JARVIS-Bench / croissant.json
WZDavid's picture
Simplify dataset title and image description note
4be7bff verified
Raw History Blame Contribute Delete
34.4 kB
{
"@context": {
"@language": "en",
"@vocab": "https://schema.org/",
"citeAs": "cr:citeAs",
"column": "cr:column",
"conformsTo": "dct:conformsTo",
"cr": "http://mlcommons.org/croissant/",
"rai": "http://mlcommons.org/croissant/RAI/",
"data": {
"@id": "cr:data",
"@type": "@json"
},
"dataType": {
"@id": "cr:dataType",
"@type": "@vocab"
},
"dct": "http://purl.org/dc/terms/",
"extract": "cr:extract",
"field": "cr:field",
"fileProperty": "cr:fileProperty",
"fileObject": "cr:fileObject",
"fileSet": "cr:fileSet",
"format": "cr:format",
"includes": "cr:includes",
"isLiveDataset": "cr:isLiveDataset",
"jsonPath": "cr:jsonPath",
"key": "cr:key",
"md5": "cr:md5",
"parentField": "cr:parentField",
"path": "cr:path",
"recordSet": "cr:recordSet",
"references": "cr:references",
"regex": "cr:regex",
"repeated": "cr:repeated",
"replace": "cr:replace",
"sc": "https://schema.org/",
"separator": "cr:separator",
"source": "cr:source",
"subField": "cr:subField",
"transform": "cr:transform",
"wd": "https://www.wikidata.org/wiki/",
"prov": "http://www.w3.org/ns/prov#",
"sha256": "sc:sha256",
"equivalentProperty": "cr:equivalentProperty",
"examples": {
"@id": "cr:examples",
"@type": "@json"
},
"samplingRate": "cr:samplingRate"
},
"@type": "sc:Dataset",
"name": "JARVIS-Bench",
"alternateName": "JARVIS-Bench: Benchmarking Personal Intelligence Agents on Long-Horizon Real-User Daily Traces",
"conformsTo": "http://mlcommons.org/croissant/1.0",
"description": "JARVIS-Bench evaluates personal intelligence agents using longitudinal daily-life records from 42 participants. This release contains 18,529 events across 1,495 user-days, with 28 to 53 days per assembled trajectory. It combines real-log-derived anchors and synthetic trajectory expansions with structured user profiles. The benchmark supports Implicit Preference Reasoning, Proactive Help Prediction, Emotion-Aware Modeling and Cross-Domain Transfer. Descriptions corresponding to the images are included in the textual records; image files are not distributed in this repository. The paper reports 18,471 events in its evaluation set; the full corpus is supplied here.",
"url": "https://huggingface.co/datasets/WZDavid/JARVIS-Bench",
"version": "1.0.0",
"license": "https://creativecommons.org/licenses/by-nc/4.0/",
"isAccessibleForFree": true,
"creator": [
{
"@type": "sc:Person",
"name": "Weizhi Zhang"
},
{
"@type": "sc:Person",
"name": "Wei-Chieh Huang"
},
{
"@type": "sc:Person",
"name": "Yueqing Liang"
},
{
"@type": "sc:Person",
"name": "Liwei Jiang"
},
{
"@type": "sc:Person",
"name": "Zhengxiang Wang"
},
{
"@type": "sc:Person",
"name": "Liangwei Yang"
},
{
"@type": "sc:Person",
"name": "Zechen Li"
},
{
"@type": "sc:Person",
"name": "Yuchen Wu"
},
{
"@type": "sc:Person",
"name": "Haozhen Zhang"
},
{
"@type": "sc:Person",
"name": "Yu Wang"
},
{
"@type": "sc:Person",
"name": "Yuanchen Bei"
},
{
"@type": "sc:Person",
"name": "Yue Zhou"
},
{
"@type": "sc:Person",
"name": "Siqi Zhu"
},
{
"@type": "sc:Person",
"name": "Henry Peng Zou"
},
{
"@type": "sc:Person",
"name": "Jiahong Liu"
},
{
"@type": "sc:Person",
"name": "Xinni Zhang"
},
{
"@type": "sc:Person",
"name": "Paul Martin"
},
{
"@type": "sc:Person",
"name": "Joseph Marvin Imperial"
},
{
"@type": "sc:Person",
"name": "Yuyang Luo"
},
{
"@type": "sc:Person",
"name": "Xiongxiao Xu"
},
{
"@type": "sc:Person",
"name": "Baixiang Huang"
},
{
"@type": "sc:Person",
"name": "Shanglin Wu"
},
{
"@type": "sc:Person",
"name": "Lucas Resck"
},
{
"@type": "sc:Person",
"name": "Yibo Wang"
},
{
"@type": "sc:Person",
"name": "Yuqing Liu"
},
{
"@type": "sc:Person",
"name": "Langzhou He"
},
{
"@type": "sc:Person",
"name": "Chengze Li"
},
{
"@type": "sc:Person",
"name": "Yuxin Tian"
},
{
"@type": "sc:Person",
"name": "Kening Zheng"
},
{
"@type": "sc:Person",
"name": "Enze Ma"
},
{
"@type": "sc:Person",
"name": "Boi Huynh"
},
{
"@type": "sc:Person",
"name": "Zheng Hui"
},
{
"@type": "sc:Person",
"name": "Rui Yang"
},
{
"@type": "sc:Person",
"name": "Tao Feng"
},
{
"@type": "sc:Person",
"name": "Jie Yang"
},
{
"@type": "sc:Person",
"name": "Shanghao Li"
},
{
"@type": "sc:Person",
"name": "Haoran Wang"
},
{
"@type": "sc:Person",
"name": "Wooseong Yang"
},
{
"@type": "sc:Person",
"name": "Hanrong Zhang"
},
{
"@type": "sc:Person",
"name": "Huanhuan Ma"
},
{
"@type": "sc:Person",
"name": "Daye Yoon"
},
{
"@type": "sc:Person",
"name": "Yaozu Wu"
},
{
"@type": "sc:Person",
"name": "Shikan Lian"
},
{
"@type": "sc:Person",
"name": "Xiaoqian Ruan"
},
{
"@type": "sc:Person",
"name": "Fangxin Wang"
},
{
"@type": "sc:Person",
"name": "Hyeonjeong Park"
},
{
"@type": "sc:Person",
"name": "Hins Hu"
},
{
"@type": "sc:Person",
"name": "Wenzhe Fan"
},
{
"@type": "sc:Person",
"name": "Chen Wang"
},
{
"@type": "sc:Person",
"name": "Yifan Gu"
},
{
"@type": "sc:Person",
"name": "Dongyuan Li"
},
{
"@type": "sc:Person",
"name": "Song Wang"
},
{
"@type": "sc:Person",
"name": "Aylin Caliskan"
},
{
"@type": "sc:Person",
"name": "Jindong Wang"
},
{
"@type": "sc:Person",
"name": "Xiao Luo"
},
{
"@type": "sc:Person",
"name": "Yankai Chen"
},
{
"@type": "sc:Person",
"name": "Kai Shu"
},
{
"@type": "sc:Person",
"name": "Xue Liu"
},
{
"@type": "sc:Person",
"name": "Philip S. Yu"
}
],
"citeAs": "Weizhi Zhang et al. (2026). JARVIS-Bench: Benchmarking Personal Intelligence Agents on Long-Horizon Real-User Daily Traces. Version 1.0.0. https://huggingface.co/datasets/WZDavid/JARVIS-Bench",
"keywords": [
"benchmark",
"personalization",
"large language models",
"LLM evaluation",
"personal intelligence agents",
"long-horizon memory",
"implicit preference",
"proactive assistance",
"AI agents",
"multilingual"
],
"rai:dataCollection": "Seventy participants enrolled in a self-tracking study approved by the Institutional Review Board of the University of Illinois Chicago. All participants provided informed consent; participation was voluntary and no monetary compensation was provided. Through a custom mobile interface, participants supplied daily logs with optional photographs, structured persona profiles, and responses to a 100-question implicit-preference survey. Daily logs include participant-reported proactive-help and mood annotations. Forty-two participants passed the three-stage quality checks and form the released evaluation set, The assembled trajectories in this release span 28 to 53 relative days. Synthetic trajectory expansions are grounded in these participants and preserve the original anchor payloads. Participants were informed about the collected daily-life traces, privacy risks, anonymization procedures, and their right to withdraw under the approved protocol.",
"rai:dataCollectionType": [
"Participant-authored daily logs with optional photographs, timing fields, and free-text descriptions, collected through a custom mobile interface",
"Structured persona profiles with self-reported demographics and free-text self-description",
"100-question, four-option implicit-preference survey across eight domain labels",
"Participant-reported proactive-help flags and rationales, mood keywords, and mood notes on real log events",
"Clue-planted synthetic trajectory expansion grounded in the retained participants"
],
"rai:dataCollectionTimeframe": "The assembled trajectories in this release span 28 to 53 relative days. This is the length of the supplied trajectories, including synthetic expansions, rather than an independently measured participant collection duration. Calendar dates are represented by relative day identifiers; the supplied documentation does not specify a calendar-date collection interval.",
"rai:dataCollectionRawData": "Participants supplied daily logs with optional photographs, timing fields, and free-text descriptions; structured persona profiles containing age band, gender, region, role, and a free-text self-description; and answers to a 100-question, four-option implicit-preference survey. Daily logs also contain participant-reported proactive-help flags with free-text rationales and, on affect-bearing events, mood keywords and free-text mood notes. The survey covers eight lifestyle-domain labels. Logged events use 12 canonical event subdomains (wake, bedtime, nap, food, work, entertainment, weather, travel, shopping, physical, exercise, event), plus a long-tail other category. Release-time privacy processing is described in dataPreprocessingProtocol.",
"rai:dataCollectionMissingData": "Of 70 enrolled participants, 42 passed the benchmark's three-stage quality checks. Fourteen participants were excluded for insufficient logging volume, six for incomplete survey responses, and eight for failure to confirm trajectory fidelity, giving 40% overall attrition. For retained participants, synthetic trajectory expansion increases behavioural coverage around original anchor events while preserving anchor payloads verbatim. Release admission requires automatic structural validation, an internal full-trajectory review, and explicit participant fidelity confirmation with a randomly sampled subset of synthetic events foregrounded for review. Task-specific annotation coverage is incomplete: Proactive Help and Emotion-Aware Modeling each use 15 of the 42 retained users under the eligibility rules in Appendix F.6, with eight users in both pools.",
"rai:dataImputationProtocol": "Sparse behavioural coverage is addressed by clue-planted synthetic trajectory expansion grounded in each retained participant's original logs, persona, and distilled survey insight. Original anchor payloads are preserved verbatim, and each event carries provenance distinguishing anchor from synthetic content. This expansion supplies behavioural context rather than replacing participant-provided survey answers or source annotations. Participant exclusion and trajectory quality checks are described in dataCollectionMissingData.",
"rai:dataPreprocessingProtocol": [
"Participant selection: retain 42 of 70 enrolled participants after excluding 14 for insufficient logging, six for incomplete surveys, and eight for failure to confirm trajectory fidelity.",
"Privacy processing: strip calendar dates at ingestion and replace them with relative Day{N}T{HH:MM:SS} timestamps; remove surnames, precise locations, and calendar-day identifiers from released artefacts.",
"Image grounding: represent photographs in trajectory text using action-focused third-person captions and abstracted captions; the prompt forbids identifying people by name or appearance and transcribing private text visible in images.",
"Schema and provenance: represent events using a shared typed schema, preserving original anchor payloads and distinguishing real anchors from synthetic events at event level; map the 12 canonical event subdomains to seven lifestyle domains. The eighth survey label, cross_domain, has no corresponding event subdomain; unmapped long-tail other events are retained in the S1 and S2 ablations.",
"Synthetic time representation: use nine time buckets and a per-day sequence index instead of explicit clock times for synthetic events, while retaining original timing fields on anchors.",
"Quality assurance: enforce anchor equality, time-bucket consistency, event-count consistency, and a privacy blacklist using a rule-based validator; follow with internal full-trajectory review and participant fidelity confirmation, regenerating and revalidating trajectories when required.",
"HF text-only packaging: select 42 complete trajectories and corresponding profiles, serialize events/day metadata/profiles into Parquet, and omit image bytes, intermediate generation files and logs. The export preserves complete event JSON and can exactly reconstruct the supplied trajectories."
],
"rai:dataManipulationProtocol": "The four-stage synthesis pipeline first converts photographs into action-focused and abstracted captions using a vision model. It then combines the participant's persona, survey answers, a local behavioural-trait digest, and sample captions into an enhanced profile, uniqueness signature, persona narrative, and distilled survey_insight. The survey_insight field is the only conduit for raw survey content into the later stages; raw question text and option letters are not reintroduced. A day-level skeleton and clue plan are generated in 14-day windows, followed by trajectory expansion in five-day chunks. Every event carries a provenance flag distinguishing real anchors from synthetic events. Anchor payloads are reproduced verbatim; synthetic events express preferences through behavioural consequences conditioned on the participant and surrounding narrative. Rule-based checks enforce anchor equality, time-bucket consistency, event-count consistency, and the privacy blacklist.",
"rai:hasSyntheticData": true,
"rai:dataAnnotationProtocol": "Each participant provides answers to a 100-question implicit-preference instrument with four options per question. The survey is organised into daily_routine, food_dining, work_study, leisure, commuting_travel, shopping, social_wellbeing, and cross_domain. These participant answers are the ground truth for Implicit Preference Reasoning and Cross-Domain Transfer. Proactive Help Prediction instead uses participant-authored agentHelp flags and agentHelpReason text from real anchor events. Its evaluation pool contains 254 image-bearing anchors from qualifying logged days across 15 users, including 111 positive and 143 negative anchors. Emotion-Aware Modeling uses participant-reported moodKeywords on real anchors and polarity derived from those annotations. Its pool contains 123 photographed, mood-annotated anchors from the last five logged days of 15 users, with 78 positive, 22 negative, and 23 mixed samples. Eight users belong to both task pools. Synthetic events provide behavioural context rather than ground-truth labels for the proactive-help and emotion tasks.",
"rai:dataAnnotationPlatform": "Participants supplied their daily logs and task supervision through the custom mobile interface described in the paper. Quality assurance includes an internal reviewer reading each assembled trajectory, followed by participant fidelity confirmation with original anchors highlighted and a randomly sampled subset of synthetic events foregrounded for review.",
"rai:dataAnnotationAnalysis": "Trajectory quality assurance combines automatic structural validation, internal full-trajectory review, and participant fidelity confirmation. A separate validation of the GPT-5 rationale-match judge used five independent human annotators on 48 Proactive Help samples. Mean pairwise human Cohen's kappa was 0.855, human Fleiss' kappa was 0.853, and the judge agreed with the human majority vote at Cohen's kappa 0.909 and 95.8% accuracy. This evaluation-judge validation is distinct from the participant-authored survey and anchor labels.",
"rai:annotationsPerItem": "Each participant supplies their own ground-truth answer to each of the 100 survey questions and their own proactive-help and mood annotations on applicable real anchor events. Separately, each of the 48 samples used to validate the rationale-match judge was independently scored by five human annotators.",
"rai:annotatorDemographics": "The original survey answers and event-level source annotations are supplied by the participants whose daily-life records they describe. The released evaluation set retains 42 participants from 70 enrolled. Appendix C and Figure 8 report a participant pool concentrated in the United States (75%), with additional participants in Southeast Asia, Europe, China, and other regions. Reported primary languages are English (51%), Chinese (40%), Korean (3%), and other languages (6%). These are primary-language demographics, not measured language shares of the released text. The paper also reports sleep timing, productive period, work style, and exercise goals; response counts vary by question because some axes were optional and some respondents selected multiple categories. It does not report an aggregate age range or gender breakdown. Separate internal trajectory reviewers and five RMR-validation annotators are described, but their demographics are not reported.",
"rai:machineAnnotationTools": "The default synthesis pipeline uses GPT-4o for image grounding and GPT-5-mini with low reasoning effort for persona and survey-insight distillation, day-skeleton and clue-plan generation, and trajectory expansion. These outputs provide derived descriptions and synthetic behavioural context. Survey answers and the proactive-help and mood source annotations are supplied by the participants. During evaluation, GPT-5 serves as the rationale-match judge; its agreement with five human annotators is reported separately in Appendix F.5.",
"rai:dataBiases": "The retained set contains 42 participants from 70 enrolled, with 40% attrition due to insufficient logging, incomplete surveys, or failure to confirm trajectory fidelity. This selection may favour participants able to sustain detailed self-tracking. The participant pool is concentrated in the United States (75%) and in English- and Chinese-primary-language speakers (51% and 40%, respectively); these demographics do not establish representativeness across cultures, languages, or populations. Self-tracking is sparse and post-hoc biased. Approximately half of the released events are synthetic expansions grounded in real anchors, so generator-specific styles and errors may influence evaluation. Proactive Help and Emotion-Aware Modeling each cover only 15 retained users under task-specific multimodal annotation eligibility rules. Findings should be interpreted within the population and task pools represented by the benchmark.",
"rai:dataUseCases": [
"Evaluating personal intelligence agents on long-horizon real-user daily traces",
"Benchmarking implicit preference reasoning, proactive-help prediction, emotion-aware modeling, and cross-domain transfer",
"Studying long-context and long-horizon memory in large language models and AI agents",
"Research on multilingual personalization and cross-cultural user modeling",
"Research on personalization, user modeling, and human-AI interaction"
],
"rai:dataLimitations": "The benchmark is intended for research evaluation of personal intelligence agents. Its 42 retained participants, 28-to-53-day assembled trajectories, and US-concentrated population limit generalization to broader populations and longer time horizons. Exclusion of 28 of 70 enrolled participants may introduce selection bias. Synthetic expansion increases coverage but introduces dependence on the trajectory-generating models. Proactive Help and Emotion-Aware Modeling each cover only 15 of the 42 users, with 254 and 123 target anchors respectively; the emotion pool is restricted to the last five logged days. Survey answers and event-level source annotations reflect participants' own self-reports. The survey-derived insight in the Lifelong Memory condition is an oracle ceiling, and results under that condition should not be interpreted as a deployable memory setting. Evaluation results do not establish safe or effective deployment of a personal assistant. This distribution omits image bytes and does not include standalone task manifests or the evaluation harness. The full corpus has 18,529 events; paper evaluation counts are not the number of rows in this distribution. Follow task-specific protocols and avoid exposing survey-derived profiles unless permitted by the evaluation condition.",
"rai:dataSocialImpact": "JARVIS-Bench supports research on agents that infer implicit preferences, anticipate assistance needs, and respond to emotional context using longitudinal personal records. Its daily-life traces and personalization targets raise privacy, consent, cultural-skew, and misuse concerns. Safeguards described in the paper include informed consent under University of Illinois Chicago IRB approval, voluntary participation and the right to withdraw, removal of identifying dates, surnames, and precise locations, privacy-constrained image captions and synthetic text, and participant fidelity confirmation. The dataset is provided under CC BY-NC 4.0, with non-commercial research use specified in the paper. Population coverage and synthetic-data limitations should accompany claims made using the benchmark.",
"rai:personalSensitiveInformation": "The source material includes daily routines, photographs, personal preferences, mood reports, and structured persona demographics, which may reveal sensitive aspects of participants' lives. The paper describes an anonymized collection protocol with informed consent under University of Illinois Chicago IRB approval. Calendar dates are stripped at ingestion and replaced by relative day identifiers; surnames, precise locations, and calendar-day identifiers are removed from released artefacts. Photographs enter trajectory text through action-focused third-person captions; the captioning prompt forbids identifying people by name or appearance and transcribing private text visible in images. A structural validator enforces a privacy blacklist for synthetic text, including calendar dates, weekday names, real names, and 12-hour-clock times. Participants were informed of the collected traces, privacy risks, anonymization procedures, voluntary participation, and their right to withdraw. The dataset is provided under CC BY-NC 4.0 for the non-commercial research use described in the paper.",
"rai:dataReleaseMaintenancePlan": "JARVIS-Bench version 1.0.0 is hosted at https://huggingface.co/datasets/WZDavid/JARVIS-Bench under CC BY-NC 4.0. Corrections will be documented through repository revisions. Users can report issues through repository discussions. Evaluation code is associated with https://github.com/AgentMemoryWorld/JARVIS-Bench. No fixed update cadence is specified.",
"distribution": [
{
"@type": "cr:FileObject",
"@id": "events-file",
"name": "data/events.parquet",
"contentUrl": "https://huggingface.co/datasets/WZDavid/JARVIS-Bench/resolve/a934eea5054927c782088900d9974536ef4887e2/data/events.parquet",
"encodingFormat": "application/x-parquet",
"sha256": "37260d0a07dfa78b747b2b3554ca16714b21bbda995aeb4c45f35412483ee96e"
},
{
"@type": "cr:FileObject",
"@id": "days-file",
"name": "data/days.parquet",
"contentUrl": "https://huggingface.co/datasets/WZDavid/JARVIS-Bench/resolve/a934eea5054927c782088900d9974536ef4887e2/data/days.parquet",
"encodingFormat": "application/x-parquet",
"sha256": "cc469b9fcecb90d58dc0833703bcc3d475b702858fc0989c788ab35e13e258cc"
},
{
"@type": "cr:FileObject",
"@id": "profiles-file",
"name": "data/profiles.parquet",
"contentUrl": "https://huggingface.co/datasets/WZDavid/JARVIS-Bench/resolve/a934eea5054927c782088900d9974536ef4887e2/data/profiles.parquet",
"encodingFormat": "application/x-parquet",
"sha256": "ece7dc72a95f5f2fadf9847ee84a68c6a77ffadaac33642f6d00dad085024c1f"
}
],
"prov:wasDerivedFrom": {
"@id": "urn:md5:2d015cf1401784d86c6dff10e0778d57",
"@type": "sc:CreativeWork",
"name": "Author-supplied jarvis-bench_full.zip"
},
"prov:wasGeneratedBy": {
"@id": "text-export",
"@type": "prov:Activity",
"description": "Lossless trajectory-to-Parquet export with image omission; see docs/provenance.md."
},
"recordSet": [
{
"@type": "cr:RecordSet",
"@id": "events",
"name": "events",
"field": [
{
"@type": "cr:Field",
"@id": "events/user_id",
"name": "user_id",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "user_id"
}
}
},
{
"@type": "cr:Field",
"@id": "events/day_key",
"name": "day_key",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "day_key"
}
}
},
{
"@type": "cr:Field",
"@id": "events/event_key",
"name": "event_key",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "event_key"
}
}
},
{
"@type": "cr:Field",
"@id": "events/event_id",
"name": "event_id",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "event_id"
}
}
},
{
"@type": "cr:Field",
"@id": "events/day_index",
"name": "day_index",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "day_index"
}
}
},
{
"@type": "cr:Field",
"@id": "events/seq",
"name": "seq",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "seq"
}
}
},
{
"@type": "cr:Field",
"@id": "events/domain",
"name": "domain",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "domain"
}
}
},
{
"@type": "cr:Field",
"@id": "events/time_bucket",
"name": "time_bucket",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "time_bucket"
}
}
},
{
"@type": "cr:Field",
"@id": "events/is_real_log",
"name": "is_real_log",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Boolean",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "is_real_log"
}
}
},
{
"@type": "cr:Field",
"@id": "events/has_image",
"name": "has_image",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Boolean",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "has_image"
}
}
},
{
"@type": "cr:Field",
"@id": "events/summary_text",
"name": "summary_text",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "summary_text"
}
}
},
{
"@type": "cr:Field",
"@id": "events/description_long",
"name": "description_long",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "description_long"
}
}
},
{
"@type": "cr:Field",
"@id": "events/image_available",
"name": "image_available",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Boolean",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "image_available"
}
}
},
{
"@type": "cr:Field",
"@id": "events/event_json",
"name": "event_json",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "events-file"
},
"extract": {
"column": "event_json"
}
}
}
]
},
{
"@type": "cr:RecordSet",
"@id": "days",
"name": "days",
"field": [
{
"@type": "cr:Field",
"@id": "days/user_id",
"name": "user_id",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "days-file"
},
"extract": {
"column": "user_id"
}
}
},
{
"@type": "cr:Field",
"@id": "days/day_key",
"name": "day_key",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "days-file"
},
"extract": {
"column": "day_key"
}
}
},
{
"@type": "cr:Field",
"@id": "days/day_index",
"name": "day_index",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "days-file"
},
"extract": {
"column": "day_index"
}
}
},
{
"@type": "cr:Field",
"@id": "days/day_metadata_json",
"name": "day_metadata_json",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "days-file"
},
"extract": {
"column": "day_metadata_json"
}
}
}
]
},
{
"@type": "cr:RecordSet",
"@id": "profiles",
"name": "profiles",
"field": [
{
"@type": "cr:Field",
"@id": "profiles/user_id",
"name": "user_id",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "profiles-file"
},
"extract": {
"column": "user_id"
}
}
},
{
"@type": "cr:Field",
"@id": "profiles/profile_json",
"name": "profile_json",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "profiles-file"
},
"extract": {
"column": "profile_json"
}
}
},
{
"@type": "cr:Field",
"@id": "profiles/trajectory_metadata_json",
"name": "trajectory_metadata_json",
"description": "See metadata/schema.json and README.md for field semantics.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "profiles-file"
},
"extract": {
"column": "trajectory_metadata_json"
}
}
}
]
}
],
"datePublished": "2026-11-12"
}