy in a process that mirrors deduction in a search-based constraint solver","support":"source-stated"},{"text":"An 800K-parameter LDT solved 99.999% of 422,786 Sudoku-Extreme test puzzles, using under 2% of prior small recurrent reasoners’ estimated training FLOPs.","excerpt":"solving $99.999\\%$ of the $422{,}786$ test puzzles","support":"source-stated"},{"text":"The authors report four wrong answers in 1.27M puzzle evaluations; otherwise, the model returned a correct answer or abstained.","excerpt":"it gives a wrong answer for $4$ of $1.27$M puzzle evaluations, and otherwise returns a correct answer or abstains.","support":"source-stated"},{"text":"A 1.8M-parameter LDT variant reached 99.8% accuracy on Maze-Hard.","excerpt":"A $1.8$M-parameter LDT variant reaches $99.8\\%$ accuracy on Maze-Hard.","support":"source-stated"}],"limitations":["The abstract reports benchmark results for Sudoku-Extreme and Maze-Hard; these results alone do not establish performance on other kinds of problems.","The authors characterize the model as “nearly always sound,” not always sound, and report four wrong answers among 1.27M puzzle evaluations."],"mediaId":"d7644790be2aecd7000ada43301d0a58810e1b4b988a603a2a4cb5e77812f643","correctionReason":null,"id":"5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030","publicationId":"eeb9f142-324c-465d-8174-1b696e3acf8d","version":1,"sourceUrl":"https://arxiv.org/abs/2605.08605","sourceTitle":"Lattice Deduction Transformers","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-05-09","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:06:33.842Z","createdAt":"2026-11-10T01:07:49.667Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"d7644790be2aecd7000ada43301d0a58810e1b4b988a603a2a4cb5e77812f643","kind":"article","url":"https://arxiv.org/abs/2605.08605","sourceUrl":"https://arxiv.org/abs/2605.08605","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-10T01:03:29.455Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2605.08605","checkedAt":"2026-11-10T01:38:07.536Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"d7644790be2aecd7000ada43301d0a58810e1b4b988a603a2a4cb5e77812f643","confidence":0.93,"provider":"typesafe","model":"jev-1.13.0","inputHash":"15edd136317b1808b1d37915de3164c2bcee2514017c10f33d47fd5b85ca6b1a","decidedAt":"2026-11-10T01:03:30.446Z"},"provenance":{"textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observationId":"a41e4ea50d556f435039c477928fd2b75613f0730b6381d48c355ebf2aad52fb","retrievedAt":"2026-11-10T01:03:29.314Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"61b77478-a361-48cb-8fe3-606a9811c1fe"},"reviewer":{"model":"gpt-6-luna","runId":"1ae29ba0-9907-4be6-8036-ea36311df051"}},"audit":{"status":"passed","reviewedAt":"2026-11-10T01:07:49.667Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Davis, Liam","url":null,"kind":"unknown"},{"name":"Haller, Leopold","url":null,"kind":"unknown"},{"name":"Alfarano, Alberto","url":null,"kind":"unknown"},{"name":"Santolucito, Mark","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Davis, Liam","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Davis, Liam","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}},{"actor":{"name":"Haller, Leopold","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Haller, Leopold","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}},{"actor":{"name":"Alfarano, Alberto","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alfarano, Alberto","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}},{"actor":{"name":"Santolucito, Mark","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Santolucito, Mark","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2605.08605","locator":"head/meta[name=author|citation_author]"},"assertionId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T01:38:07.536Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","sourceUrl":"https://arxiv.org/abs/2605.08605","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2605.08605","snapshotId":"a6d2915a7e5450d3e7f538cf3f78737e967bb58ea599205cd0dced64cc8fc375","contentSha256":"e6e94338c100c5b8ca765f1780661aaf1a663591234b26fe9564af12c5e76bca","textHash":"7a8f1186c59a8d708aeb4117a1bf629bf6a9e54b90edb91a0e301801975da845","observedAt":"2026-11-10T01:38:07.536Z","basis":"publisher-page-metadata","sourceLinks":48,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/5a4e12e1dd6d58bad4741314d81092b8d08b19e3cc78de039229900453326030/source"}}}],"audits":[{"publicationId":"eeb9f142-324c-465d-8174-1b696e3acf8d","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft correctly credits the four listed authors and keeps the results attributed to them as reported outcomes (\"The authors report\", \"They describe it as\").","excerpt":"Authors:Liam Davis, Leopold Haller, Alberto Alfarano, Mark Santolucito"},{"dimension":"summary","status":"pass","explanation":"The draft leads with the actor and the method, then gives the key results and the paper's own scope limits. It keeps the technical terms (recurrent transformer, latent state, lattice, forward passes, FLOPs) exactly, which suits tooltip explanation. It conveys soundness concretely: wrong answers versus correct answers or abstentions. The significance is framed as the paper's case for a small, efficient approach and is bounded to the reported puzzles. The title and question are fair, and the question is a genuine open issue rather than a leading claim.","excerpt":"The model is nearly always sound, in that it gives a wrong answer for $4$ of $1.27$M puzzle evaluations, and otherwise returns a correct answer or abstains."},{"dimension":"date","status":"pass","explanation":"originalPublished is 2026-05-09, which matches the v1 submission date. It is not the 3 Nov 2026 revision or the November feed date.","excerpt":"Submitted on 9 May 2026 (v1), last revised 3 Nov 2026 (this version, v2)"},{"dimension":"support","status":"pass","explanation":"All the figures match the abstract: 800K parameters, 99.999% of 422,786 puzzles, under 2% of estimated training FLOPs, 4 wrong answers in 1.27M evaluations, and 1.8M parameters at 99.8% on Maze-Hard. The quoted phrase \"nearly always sound\" is exact. The draft does not stretch the benchmark results into real-world reliability, and its limitations stay within the abstract's scope.","excerpt":"An $800$K-parameter LDT outperforms prior small recurrent reasoners on Sudoku-Extreme with under $2\\%$ of their estimated training FLOPs"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way to reach the source.","excerpt":null}],"createdAt":"2026-11-10T01:07:49.667Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}