{"id":"8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328","title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","argument":"Joseph Q. Zales, Pragya Sharma and Mani Srivastava argue that choosing an energy-efficient model for periodic sensing requires measuring the deployment, not inference alone. Their system, CREST, evaluates model and hardware configurations under a specified release schedule, accounting for the time between inferences. In experiments on STM32 N657, waiting between releases offset 97% of the energy saving associated with choosing 800 MHz rather than 300 MHz for one architecture; with STOP between releases, 300 MHz used slightly less board energy per release at 100 and 500 ms. The preferred architecture also changed with the release period. The authors report that latency predicted continuous energy closely, but most release-energy error remained in scheduled terms. Other case studies found that FLOPs-selected candidates sometimes did not fit or used more energy than measured selections, and that cross-target replay changed an audio-classification configuration’s energy rank. The results illustrate why a model’s inference cost alone may not predict its energy use in a scheduled deployment. The abstract reports three case studies; how broadly their findings apply beyond the tested configurations remains open.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports three case studies; the findings described here concern the tested targets, configurations and schedules. The abstract does not establish how broadly the results apply beyond those case studies.","speaker":{"name":"Zales, Joseph Q., Sharma, Pragya, Srivastava, Mani","roleAtSource":"Article author"},"source":{"title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","publisher":"arXiv eess.SY","format":"Article","url":"https://arxiv.org/abs/2606.15004","publishedAt":"2026-06-12","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that model selection for periodic sensing should account for the time between inferences, not just inference energy.","excerpt":"Selecting an energy-efficient model for periodic sensing requires accounting for time between inferences.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"CREST evaluates model and hardware configurations under a specified release schedule.","excerpt":"an open-source hardware-in-the-loop experimentation system that evaluates model and hardware configurations under a specified release schedule.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"In one STM32 N657 case, waiting offset 97% of the continuous-inference energy saving in a fitted decomposition; with STOP between releases, 300 MHz used slightly less board energy per release at the reported periods.","excerpt":"additional waiting offsets 97% of this saving in a fitted decomposition. With STOP between releases, 300 MHz uses about 0.3-0.4% less board energy per release at 100 and 500 ms.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"The authors report that latency predicted continuous energy closely, while most release-energy error remained in scheduled terms.","excerpt":"Latency predicts continuous energy within a median 1.7% on held-out observations, yet most release-energy error remains in scheduled terms.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"In the Arduino case study, some FLOPs-selected candidates did not fit the most memory-constrained board or used more energy than measured selections.","excerpt":"proxy-selected inertial-odometry candidates can fail to fit or use more energy than measured selections, with only 9 of 16 FLOPs-selected candidates fitting the most memory-constrained board.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"Cross-target replay changed the selected audio-classification configuration from third to last in energy rank.","excerpt":"cross-target replay moves a selected configuration from third to last in energy rank.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2606.15004","checkedAt":"2026-11-13T01:50:10.206Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well do the reported findings hold for model and hardware configurations beyond the three case studies?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","summary":"Joseph Q. Zales, Pragya Sharma and Mani Srivastava argue that choosing an energy-efficient model for periodic sensing requires measuring the deployment, not inference alone. Their system, CREST, evaluates model and hardware configurations under a specified release schedule, accounting for the time between inferences. In experiments on STM32 N657, waiting between releases offset 97% of the energy saving associated with choosing 800 MHz rather than 300 MHz for one architecture; with STOP between releases, 300 MHz used slightly less board energy per release at 100 and 500 ms. The preferred architecture also changed with the release period. The authors report that latency predicted continuous energy closely, but most release-energy error remained in scheduled terms. Other case studies found that FLOPs-selected candidates sometimes did not fit or used more energy than measured selections, and that cross-target replay changed an audio-classification configuration’s energy rank. The results illustrate why a model’s inference cost alone may not predict its energy use in a scheduled deployment. The abstract reports three case studies; how broadly their findings apply beyond the tested configurations remains open.","question":"How well do the reported findings hold for model and hardware configurations beyond the three case studies?","claims":[{"text":"The authors argue that model selection for periodic sensing should account for the time between inferences, not just inference energy.","excerpt":"Selecting an energy-efficient model for periodic sensing requires accounting for time between inferences.","support":"source-stated"},{"text":"CREST evaluates model and hardware configurations under a specified release schedule.","excerpt":"an open-source hardware-in-the-loop experimentation system that evaluates model and hardware configurations under a specified release schedule.","support":"source-stated"},{"text":"In one STM32 N657 case, waiting offset 97% of the continuous-inference energy saving in a fitted decomposition; with STOP between releases, 300 MHz used slightly less board energy per release at the reported periods.","excerpt":"additional waiting offsets 97% of this saving in a fitted decomposition. With STOP between releases, 300 MHz uses about 0.3-0.4% less board energy per release at 100 and 500 ms.","support":"source-stated"},{"text":"The authors report that latency predicted continuous energy closely, while most release-energy error remained in scheduled terms.","excerpt":"Latency predicts continuous energy within a median 1.7% on held-out observations, yet most release-energy error remains in scheduled terms.","support":"source-stated"},{"text":"In the Arduino case study, some FLOPs-selected candidates did not fit the most memory-constrained board or used more energy than measured selections.","excerpt":"proxy-selected inertial-odometry candidates can fail to fit or use more energy than measured selections, with only 9 of 16 FLOPs-selected candidates fitting the most memory-constrained board.","support":"source-stated"},{"text":"Cross-target replay changed the selected audio-classification configuration from third to last in energy rank.","excerpt":"cross-target replay moves a selected configuration from third to last in energy rank.","support":"source-stated"}],"limitations":["The abstract reports three case studies; the findings described here concern the tested targets, configurations and schedules.","The abstract does not establish how broadly the results apply beyond those case studies."],"mediaId":"bf36d4ce38d751eee1e8bf5004baf43a6235882a03467a182ea427926559bdfb","correctionReason":null,"id":"8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328","publicationId":"1eb9fa6c-8f00-4fd9-841a-39d1c047d513","version":1,"sourceUrl":"https://arxiv.org/abs/2606.15004","sourceTitle":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","publisher":"arXiv eess.SY","publishedAt":null,"originalPublished":{"value":"2026-06-12","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:34:05.517Z","createdAt":"2026-11-12T08:46:34.241Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"bf36d4ce38d751eee1e8bf5004baf43a6235882a03467a182ea427926559bdfb","kind":"article","url":"https://arxiv.org/abs/2606.15004","sourceUrl":"https://arxiv.org/abs/2606.15004","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T08:45:44.560Z","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/2606.15004","checkedAt":"2026-11-13T01:50:10.206Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"bf36d4ce38d751eee1e8bf5004baf43a6235882a03467a182ea427926559bdfb","confidence":0.65,"provider":"typesafe","model":"jev-1.13.0","inputHash":"faabe4ad2e970f9d54c3dfbcfabb0c8de56a3ce5c6619a62b7a7a7b54c1716c1","decidedAt":"2026-11-12T08:45:45.319Z"},"provenance":{"textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observationId":"238674fae4525d8ed65635bce77b3771e5f30794a19f7ab752b702ecad27a4b0","retrievedAt":"2026-11-12T08:45:44.483Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"5d036785-ea76-4fc9-805d-8c67230d8bce"},"reviewer":{"model":"gpt-6-luna","runId":"58056912-7c88-4cc4-8ad5-0d900a58e546"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T08:46:34.241Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Zales, Joseph Q.","url":null,"kind":"unknown"},{"name":"Sharma, Pragya","url":null,"kind":"unknown"},{"name":"Srivastava, Mani","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Zales, Joseph Q.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zales, Joseph Q.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2606.15004","locator":"head/meta[name=author|citation_author]"},"assertionId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:50:10.206Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","sourceUrl":"https://arxiv.org/abs/2606.15004","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z"}},{"actor":{"name":"Sharma, Pragya","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sharma, Pragya","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2606.15004","locator":"head/meta[name=author|citation_author]"},"assertionId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:50:10.206Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","sourceUrl":"https://arxiv.org/abs/2606.15004","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z"}},{"actor":{"name":"Srivastava, Mani","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Srivastava, Mani","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2606.15004","locator":"head/meta[name=author|citation_author]"},"assertionId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:50:10.206Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","sourceUrl":"https://arxiv.org/abs/2606.15004","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2606.15004","snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328/source"}},"canonicalUrl":"https://escapemeridian.com/readings/measure-the-deployment-not-just-the-model-rethinking-hardware-8ce8d625","formats":{"html":"/readings/measure-the-deployment-not-just-the-model-rethinking-hardware-8ce8d625","markdown":"/readings/measure-the-deployment-not-just-the-model-rethinking-hardware-8ce8d625.md","json":"/readings/measure-the-deployment-not-just-the-model-rethinking-hardware-8ce8d625.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-latency","name":"Latency","explanation":"The delay between sending a request and receiving the result.","referenceUrl":"https://developers.google.com/machine-learning/glossary#latency"},{"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"},{"id":"time","name":"TIME","kind":"organization","context":"TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.","referenceUrls":["https://time.com/"],"path":"/entities/time","json":"/entities/time.json","markdown":"/entities/time.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328","revision":"1eb9fa6c-8f00-4fd9-841a-39d1c047d513","revisionKind":"publication-id","title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","url":"https://escapemeridian.com/readings/measure-the-deployment-not-just-the-model-rethinking-hardware-8ce8d625"},"source":{"type":"source","id":"https://arxiv.org/abs/2606.15004","revision":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","revisionKind":"sha256-source-text","title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","url":"https://arxiv.org/abs/2606.15004"},"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":"8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328","title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","argument":"Joseph Q. Zales, Pragya Sharma and Mani Srivastava argue that choosing an energy-efficient model for periodic sensing requires measuring the deployment, not inference alone. Their system, CREST, evaluates model and hardware configurations under a specified release schedule, accounting for the time between inferences. In experiments on STM32 N657, waiting between releases offset 97% of the energy saving associated with choosing 800 MHz rather than 300 MHz for one architecture; with STOP between releases, 300 MHz used slightly less board energy per release at 100 and 500 ms. The preferred architecture also changed with the release period. The authors report that latency predicted continuous energy closely, but most release-energy error remained in scheduled terms. Other case studies found that FLOPs-selected candidates sometimes did not fit or used more energy than measured selections, and that cross-target replay changed an audio-classification configuration’s energy rank. The results illustrate why a model’s inference cost alone may not predict its energy use in a scheduled deployment. The abstract reports three case studies; how broadly their findings apply beyond the tested configurations remains open.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports three case studies; the findings described here concern the tested targets, configurations and schedules. The abstract does not establish how broadly the results apply beyond those case studies.","speaker":{"name":"Zales, Joseph Q., Sharma, Pragya, Srivastava, Mani","roleAtSource":"Article author"},"source":{"title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","publisher":"arXiv eess.SY","format":"Article","url":"https://arxiv.org/abs/2606.15004","publishedAt":"2026-06-12","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that model selection for periodic sensing should account for the time between inferences, not just inference energy.","excerpt":"Selecting an energy-efficient model for periodic sensing requires accounting for time between inferences.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"CREST evaluates model and hardware configurations under a specified release schedule.","excerpt":"an open-source hardware-in-the-loop experimentation system that evaluates model and hardware configurations under a specified release schedule.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"In one STM32 N657 case, waiting offset 97% of the continuous-inference energy saving in a fitted decomposition; with STOP between releases, 300 MHz used slightly less board energy per release at the reported periods.","excerpt":"additional waiting offsets 97% of this saving in a fitted decomposition. With STOP between releases, 300 MHz uses about 0.3-0.4% less board energy per release at 100 and 500 ms.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"The authors report that latency predicted continuous energy closely, while most release-energy error remained in scheduled terms.","excerpt":"Latency predicts continuous energy within a median 1.7% on held-out observations, yet most release-energy error remains in scheduled terms.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"In the Arduino case study, some FLOPs-selected candidates did not fit the most memory-constrained board or used more energy than measured selections.","excerpt":"proxy-selected inertial-odometry candidates can fail to fit or use more energy than measured selections, with only 9 of 16 FLOPs-selected candidates fitting the most memory-constrained board.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"},{"text":"Cross-target replay changed the selected audio-classification configuration from third to last in energy rank.","excerpt":"cross-target replay moves a selected configuration from third to last in energy rank.","url":"https://arxiv.org/abs/2606.15004","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2606.15004","checkedAt":"2026-11-13T01:50:10.206Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well do the reported findings hold for model and hardware configurations beyond the three case studies?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","summary":"Joseph Q. Zales, Pragya Sharma and Mani Srivastava argue that choosing an energy-efficient model for periodic sensing requires measuring the deployment, not inference alone. Their system, CREST, evaluates model and hardware configurations under a specified release schedule, accounting for the time between inferences. In experiments on STM32 N657, waiting between releases offset 97% of the energy saving associated with choosing 800 MHz rather than 300 MHz for one architecture; with STOP between releases, 300 MHz used slightly less board energy per release at 100 and 500 ms. The preferred architecture also changed with the release period. The authors report that latency predicted continuous energy closely, but most release-energy error remained in scheduled terms. Other case studies found that FLOPs-selected candidates sometimes did not fit or used more energy than measured selections, and that cross-target replay changed an audio-classification configuration’s energy rank. The results illustrate why a model’s inference cost alone may not predict its energy use in a scheduled deployment. The abstract reports three case studies; how broadly their findings apply beyond the tested configurations remains open.","question":"How well do the reported findings hold for model and hardware configurations beyond the three case studies?","claims":[{"text":"The authors argue that model selection for periodic sensing should account for the time between inferences, not just inference energy.","excerpt":"Selecting an energy-efficient model for periodic sensing requires accounting for time between inferences.","support":"source-stated"},{"text":"CREST evaluates model and hardware configurations under a specified release schedule.","excerpt":"an open-source hardware-in-the-loop experimentation system that evaluates model and hardware configurations under a specified release schedule.","support":"source-stated"},{"text":"In one STM32 N657 case, waiting offset 97% of the continuous-inference energy saving in a fitted decomposition; with STOP between releases, 300 MHz used slightly less board energy per release at the reported periods.","excerpt":"additional waiting offsets 97% of this saving in a fitted decomposition. With STOP between releases, 300 MHz uses about 0.3-0.4% less board energy per release at 100 and 500 ms.","support":"source-stated"},{"text":"The authors report that latency predicted continuous energy closely, while most release-energy error remained in scheduled terms.","excerpt":"Latency predicts continuous energy within a median 1.7% on held-out observations, yet most release-energy error remains in scheduled terms.","support":"source-stated"},{"text":"In the Arduino case study, some FLOPs-selected candidates did not fit the most memory-constrained board or used more energy than measured selections.","excerpt":"proxy-selected inertial-odometry candidates can fail to fit or use more energy than measured selections, with only 9 of 16 FLOPs-selected candidates fitting the most memory-constrained board.","support":"source-stated"},{"text":"Cross-target replay changed the selected audio-classification configuration from third to last in energy rank.","excerpt":"cross-target replay moves a selected configuration from third to last in energy rank.","support":"source-stated"}],"limitations":["The abstract reports three case studies; the findings described here concern the tested targets, configurations and schedules.","The abstract does not establish how broadly the results apply beyond those case studies."],"mediaId":"bf36d4ce38d751eee1e8bf5004baf43a6235882a03467a182ea427926559bdfb","correctionReason":null,"id":"8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328","publicationId":"1eb9fa6c-8f00-4fd9-841a-39d1c047d513","version":1,"sourceUrl":"https://arxiv.org/abs/2606.15004","sourceTitle":"Measure the Deployment, Not Just the Model: Rethinking Hardware-Aware Model Selection for Microcontrollers","publisher":"arXiv eess.SY","publishedAt":null,"originalPublished":{"value":"2026-06-12","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:34:05.517Z","createdAt":"2026-11-12T08:46:34.241Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"bf36d4ce38d751eee1e8bf5004baf43a6235882a03467a182ea427926559bdfb","kind":"article","url":"https://arxiv.org/abs/2606.15004","sourceUrl":"https://arxiv.org/abs/2606.15004","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T08:45:44.560Z","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/2606.15004","checkedAt":"2026-11-13T01:50:10.206Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"bf36d4ce38d751eee1e8bf5004baf43a6235882a03467a182ea427926559bdfb","confidence":0.65,"provider":"typesafe","model":"jev-1.13.0","inputHash":"faabe4ad2e970f9d54c3dfbcfabb0c8de56a3ce5c6619a62b7a7a7b54c1716c1","decidedAt":"2026-11-12T08:45:45.319Z"},"provenance":{"textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observationId":"238674fae4525d8ed65635bce77b3771e5f30794a19f7ab752b702ecad27a4b0","retrievedAt":"2026-11-12T08:45:44.483Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"5d036785-ea76-4fc9-805d-8c67230d8bce"},"reviewer":{"model":"gpt-6-luna","runId":"58056912-7c88-4cc4-8ad5-0d900a58e546"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T08:46:34.241Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Zales, Joseph Q.","url":null,"kind":"unknown"},{"name":"Sharma, Pragya","url":null,"kind":"unknown"},{"name":"Srivastava, Mani","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Zales, Joseph Q.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zales, Joseph Q.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2606.15004","locator":"head/meta[name=author|citation_author]"},"assertionId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:50:10.206Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","sourceUrl":"https://arxiv.org/abs/2606.15004","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z"}},{"actor":{"name":"Sharma, Pragya","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sharma, Pragya","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2606.15004","locator":"head/meta[name=author|citation_author]"},"assertionId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:50:10.206Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","sourceUrl":"https://arxiv.org/abs/2606.15004","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z"}},{"actor":{"name":"Srivastava, Mani","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Srivastava, Mani","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2606.15004","locator":"head/meta[name=author|citation_author]"},"assertionId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-13T01:50:10.206Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","sourceUrl":"https://arxiv.org/abs/2606.15004","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2606.15004","snapshotId":"e206ca41b6c44dba7f4077ea740f437c407fb70eb77e0c95d22ec26a0a5aa70a","contentSha256":"55adeee39a471d93fe62c91d3b1509b8fb74a8bbfcca15ad82d5165194c5be15","textHash":"b820e274a5cbbada1f6553a1594f12290f76b4f06a81aa86252906e710379371","observedAt":"2026-11-13T01:50:10.206Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/8ce8d625632e7a3af88c121fe6e946366e4c4a6cf4c3d602f9321de6ac0a5328/source"}}}],"audits":[{"publicationId":"1eb9fa6c-8f00-4fd9-841a-39d1c047d513","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the argument and CREST to Joseph Q. Zales, Pragya Sharma and Mani Srivastava, matching the arXiv author list. It frames the results as the authors' reports.","excerpt":"Authors:Joseph Q. Zales, Pragya Sharma, Mani Srivastava"},{"dimension":"summary","status":"uncertain","explanation":"The draft leads with the authors' thesis, but central terms are left unexplained for non-specialists: microcontroller (only in the title), release and release schedule, STOP, FLOPs, cross-target replay, scheduled terms and hardware-in-the-loop. Without them, a general reader cannot see why waiting changes energy, for example a chip sleeping between scheduled sensor readings. Smallest repair: keep the terms and add tooltips, or add a short gloss for release, STOP and FLOPs. Add those missing from the term library. The title and the open question are faithful.","excerpt":"Selecting an energy-efficient model for periodic sensing requires accounting for time between inferences."},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-06-12 matches the v1 submission date. It is correctly distinct from the v2 revision (9 Nov 2026) and the feed date.","excerpt":"Submitted on 12 Jun 2026 (v1), last revised 9 Nov 2026 (this version, v2)"},{"dimension":"support","status":"uncertain","explanation":"Two conditions are blurred. (1) The summary says waiting offset 97% of the saving \"in experiments\" and drops the source's qualifier that this comes from a fitted decomposition, a model-based estimate rather than a direct measurement. The claim keeps the qualifier; the summary should too. (2) The source attributes failing to fit or using more energy to proxy-selected candidates generally, and attributes only the fit count (9 of 16) to FLOPs-selected ones. The summary and claim assign both failures to FLOPs-selected candidates. Repair: say \"proxy-selected\" and cite FLOPs only for the 9-of-16 fit result.","excerpt":"proxy-selected inertial-odometry candidates can fail to fit or use more energy than measured selections, with only 9 of 16 FLOPs-selected candidates fitting the most memory-constrained board."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-12T08:46:34.241Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}