{"id":"e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6","title":"arXiv cs.AI authors report an offline first-aid assistant with modest scores","argument":"Authors Rezwan Islam Salvi, Saiyma Sittul Muna and Mushfiqur Rahman Mushfique describe HandsOnAid, a phone-based assistant designed to answer first-aid questions without a network connection. They tested seven configurations using 41 question-answer pairs with medically reviewed references following ANZCOR guidelines, and four frontier LLM judges scored the answers. Fine-tuning raised the mean score over the base model, but the best configuration—fine-tuning with retrieval—scored 2.51 out of 5. A medical professional’s blind evaluation of 40 answers produced scores similar to the judges’ averages. The authors also report that a keyword audit found limited coverage of several guideline details required by their rubric, suggesting gaps in training data could help explain low scores. On a Pixel 7 Pro without network access, the fine-tuned 4-bit model produced its first displayable text in an average of 6.86 seconds and a complete answer in 14.88 seconds over ten questions. The paper presents an offline prototype and bounded evaluations, not evidence that its answers are safe or effective in real emergencies.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported scores come from a 41-pair question-answer bank and a blind evaluation of 40 answers; the evidence does not establish performance in real emergency situations. The reported timing is from ten questions on a Pixel 7 Pro with six CPU threads and no network connection.","speaker":{"name":"Salvi, Rezwan Islam, Muna, Saiyma Sittul, Mushfique, Mushfiqur Rahman","roleAtSource":"Article author"},"source":{"title":"HandsOnAid: Building and Evaluating an Offline First-Aid Assistant for Emergency Response Scenarios","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.14059","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors built HandsOnAid as a phone-based first-aid assistant intended to work offline, using a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA.","excerpt":"We built HandsOnAid, a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA dataset","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"The authors evaluated seven configurations with a 41-pair question-answer bank and four LLM judges; the best configuration, fine-tuning with retrieval, scored 2.51 out of 5.","excerpt":"The best configuration scored 2.51 out of 5 which was the fine-tuned 4-bit model with retrieval.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"A medical professional’s blind evaluation of 40 answers produced scores similar to the average scores from the LLM judges.","excerpt":"A medical professional also conducted a blind evaluation of 40 answers across the base, the 4-bit fine-tuned, and the retrieval configurations, and the scores were similar to the LLM judges' average scores.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"The authors report limited coverage of several guideline details in the training corpus and suggest missing information could contribute to low scores.","excerpt":"A keyword audit found limited coverage of several details of ANZCOR guidelines in the training corpus, that are required by the rubric, which suggests that missing information in the training data could be a cause of the low scores.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"In a ten-question test on a Pixel 7 Pro without network access, the fine-tuned 4-bit model took an average of 6.86 seconds to show text and 14.88 seconds to complete an answer.","excerpt":"the 4-bit fine-tuned model produced its first displayable text in 6.86 seconds and a complete answer in 14.88 seconds on average over ten questions.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.14059","checkedAt":"2026-11-12T04:18:10.436Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would the system perform in emergency situations beyond the paper’s question-and-answer evaluations?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI authors report an offline first-aid assistant with modest scores","summary":"Authors Rezwan Islam Salvi, Saiyma Sittul Muna and Mushfiqur Rahman Mushfique describe HandsOnAid, a phone-based assistant designed to answer first-aid questions without a network connection. They tested seven configurations using 41 question-answer pairs with medically reviewed references following ANZCOR guidelines, and four frontier LLM judges scored the answers. Fine-tuning raised the mean score over the base model, but the best configuration—fine-tuning with retrieval—scored 2.51 out of 5. A medical professional’s blind evaluation of 40 answers produced scores similar to the judges’ averages. The authors also report that a keyword audit found limited coverage of several guideline details required by their rubric, suggesting gaps in training data could help explain low scores. On a Pixel 7 Pro without network access, the fine-tuned 4-bit model produced its first displayable text in an average of 6.86 seconds and a complete answer in 14.88 seconds over ten questions. The paper presents an offline prototype and bounded evaluations, not evidence that its answers are safe or effective in real emergencies.","question":"How would the system perform in emergency situations beyond the paper’s question-and-answer evaluations?","claims":[{"text":"The authors built HandsOnAid as a phone-based first-aid assistant intended to work offline, using a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA.","excerpt":"We built HandsOnAid, a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA dataset","support":"source-stated"},{"text":"The authors evaluated seven configurations with a 41-pair question-answer bank and four LLM judges; the best configuration, fine-tuning with retrieval, scored 2.51 out of 5.","excerpt":"The best configuration scored 2.51 out of 5 which was the fine-tuned 4-bit model with retrieval.","support":"source-stated"},{"text":"A medical professional’s blind evaluation of 40 answers produced scores similar to the average scores from the LLM judges.","excerpt":"A medical professional also conducted a blind evaluation of 40 answers across the base, the 4-bit fine-tuned, and the retrieval configurations, and the scores were similar to the LLM judges' average scores.","support":"source-stated"},{"text":"The authors report limited coverage of several guideline details in the training corpus and suggest missing information could contribute to low scores.","excerpt":"A keyword audit found limited coverage of several details of ANZCOR guidelines in the training corpus, that are required by the rubric, which suggests that missing information in the training data could be a cause of the low scores.","support":"source-stated"},{"text":"In a ten-question test on a Pixel 7 Pro without network access, the fine-tuned 4-bit model took an average of 6.86 seconds to show text and 14.88 seconds to complete an answer.","excerpt":"the 4-bit fine-tuned model produced its first displayable text in 6.86 seconds and a complete answer in 14.88 seconds on average over ten questions.","support":"source-stated"}],"limitations":["The reported scores come from a 41-pair question-answer bank and a blind evaluation of 40 answers; the evidence does not establish performance in real emergency situations.","The reported timing is from ten questions on a Pixel 7 Pro with six CPU threads and no network connection."],"mediaId":"5da77e34efba0af21c9d68ad744cfdcdd13250145e2f2a38dacf8a2a734dab64","correctionReason":null,"id":"e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6","publicationId":"f18d86a0-ab05-48ef-8d1f-796a30774ad9","version":1,"sourceUrl":"https://arxiv.org/abs/2611.14059","sourceTitle":"HandsOnAid: Building and Evaluating an Offline First-Aid Assistant for Emergency Response Scenarios","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:06:40.100Z","createdAt":"2026-11-12T03:47:28.535Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"5da77e34efba0af21c9d68ad744cfdcdd13250145e2f2a38dacf8a2a734dab64","kind":"article","url":"https://arxiv.org/abs/2611.14059","sourceUrl":"https://arxiv.org/abs/2611.14059","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T03:46:27.324Z","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/2611.14059","checkedAt":"2026-11-12T04:18:10.436Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"5da77e34efba0af21c9d68ad744cfdcdd13250145e2f2a38dacf8a2a734dab64","confidence":0.59,"provider":"typesafe","model":"jev-1.13.0","inputHash":"f7d78445afe219444215c3d1495d5013d536041d3cfc9e0a9e2b3804c2537a93","decidedAt":"2026-11-12T03:46:29.517Z"},"provenance":{"textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observationId":"7e0703104a59bd48940c21e7e4e3d3ba2a5ba491ba0165edcf959d4304390fce","retrievedAt":"2026-11-12T03:46:26.979Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"a7408383-9306-4dac-8aad-d71e55a86a39"},"reviewer":{"model":"gpt-6-luna","runId":"f053a2dc-ef97-47b9-8c52-f3a9e06f9c48"}},"audit":{"status":"passed","reviewedAt":"2026-11-12T03:47:28.535Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Salvi, Rezwan Islam","url":null,"kind":"unknown"},{"name":"Muna, Saiyma Sittul","url":null,"kind":"unknown"},{"name":"Mushfique, Mushfiqur Rahman","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Salvi, Rezwan Islam","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Salvi, Rezwan Islam","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14059","locator":"head/meta[name=author|citation_author]"},"assertionId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T04:18:10.436Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","sourceUrl":"https://arxiv.org/abs/2611.14059","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z"}},{"actor":{"name":"Muna, Saiyma Sittul","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Muna, Saiyma Sittul","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14059","locator":"head/meta[name=author|citation_author]"},"assertionId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T04:18:10.436Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","sourceUrl":"https://arxiv.org/abs/2611.14059","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z"}},{"actor":{"name":"Mushfique, Mushfiqur Rahman","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Mushfique, Mushfiqur Rahman","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14059","locator":"head/meta[name=author|citation_author]"},"assertionId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T04:18:10.436Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","sourceUrl":"https://arxiv.org/abs/2611.14059","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.14059","snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-authors-report-an-offline-first-aid-assistant-with-e07029bd","formats":{"html":"/readings/arxiv-cs-ai-authors-report-an-offline-first-aid-assistant-with-e07029bd","markdown":"/readings/arxiv-cs-ai-authors-report-an-offline-first-aid-assistant-with-e07029bd.md","json":"/readings/arxiv-cs-ai-authors-report-an-offline-first-aid-assistant-with-e07029bd.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-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"},{"id":"term-fine-tuning","name":"Fine-tuning","explanation":"Further training of an existing model to adapt it to a particular task or use.","referenceUrl":"https://developers.google.com/machine-learning/glossary#fine-tuning"},{"id":"term-cpu","name":"CPU","explanation":"The central processing unit: a general-purpose processor that carries out a computer's instructions.","referenceUrl":"https://aws.amazon.com/what-is/cpu/"}],"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":"e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6","revision":"f18d86a0-ab05-48ef-8d1f-796a30774ad9","revisionKind":"publication-id","title":"arXiv cs.AI authors report an offline first-aid assistant with modest scores","url":"https://escapemeridian.com/readings/arxiv-cs-ai-authors-report-an-offline-first-aid-assistant-with-e07029bd"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.14059","revision":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","revisionKind":"sha256-source-text","title":"HandsOnAid: Building and Evaluating an Offline First-Aid Assistant for Emergency Response Scenarios","url":"https://arxiv.org/abs/2611.14059"},"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":"e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6","title":"arXiv cs.AI authors report an offline first-aid assistant with modest scores","argument":"Authors Rezwan Islam Salvi, Saiyma Sittul Muna and Mushfiqur Rahman Mushfique describe HandsOnAid, a phone-based assistant designed to answer first-aid questions without a network connection. They tested seven configurations using 41 question-answer pairs with medically reviewed references following ANZCOR guidelines, and four frontier LLM judges scored the answers. Fine-tuning raised the mean score over the base model, but the best configuration—fine-tuning with retrieval—scored 2.51 out of 5. A medical professional’s blind evaluation of 40 answers produced scores similar to the judges’ averages. The authors also report that a keyword audit found limited coverage of several guideline details required by their rubric, suggesting gaps in training data could help explain low scores. On a Pixel 7 Pro without network access, the fine-tuned 4-bit model produced its first displayable text in an average of 6.86 seconds and a complete answer in 14.88 seconds over ten questions. The paper presents an offline prototype and bounded evaluations, not evidence that its answers are safe or effective in real emergencies.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported scores come from a 41-pair question-answer bank and a blind evaluation of 40 answers; the evidence does not establish performance in real emergency situations. The reported timing is from ten questions on a Pixel 7 Pro with six CPU threads and no network connection.","speaker":{"name":"Salvi, Rezwan Islam, Muna, Saiyma Sittul, Mushfique, Mushfiqur Rahman","roleAtSource":"Article author"},"source":{"title":"HandsOnAid: Building and Evaluating an Offline First-Aid Assistant for Emergency Response Scenarios","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.14059","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors built HandsOnAid as a phone-based first-aid assistant intended to work offline, using a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA.","excerpt":"We built HandsOnAid, a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA dataset","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"The authors evaluated seven configurations with a 41-pair question-answer bank and four LLM judges; the best configuration, fine-tuning with retrieval, scored 2.51 out of 5.","excerpt":"The best configuration scored 2.51 out of 5 which was the fine-tuned 4-bit model with retrieval.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"A medical professional’s blind evaluation of 40 answers produced scores similar to the average scores from the LLM judges.","excerpt":"A medical professional also conducted a blind evaluation of 40 answers across the base, the 4-bit fine-tuned, and the retrieval configurations, and the scores were similar to the LLM judges' average scores.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"The authors report limited coverage of several guideline details in the training corpus and suggest missing information could contribute to low scores.","excerpt":"A keyword audit found limited coverage of several details of ANZCOR guidelines in the training corpus, that are required by the rubric, which suggests that missing information in the training data could be a cause of the low scores.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"},{"text":"In a ten-question test on a Pixel 7 Pro without network access, the fine-tuned 4-bit model took an average of 6.86 seconds to show text and 14.88 seconds to complete an answer.","excerpt":"the 4-bit fine-tuned model produced its first displayable text in 6.86 seconds and a complete answer in 14.88 seconds on average over ten questions.","url":"https://arxiv.org/abs/2611.14059","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.14059","checkedAt":"2026-11-12T04:18:10.436Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would the system perform in emergency situations beyond the paper’s question-and-answer evaluations?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI authors report an offline first-aid assistant with modest scores","summary":"Authors Rezwan Islam Salvi, Saiyma Sittul Muna and Mushfiqur Rahman Mushfique describe HandsOnAid, a phone-based assistant designed to answer first-aid questions without a network connection. They tested seven configurations using 41 question-answer pairs with medically reviewed references following ANZCOR guidelines, and four frontier LLM judges scored the answers. Fine-tuning raised the mean score over the base model, but the best configuration—fine-tuning with retrieval—scored 2.51 out of 5. A medical professional’s blind evaluation of 40 answers produced scores similar to the judges’ averages. The authors also report that a keyword audit found limited coverage of several guideline details required by their rubric, suggesting gaps in training data could help explain low scores. On a Pixel 7 Pro without network access, the fine-tuned 4-bit model produced its first displayable text in an average of 6.86 seconds and a complete answer in 14.88 seconds over ten questions. The paper presents an offline prototype and bounded evaluations, not evidence that its answers are safe or effective in real emergencies.","question":"How would the system perform in emergency situations beyond the paper’s question-and-answer evaluations?","claims":[{"text":"The authors built HandsOnAid as a phone-based first-aid assistant intended to work offline, using a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA.","excerpt":"We built HandsOnAid, a 4-bit Gemma 2B model fine-tuned with QLoRA on FirstAidQA dataset","support":"source-stated"},{"text":"The authors evaluated seven configurations with a 41-pair question-answer bank and four LLM judges; the best configuration, fine-tuning with retrieval, scored 2.51 out of 5.","excerpt":"The best configuration scored 2.51 out of 5 which was the fine-tuned 4-bit model with retrieval.","support":"source-stated"},{"text":"A medical professional’s blind evaluation of 40 answers produced scores similar to the average scores from the LLM judges.","excerpt":"A medical professional also conducted a blind evaluation of 40 answers across the base, the 4-bit fine-tuned, and the retrieval configurations, and the scores were similar to the LLM judges' average scores.","support":"source-stated"},{"text":"The authors report limited coverage of several guideline details in the training corpus and suggest missing information could contribute to low scores.","excerpt":"A keyword audit found limited coverage of several details of ANZCOR guidelines in the training corpus, that are required by the rubric, which suggests that missing information in the training data could be a cause of the low scores.","support":"source-stated"},{"text":"In a ten-question test on a Pixel 7 Pro without network access, the fine-tuned 4-bit model took an average of 6.86 seconds to show text and 14.88 seconds to complete an answer.","excerpt":"the 4-bit fine-tuned model produced its first displayable text in 6.86 seconds and a complete answer in 14.88 seconds on average over ten questions.","support":"source-stated"}],"limitations":["The reported scores come from a 41-pair question-answer bank and a blind evaluation of 40 answers; the evidence does not establish performance in real emergency situations.","The reported timing is from ten questions on a Pixel 7 Pro with six CPU threads and no network connection."],"mediaId":"5da77e34efba0af21c9d68ad744cfdcdd13250145e2f2a38dacf8a2a734dab64","correctionReason":null,"id":"e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6","publicationId":"f18d86a0-ab05-48ef-8d1f-796a30774ad9","version":1,"sourceUrl":"https://arxiv.org/abs/2611.14059","sourceTitle":"HandsOnAid: Building and Evaluating an Offline First-Aid Assistant for Emergency Response Scenarios","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:06:40.100Z","createdAt":"2026-11-12T03:47:28.535Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"5da77e34efba0af21c9d68ad744cfdcdd13250145e2f2a38dacf8a2a734dab64","kind":"article","url":"https://arxiv.org/abs/2611.14059","sourceUrl":"https://arxiv.org/abs/2611.14059","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T03:46:27.324Z","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/2611.14059","checkedAt":"2026-11-12T04:18:10.436Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"5da77e34efba0af21c9d68ad744cfdcdd13250145e2f2a38dacf8a2a734dab64","confidence":0.59,"provider":"typesafe","model":"jev-1.13.0","inputHash":"f7d78445afe219444215c3d1495d5013d536041d3cfc9e0a9e2b3804c2537a93","decidedAt":"2026-11-12T03:46:29.517Z"},"provenance":{"textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observationId":"7e0703104a59bd48940c21e7e4e3d3ba2a5ba491ba0165edcf959d4304390fce","retrievedAt":"2026-11-12T03:46:26.979Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"a7408383-9306-4dac-8aad-d71e55a86a39"},"reviewer":{"model":"gpt-6-luna","runId":"f053a2dc-ef97-47b9-8c52-f3a9e06f9c48"}},"audit":{"status":"passed","reviewedAt":"2026-11-12T03:47:28.535Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Salvi, Rezwan Islam","url":null,"kind":"unknown"},{"name":"Muna, Saiyma Sittul","url":null,"kind":"unknown"},{"name":"Mushfique, Mushfiqur Rahman","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Salvi, Rezwan Islam","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Salvi, Rezwan Islam","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14059","locator":"head/meta[name=author|citation_author]"},"assertionId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T04:18:10.436Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","sourceUrl":"https://arxiv.org/abs/2611.14059","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z"}},{"actor":{"name":"Muna, Saiyma Sittul","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Muna, Saiyma Sittul","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14059","locator":"head/meta[name=author|citation_author]"},"assertionId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T04:18:10.436Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","sourceUrl":"https://arxiv.org/abs/2611.14059","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z"}},{"actor":{"name":"Mushfique, Mushfiqur Rahman","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Mushfique, Mushfiqur Rahman","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.14059","locator":"head/meta[name=author|citation_author]"},"assertionId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T04:18:10.436Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","sourceUrl":"https://arxiv.org/abs/2611.14059","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.14059","snapshotId":"0ce0af26f4d736e5cb9b742e4ab5144c83d0dd16173d0125cfe09eac1eed4100","contentSha256":"e35356012698cb4d91d6dbbec505e83e0f2d2c0152950b7ae13f4bec5a4f2f11","textHash":"a9e1c3d6faab879f05ff818d54d4e83bb0d1c073df9431d747a7e0214620e09f","observedAt":"2026-11-12T04:18:10.436Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/e07029bdc851b6cd97a21b8648bfcb55a6fbbe3cfab2a05e731634d377a78dc6/source"}}}],"audits":[{"publicationId":"f18d86a0-ab05-48ef-8d1f-796a30774ad9","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the work to the three listed authors. It keeps the results as the authors' reported findings, including the hedged keyword-audit interpretation. The arXiv cs.AI label matches the feed and the paper's cross-listing.","excerpt":"Authors:Rezwan Islam Salvi, Saiyma Sittul Muna, Mushfiqur Rahman Mushfique"},{"dimension":"summary","status":"pass","explanation":"The title and summary faithfully state what was built and how it was tested. They also report the modest best score (2.51/5), the medical-professional comparison, the training-data gap and the on-phone timing. A general reader can follow the central idea: an offline phone first-aid assistant that scores modestly on a strict evaluation. Terms such as ANZCOR, fine-tuning and retrieval are kept exactly and can be explained with tooltips. Omitting the base-to-fine-tuned figures (+1.08, to 2.34) and the safety overrides does not distort the meaning. The question follows from the evidence.","excerpt":"Fine-tuning improved the mean score by +1.08 over the base model to 2.34."},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-11-10 matches the arXiv v1 submission date. The later feed timestamp is not used.","excerpt":"[Submitted on 10 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Every claim matches its verbatim excerpt. The closing limitation, that the paper is not evidence of safety or effectiveness in real emergencies, is appropriately cautious and follows from the bounded question-answer and timing evaluations. It invents no benefits, costs or stakeholders. The authors' tentative wording, that missing training information 'could be a cause' of the low scores, is preserved.","excerpt":"which suggests that missing information in the training data could be a cause of the low scores"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page. It is the original source and the most relevant item available.","excerpt":null}],"createdAt":"2026-11-12T03:47:28.535Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}