RFSchemBench / croissant.json
anonymous-submission042's picture
v1.1: drop question_zh and task_id; add hasSyntheticData + provenance fields
d4c3c2d verified
{
"@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/",
"examples": {
"@id": "cr:examples",
"@type": "@json"
},
"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",
"prov": "http://www.w3.org/ns/prov#",
"wasDerivedFrom": "prov:wasDerivedFrom",
"wasGeneratedBy": "prov:wasGeneratedBy"
},
"@type": "sc:Dataset",
"conformsTo": "http://mlcommons.org/croissant/1.0",
"name": "RFSchemBench",
"description": "A multimodal LLM evaluation benchmark for RF circuit schematic understanding. 2,348 questions across 590 schematic pages from 5 publicly available RF data sources, organized in a four-level semantic hierarchy: Component Understanding, Structural Understanding, Functional Understanding, and Dynamic Reasoning (simulation-grounded counterfactual plot reasoning). Constructed via expert-rule-guided programmatic generation; LLMs are deliberately excluded from the gold-answer path.",
"license": "https://creativecommons.org/licenses/by/4.0/",
"url": "https://huggingface.co/datasets/anonymous-submission042/RFSchemBench",
"version": "1.0.0",
"citeAs": "@misc{rfschembench2026, title={RFSchemBench: A Multi-Source, Hierarchically-Structured Multimodal Benchmark for RF Circuit Schematic Understanding}, author={Anonymous}, year={2026}, note={Submitted to NeurIPS 2026 Evaluations \\& Datasets Track}}",
"creator": {
"@type": "sc:Organization",
"name": "Anonymous (double-blind submission)"
},
"keywords": [
"RF",
"circuit",
"schematic",
"multimodal",
"benchmark",
"VQA",
"engineering"
],
"datePublished": "2026-05-01",
"isLiveDataset": false,
"distribution": [
{
"@type": "cr:FileObject",
"@id": "permissive-parquet",
"name": "permissive-parquet",
"description": "Parquet shard for the `permissive` configuration.",
"contentUrl": "data/permissive/test-00000-of-00001.parquet",
"encodingFormat": "application/vnd.apache.parquet",
"license": "https://creativecommons.org/licenses/by/4.0/",
"sha256": "8bb150251f50574342264ae3a973aff0d7fa74f8d6a9192bbd20ea7df468d866"
},
{
"@type": "cr:FileObject",
"@id": "nc_allowed-parquet",
"name": "nc_allowed-parquet",
"description": "Parquet shard for the `nc_allowed` configuration.",
"contentUrl": "data/nc_allowed/test-00000-of-00001.parquet",
"encodingFormat": "application/vnd.apache.parquet",
"license": "https://creativecommons.org/licenses/by-nc-sa/4.0/",
"sha256": "4b1f90bf3436d76c1d95792e536f64fd0723bd5a1b97e52c0c923b9b9703d9c2"
}
],
"recordSet": [
{
"@type": "cr:RecordSet",
"@id": "permissive",
"name": "RFSchemBench - permissive subset",
"description": "RFSchemBench rows from sources licensed under CC-BY-4.0-compatible terms (excludes the m17 NC-licensed source class). N=2,258.",
"field": [
{
"@type": "cr:Field",
"@id": "permissive/question_id",
"name": "question_id",
"description": "Stable unique identifier",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "question_id"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/item_id",
"name": "item_id",
"description": "Source schematic identifier",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "item_id"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/source",
"name": "source",
"description": "Source class: qucs / kicad / myriadrf / m17 / oresat",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "source"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/level",
"name": "level",
"description": "One of: Component Understanding / Structural Understanding / Functional Understanding / Dynamic Reasoning",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "level"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/category",
"name": "category",
"description": "Coarse-grained category tag",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "category"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/question",
"name": "question",
"description": "English question text (what models are evaluated on)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "question"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/image",
"name": "image",
"description": "Primary schematic rendering (PNG bytes embedded)",
"dataType": "sc:ImageObject",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "image"
},
"transform": {
"jsonPath": "$.bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/context_images",
"name": "context_images",
"description": "Auxiliary context images (Dynamic Reasoning rows only). List of {caption, image-bytes}.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "context_images"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/options",
"name": "options",
"description": "Multi-choice options for Dynamic Reasoning rows. List of {label, text, image-bytes}.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "options"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/answer_type",
"name": "answer_type",
"description": "enum_label / comma_separated_list / integer / short_text",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "answer_type"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/answer_allowed",
"name": "answer_allowed",
"description": "Permitted enum values (empty for non-enum types)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "answer_allowed"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/answer",
"name": "answer",
"description": "Gold answer; for list-type, comma-separated",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "answer"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/source_schematic",
"name": "source_schematic",
"description": "Provenance: original .kicad_sch / .sch path",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "source_schematic"
}
}
},
{
"@type": "cr:Field",
"@id": "permissive/license",
"name": "license",
"description": "Per-row license tag (CC-BY-4.0 or CC-BY-NC-SA-4.0)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "permissive-parquet"
},
"extract": {
"column": "license"
}
}
}
]
},
{
"@type": "cr:RecordSet",
"@id": "nc_allowed",
"name": "RFSchemBench - NC-allowed full subset",
"description": "Full RFSchemBench (all 5 source classes, N=2,348). Includes 90 rows under CC-BY-NC-SA-4.0. NonCommercial usage only.",
"field": [
{
"@type": "cr:Field",
"@id": "nc_allowed/question_id",
"name": "question_id",
"description": "Stable unique identifier",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "question_id"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/item_id",
"name": "item_id",
"description": "Source schematic identifier",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "item_id"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/source",
"name": "source",
"description": "Source class: qucs / kicad / myriadrf / m17 / oresat",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "source"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/level",
"name": "level",
"description": "One of: Component Understanding / Structural Understanding / Functional Understanding / Dynamic Reasoning",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "level"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/category",
"name": "category",
"description": "Coarse-grained category tag",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "category"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/question",
"name": "question",
"description": "English question text (what models are evaluated on)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "question"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/image",
"name": "image",
"description": "Primary schematic rendering (PNG bytes embedded)",
"dataType": "sc:ImageObject",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "image"
},
"transform": {
"jsonPath": "$.bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/context_images",
"name": "context_images",
"description": "Auxiliary context images (Dynamic Reasoning rows only). List of {caption, image-bytes}.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "context_images"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/options",
"name": "options",
"description": "Multi-choice options for Dynamic Reasoning rows. List of {label, text, image-bytes}.",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "options"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/answer_type",
"name": "answer_type",
"description": "enum_label / comma_separated_list / integer / short_text",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "answer_type"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/answer_allowed",
"name": "answer_allowed",
"description": "Permitted enum values (empty for non-enum types)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "answer_allowed"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/answer",
"name": "answer",
"description": "Gold answer; for list-type, comma-separated",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "answer"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/source_schematic",
"name": "source_schematic",
"description": "Provenance: original .kicad_sch / .sch path",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "source_schematic"
}
}
},
{
"@type": "cr:Field",
"@id": "nc_allowed/license",
"name": "license",
"description": "Per-row license tag (CC-BY-4.0 or CC-BY-NC-SA-4.0)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "nc_allowed-parquet"
},
"extract": {
"column": "license"
}
}
}
]
}
],
"rai:dataCollection": "Schematic source files (.kicad_sch, .qucs) were collected from publicly available GitHub repositories for the qucs / kicad / myriadrf / m17 / oresat ecosystems. Pages were rendered to PNG via KiCad CLI and Qucs native renderers. A page-level Qwen3.6 RF-relevance gate filtered out non-RF pages.",
"rai:dataCollectionType": "Crawled / programmatically rendered from authoritative open-source schematic repositories.",
"rai:dataCollectionRawData": "Original `.kicad_sch` and `.qucs` files plus their rendered images.",
"rai:dataPreprocessingProtocol": "Each schematic page is rendered to a primary PNG image. For Dynamic Reasoning, ngspice simulation outputs and counterfactual variant plots are pre-computed. RF-relevance gating applied at the page level.",
"rai:dataAnnotationProtocol": "Gold answers are produced deterministically by Python rules authored by domain experts, extracting structured ground truth from KiCad CLI / Qucs / ngspice outputs. LLMs are NOT used in the gold-answer path. Iterative rule refinement: experts review generated questions and patch rules when systematic errors are found; the released release reflects the latest revisions.",
"rai:dataAnnotationPlatform": "Custom Python pipelines under `qa_pipelines/<source>/` in the source repository.",
"rai:dataAnnotationAnalysis": "An independent post-hoc gold audit on a stratified sample is planned as part of the validity protocol; results will be reported in the camera-ready paper.",
"rai:annotationsPerItem": "1 gold answer per question, deterministic from rules.",
"rai:annotatorDemographics": "Gold answers produced by deterministic rule code; no human annotators per item.",
"rai:dataUseCases": [
"Evaluation of multimodal LLMs on RF circuit schematic understanding.",
"Diagnostic analysis of structural connectivity reasoning in vision-language models.",
"Reasoning-mode (thinking on/off) ablation studies."
],
"rai:dataLimitations": "Per-source-class question counts are imbalanced (40 to 974); per-source claims should be reported with N. Dynamic Reasoning subset (N=55) exists only on simulation-capable source class. Questions evaluated in English; Chinese parallel translations provided for reference. Single-image-per-question protocol.",
"rai:dataBiases": "Source-class bias: the largest source class contributes ~41% of questions, which can dominate micro accuracy. We provide a `permissive` config that excludes the NonCommercial source and recommend reporting both micro and source-balanced macro accuracy. Domain bias: schematics are biased toward open-source SDR / amateur radio / satellite hardware; commercial proprietary RF designs are not represented.",
"rai:dataReleaseMaintenancePlan": "Versioned releases under semver. Discussion thread is open for gold-answer corrections and parser / scorer disputes. After acceptance, this anonymous repository will be transferred to a maintainer account.",
"rai:personalSensitiveInformation": "None. The dataset contains schematic images from open-source hardware projects, with no personally identifiable information, biometric data, or sensitive content.",
"rai:dataSocialImpact": "Positive: enables systematic evaluation of multimodal LLMs on RF engineering tasks, supporting safer deployment in EE / RF design workflows. Risk: LLMs that perform well on this benchmark are not certified for production EE design. Use as evaluation tool, not deployment certification.",
"rai:hasSyntheticData": false,
"prov:wasDerivedFrom": [
{
"@type": "sc:Dataset",
"name": "Qucs-S example library (RF subset)",
"description": "Educational RF circuit examples shipped with Qucs-S; rendered to PNG via Qucs native renderer.",
"url": "https://github.com/ra3xdh/qucs_s"
},
{
"@type": "sc:Dataset",
"name": "KiCad public RF schematic projects",
"description": "Curated set of public GitHub KiCad projects with RF subsystems; rendered to PNG via KiCad CLI."
},
{
"@type": "sc:Dataset",
"name": "MyriadRF / LimeSDR hardware schematics",
"description": "Professional SDR board schematics released by Lime Microsystems and the MyriadRF community.",
"url": "https://github.com/myriadrf"
},
{
"@type": "sc:Dataset",
"name": "M17 Project digital radio hardware",
"description": "Open-source digital radio community hardware schematics (CC-BY-NC-SA-4.0 — used in nc_allowed config only).",
"url": "https://github.com/M17-Project"
},
{
"@type": "sc:Dataset",
"name": "OreSat satellite GPS module",
"description": "OreSat space hardware schematics covering RF / GPS subsystems.",
"url": "https://github.com/oresat"
}
],
"prov:wasGeneratedBy": {
"@type": "prov:Activity",
"name": "RFSchemBench construction pipeline",
"description": "(1) Schematic source files from open-source repos rendered to PNG via KiCad CLI / Qucs native; (2) Page-level RF-relevance gate (Qwen3.6, used only as filter, not for gold answers); (3) Domain-expert-authored Python rules (`qa_pipelines/<source>/build_tier{1..4}_question_bank.py`) deterministically extract structured ground truth from KiCad netlist / Qucs schematic graph / ngspice simulation outputs and emit (question, gold) tuples; (4) Iterative rule refinement: experts review samples and patch rules upon discovering systematic errors."
}
}