{"id":"789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948","title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation | arXiv cs.AI","argument":"The authors of “Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation” report that Gemma4-12B-it represents arithmetic results from text and images in different internal coordinates. A rotation and rescaling can map the representations onto one another. They use causal interventions—changing internal model states and observing the effects—to test whether the representations share more than a similar appearance. Their experiments find that some directions in the model’s residual stream carry the number without affecting the digit it is about to write. In cases where two numbers share their first digit, changing this hidden information can redirect the next digit while leaving the first unchanged. The authors interpret this as evidence that the model can hold an answer before spelling it out, and that the modalities share numerical meaning through related internal geometry. The findings concern arithmetic tasks in one model; the abstract does not establish whether the same pattern applies to other models, tasks, or practical uses. How far does this shared geometry extend beyond the tested model and arithmetic conditions?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports experiments on Gemma4-12B-it and arithmetic across text or image input, using addition or subtraction; it does not establish whether the findings generalize to other models or tasks. The abstract does not establish how the reported internal representations affect accuracy or reliability in practical uses.","speaker":{"name":"Sabater, Oriol Juan, Morenas, Alex Ferrando de las, Rodríguez, Pau, Suau, Xavier","roleAtSource":"Article author"},"source":{"title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.37422","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors report that text and image arithmetic use different coordinate directions, while a rotation and rescaling maps one representation onto the other with higher causal-effect recovery.","excerpt":"A single rotation and rescaling maps one subspace onto the other and raises recovery to 91.4%, so the conditions share one geometry expressed in different coordinates.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"},{"text":"The authors find that directions the output layer cannot see still carry information about the number, while changing them barely alters the digit being written.","excerpt":"The 13 hidden directions still carry the number, yet changing them barely alters the digit being written.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"},{"text":"In the reported cases where two numbers share their first digit, changing the number represented in hidden directions redirects the next digit in many cases without changing the digit being written.","excerpt":"When two numbers share their first digit, swapping the number in the hidden directions leaves the digit being written intact and redirects the next digit in 77.6% of cases.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"},{"text":"The authors interpret their results as showing that the model forms a numerical representation before producing the answer digit by digit.","excerpt":"The model thus forms an abstract numerical concept upstream and unfolds it across time: multimodal representations share meaning through relational geometry across distinct coordinate frames.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.37422","checkedAt":"2026-10-30T04:40:34.607Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How far does this shared geometry extend beyond the tested model and arithmetic conditions?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation | arXiv cs.AI","summary":"The authors of “Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation” report that Gemma4-12B-it represents arithmetic results from text and images in different internal coordinates. A rotation and rescaling can map the representations onto one another. They use causal interventions—changing internal model states and observing the effects—to test whether the representations share more than a similar appearance. Their experiments find that some directions in the model’s residual stream carry the number without affecting the digit it is about to write. In cases where two numbers share their first digit, changing this hidden information can redirect the next digit while leaving the first unchanged. The authors interpret this as evidence that the model can hold an answer before spelling it out, and that the modalities share numerical meaning through related internal geometry. The findings concern arithmetic tasks in one model; the abstract does not establish whether the same pattern applies to other models, tasks, or practical uses. How far does this shared geometry extend beyond the tested model and arithmetic conditions?","question":"How far does this shared geometry extend beyond the tested model and arithmetic conditions?","claims":[{"text":"The authors report that text and image arithmetic use different coordinate directions, while a rotation and rescaling maps one representation onto the other with higher causal-effect recovery.","excerpt":"A single rotation and rescaling maps one subspace onto the other and raises recovery to 91.4%, so the conditions share one geometry expressed in different coordinates.","support":"source-stated"},{"text":"The authors find that directions the output layer cannot see still carry information about the number, while changing them barely alters the digit being written.","excerpt":"The 13 hidden directions still carry the number, yet changing them barely alters the digit being written.","support":"source-stated"},{"text":"In the reported cases where two numbers share their first digit, changing the number represented in hidden directions redirects the next digit in many cases without changing the digit being written.","excerpt":"When two numbers share their first digit, swapping the number in the hidden directions leaves the digit being written intact and redirects the next digit in 77.6% of cases.","support":"source-stated"},{"text":"The authors interpret their results as showing that the model forms a numerical representation before producing the answer digit by digit.","excerpt":"The model thus forms an abstract numerical concept upstream and unfolds it across time: multimodal representations share meaning through relational geometry across distinct coordinate frames.","support":"source-stated"}],"limitations":["The abstract reports experiments on Gemma4-12B-it and arithmetic across text or image input, using addition or subtraction; it does not establish whether the findings generalize to other models or tasks.","The abstract does not establish how the reported internal representations affect accuracy or reliability in practical uses."],"mediaId":"737aa075f2b47f75420732f5e86cd0e3667fe71b8ac4288cc1f1406eeaaaad70","correctionReason":null,"id":"789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948","publicationId":"47045bb0-df77-479c-8b4c-702c16a66614","version":1,"sourceUrl":"https://arxiv.org/abs/2610.37422","sourceTitle":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T04:42:32.661Z","createdAt":"2026-10-30T02:32:58.047Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"737aa075f2b47f75420732f5e86cd0e3667fe71b8ac4288cc1f1406eeaaaad70","kind":"article","url":"https://arxiv.org/abs/2610.37422","sourceUrl":"https://arxiv.org/abs/2610.37422","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T02:32:17.060Z","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/2610.37422","checkedAt":"2026-10-30T04:40:34.607Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"737aa075f2b47f75420732f5e86cd0e3667fe71b8ac4288cc1f1406eeaaaad70","confidence":0.8400000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"79186c3f92373c99aebd88a3dd241119fc0707d266317b6a8c9b9b0b6074c32f","decidedAt":"2026-10-30T02:32:17.605Z"},"provenance":{"textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observationId":"c31f556a95ee4a695c5e79220f358f30e5dc48d1983cddab64f42a738098e764","retrievedAt":"2026-10-30T02:32:16.980Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"00a19099-de63-4676-8f96-b7d46b17bec8"},"reviewer":{"model":"gpt-6-luna","runId":"ff6d5c5f-e08f-40b7-8a33-fe77740e174a"}},"audit":{"status":"passed","reviewedAt":"2026-10-30T02:32:58.047Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Sabater, Oriol Juan","url":null,"kind":"unknown"},{"name":"Morenas, Alex Ferrando de las","url":null,"kind":"unknown"},{"name":"Rodríguez, Pau","url":null,"kind":"unknown"},{"name":"Suau, Xavier","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Sabater, Oriol Juan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sabater, Oriol Juan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}},{"actor":{"name":"Morenas, Alex Ferrando de las","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Morenas, Alex Ferrando de las","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}},{"actor":{"name":"Rodríguez, Pau","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Rodríguez, Pau","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}},{"actor":{"name":"Suau, Xavier","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Suau, Xavier","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.37422","snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948/source"}},"canonicalUrl":"https://escapemeridian.com/readings/beyond-the-readout-a-latent-numerical-frame-shared-across-789b6434","formats":{"html":"/readings/beyond-the-readout-a-latent-numerical-frame-shared-across-789b6434","markdown":"/readings/beyond-the-readout-a-latent-numerical-frame-shared-across-789b6434.md","json":"/readings/beyond-the-readout-a-latent-numerical-frame-shared-across-789b6434.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.","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":[],"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":"789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948","revision":"47045bb0-df77-479c-8b4c-702c16a66614","revisionKind":"publication-id","title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation | arXiv cs.AI","url":"https://escapemeridian.com/readings/beyond-the-readout-a-latent-numerical-frame-shared-across-789b6434"},"source":{"type":"source","id":"https://arxiv.org/abs/2610.37422","revision":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","revisionKind":"sha256-source-text","title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation","url":"https://arxiv.org/abs/2610.37422"},"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":"789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948","title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation | arXiv cs.AI","argument":"The authors of “Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation” report that Gemma4-12B-it represents arithmetic results from text and images in different internal coordinates. A rotation and rescaling can map the representations onto one another. They use causal interventions—changing internal model states and observing the effects—to test whether the representations share more than a similar appearance. Their experiments find that some directions in the model’s residual stream carry the number without affecting the digit it is about to write. In cases where two numbers share their first digit, changing this hidden information can redirect the next digit while leaving the first unchanged. The authors interpret this as evidence that the model can hold an answer before spelling it out, and that the modalities share numerical meaning through related internal geometry. The findings concern arithmetic tasks in one model; the abstract does not establish whether the same pattern applies to other models, tasks, or practical uses. How far does this shared geometry extend beyond the tested model and arithmetic conditions?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports experiments on Gemma4-12B-it and arithmetic across text or image input, using addition or subtraction; it does not establish whether the findings generalize to other models or tasks. The abstract does not establish how the reported internal representations affect accuracy or reliability in practical uses.","speaker":{"name":"Sabater, Oriol Juan, Morenas, Alex Ferrando de las, Rodríguez, Pau, Suau, Xavier","roleAtSource":"Article author"},"source":{"title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.37422","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors report that text and image arithmetic use different coordinate directions, while a rotation and rescaling maps one representation onto the other with higher causal-effect recovery.","excerpt":"A single rotation and rescaling maps one subspace onto the other and raises recovery to 91.4%, so the conditions share one geometry expressed in different coordinates.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"},{"text":"The authors find that directions the output layer cannot see still carry information about the number, while changing them barely alters the digit being written.","excerpt":"The 13 hidden directions still carry the number, yet changing them barely alters the digit being written.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"},{"text":"In the reported cases where two numbers share their first digit, changing the number represented in hidden directions redirects the next digit in many cases without changing the digit being written.","excerpt":"When two numbers share their first digit, swapping the number in the hidden directions leaves the digit being written intact and redirects the next digit in 77.6% of cases.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"},{"text":"The authors interpret their results as showing that the model forms a numerical representation before producing the answer digit by digit.","excerpt":"The model thus forms an abstract numerical concept upstream and unfolds it across time: multimodal representations share meaning through relational geometry across distinct coordinate frames.","url":"https://arxiv.org/abs/2610.37422","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.37422","checkedAt":"2026-10-30T04:40:34.607Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How far does this shared geometry extend beyond the tested model and arithmetic conditions?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation | arXiv cs.AI","summary":"The authors of “Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation” report that Gemma4-12B-it represents arithmetic results from text and images in different internal coordinates. A rotation and rescaling can map the representations onto one another. They use causal interventions—changing internal model states and observing the effects—to test whether the representations share more than a similar appearance. Their experiments find that some directions in the model’s residual stream carry the number without affecting the digit it is about to write. In cases where two numbers share their first digit, changing this hidden information can redirect the next digit while leaving the first unchanged. The authors interpret this as evidence that the model can hold an answer before spelling it out, and that the modalities share numerical meaning through related internal geometry. The findings concern arithmetic tasks in one model; the abstract does not establish whether the same pattern applies to other models, tasks, or practical uses. How far does this shared geometry extend beyond the tested model and arithmetic conditions?","question":"How far does this shared geometry extend beyond the tested model and arithmetic conditions?","claims":[{"text":"The authors report that text and image arithmetic use different coordinate directions, while a rotation and rescaling maps one representation onto the other with higher causal-effect recovery.","excerpt":"A single rotation and rescaling maps one subspace onto the other and raises recovery to 91.4%, so the conditions share one geometry expressed in different coordinates.","support":"source-stated"},{"text":"The authors find that directions the output layer cannot see still carry information about the number, while changing them barely alters the digit being written.","excerpt":"The 13 hidden directions still carry the number, yet changing them barely alters the digit being written.","support":"source-stated"},{"text":"In the reported cases where two numbers share their first digit, changing the number represented in hidden directions redirects the next digit in many cases without changing the digit being written.","excerpt":"When two numbers share their first digit, swapping the number in the hidden directions leaves the digit being written intact and redirects the next digit in 77.6% of cases.","support":"source-stated"},{"text":"The authors interpret their results as showing that the model forms a numerical representation before producing the answer digit by digit.","excerpt":"The model thus forms an abstract numerical concept upstream and unfolds it across time: multimodal representations share meaning through relational geometry across distinct coordinate frames.","support":"source-stated"}],"limitations":["The abstract reports experiments on Gemma4-12B-it and arithmetic across text or image input, using addition or subtraction; it does not establish whether the findings generalize to other models or tasks.","The abstract does not establish how the reported internal representations affect accuracy or reliability in practical uses."],"mediaId":"737aa075f2b47f75420732f5e86cd0e3667fe71b8ac4288cc1f1406eeaaaad70","correctionReason":null,"id":"789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948","publicationId":"47045bb0-df77-479c-8b4c-702c16a66614","version":1,"sourceUrl":"https://arxiv.org/abs/2610.37422","sourceTitle":"Beyond the Readout: A Latent Numerical Frame Shared Across Modality and Operation","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T04:42:32.661Z","createdAt":"2026-10-30T02:32:58.047Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"737aa075f2b47f75420732f5e86cd0e3667fe71b8ac4288cc1f1406eeaaaad70","kind":"article","url":"https://arxiv.org/abs/2610.37422","sourceUrl":"https://arxiv.org/abs/2610.37422","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T02:32:17.060Z","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/2610.37422","checkedAt":"2026-10-30T04:40:34.607Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"737aa075f2b47f75420732f5e86cd0e3667fe71b8ac4288cc1f1406eeaaaad70","confidence":0.8400000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"79186c3f92373c99aebd88a3dd241119fc0707d266317b6a8c9b9b0b6074c32f","decidedAt":"2026-10-30T02:32:17.605Z"},"provenance":{"textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observationId":"c31f556a95ee4a695c5e79220f358f30e5dc48d1983cddab64f42a738098e764","retrievedAt":"2026-10-30T02:32:16.980Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"00a19099-de63-4676-8f96-b7d46b17bec8"},"reviewer":{"model":"gpt-6-luna","runId":"ff6d5c5f-e08f-40b7-8a33-fe77740e174a"}},"audit":{"status":"passed","reviewedAt":"2026-10-30T02:32:58.047Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Sabater, Oriol Juan","url":null,"kind":"unknown"},{"name":"Morenas, Alex Ferrando de las","url":null,"kind":"unknown"},{"name":"Rodríguez, Pau","url":null,"kind":"unknown"},{"name":"Suau, Xavier","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Sabater, Oriol Juan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sabater, Oriol Juan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}},{"actor":{"name":"Morenas, Alex Ferrando de las","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Morenas, Alex Ferrando de las","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}},{"actor":{"name":"Rodríguez, Pau","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Rodríguez, Pau","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}},{"actor":{"name":"Suau, Xavier","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Suau, Xavier","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.37422","locator":"head/meta[name=author|citation_author]"},"assertionId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-30T04:40:34.607Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","sourceUrl":"https://arxiv.org/abs/2610.37422","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.37422","snapshotId":"4e20337faeab5fc1149240627b8210333961e9b0cd5a60037022c6abc31319b5","contentSha256":"8d1901e153c28c67c34121d44857cbae04d70f1b5ef11c6f2ed1cca97cc203ad","textHash":"0ae9a1ec5bffcf2e35b19420336fa0b63060a0071a2597979f5646b00761f15b","observedAt":"2026-10-30T04:40:34.607Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/789b64342a91162a2acf3db4fc257401a3744591143e08a3d7c3ea52d5a58948/source"}}}],"audits":[{"publicationId":"47045bb0-df77-479c-8b4c-702c16a66614","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft attributes the findings and interpretation to the paper's authors, matching the arXiv abstract. It does not present the results as independently verified.","excerpt":"Authors:Oriol Juan Sabater, Alex Ferrando de las Morenas, Pau Rodríguez, Xavier Suau"},{"dimension":"summary","status":"pass","explanation":"The summary leads with what the authors report and explains causal interventions in plain language. It accurately conveys the three findings and the authors' interpretation, and states the scope limit of one model and arithmetic tasks. 'Residual stream' is not explained, but the surrounding text makes clear it means internal model information. The question is substantive and follows from the evidence.","excerpt":"we learn a small subspace of Gemma4-12B-it's residual stream in which swapping coordinates swaps the model's answer"},{"dimension":"date","status":"pass","explanation":"The displayed original publication date of 2026-10-27 matches the arXiv v1 submission date. The later feed date was correctly not used.","excerpt":"[Submitted on 27 Oct 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is directly supported by the abstract, including the move from 53.2% to 91.4% recovery, the hidden directions, and the 77.6% next-digit redirection. The interpretation is labeled as the authors'. The limitations do not overstate generalization or practical reliability.","excerpt":"Reusing a subspace from text in an image on the same operation recovers only 53.2% of its causal effect."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page of the paper itself, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-10-30T02:32:58.047Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}