{ "@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" }, "@type": "sc:Dataset", "conformsTo": "http://mlcommons.org/croissant/1.0", "name": "Calibration Routing Audit", "alternateName": [ "calibration-routing-audit" ], "description": "Parsed trial records for the two retained calibration records (Study 1 and Study 2), the machine-readable audit record with its schema, the Study 3 items, pilot and main call logs, protocols (including superseded versions), manifest and summary, the supplement README, and the benchmark, experiment, re-analysis and Study 3 code.", "url": "https://doi.org/10.5281/zenodo.24179075", "sameAs": [ "https://huggingface.co/datasets/ai4he/calibration-routing-audit", "https://openreview.net/forum?id=uPtp2butrm" ], "license": "https://creativecommons.org/licenses/by/4.0/", "version": "1.0.0", "datePublished": "2026-11-12", "inLanguage": "en", "isLiveDataset": false, "keywords": [ "calibration", "verbalized-confidence", "routing", "llm-evaluation", "audit" ], "creator": [ { "@type": "sc:Person", "name": "Carlos Toxtli", "affiliation": { "@type": "sc:Organization", "name": "Clemson University" }, "identifier": "https://orcid.org/0000-0002-9430-2661" }, { "@type": "sc:Person", "name": "Manuel Delaflor", "affiliation": { "@type": "sc:Organization", "name": "Metacognition Institute" } } ], "citeAs": "@inproceedings{Toxtli2026Calibration,\n title = {When Calibration Records Cannot Support Routing: A Denominator and Artifact Audit},\n author = {Toxtli, Carlos and Delaflor, Manuel},\n booktitle = {TAE (Trust-AI-Eval): Can We Trust AI Evaluation? Workshop at NeurIPS 2026},\n address = {Sydney, Australia},\n year = {2026},\n month = dec,\n note = {Poster},\n url = {https://openreview.net/forum?id=uPtp2butrm}\n}", "distribution": [ { "@type": "cr:FileObject", "@id": "code/reanalyze_records.py", "name": "code/reanalyze_records.py", "contentUrl": "code/reanalyze_records.py", "contentSize": "22102 B", "encodingFormat": "text/x-python", "sha256": "5733a496a0ab7bf4659217a6b403e40e56af55f1d60630c54de14c793fcef0b8" }, { "@type": "cr:FileObject", "@id": "code/study2_experiment.py", "name": "code/study2_experiment.py", "contentUrl": "code/study2_experiment.py", "contentSize": "6679 B", "encodingFormat": "text/x-python", "sha256": "a6ac4f3eac3a0e4a47abd7275239a54896fee97dc9a51881622dff49e1fa36cb" }, { "@type": "cr:FileObject", "@id": "code/tcb_benchmark.py", "name": "code/tcb_benchmark.py", "contentUrl": "code/tcb_benchmark.py", "contentSize": "10241 B", "encodingFormat": "text/x-python", "sha256": "51d5fa8c2677ad760010e31e373c4508dc7e2a54ee5616c30ac5573b684dcf2e" }, { "@type": "cr:FileObject", "@id": "code/analysis/verify_study2_gold.py", "name": "code/analysis/verify_study2_gold.py", "contentUrl": "code/analysis/verify_study2_gold.py", "contentSize": "7383 B", "encodingFormat": "text/x-python", "sha256": "b95cc6183105031d6a12f56346adc2b3f45bc04d9c9e5ab233f6d4e225a35030" }, { "@type": "cr:FileObject", "@id": "code/study3/analyze_study3.py", "name": "code/study3/analyze_study3.py", "contentUrl": "code/study3/analyze_study3.py", "contentSize": "14006 B", "encodingFormat": "text/x-python", "sha256": "64048385408a40a08379bd1cfbff8db4c82d05a523d352740aa9a84f8772d1fe" }, { "@type": "cr:FileObject", "@id": "code/study3/study3_record.py", "name": "code/study3/study3_record.py", "contentUrl": "code/study3/study3_record.py", "contentSize": "24903 B", "encodingFormat": "text/x-python", "sha256": "e593833b16de433c2a603e0ae10c90a051f515248e3148eb1abc057587fe0088" }, { "@type": "cr:FileObject", "@id": "data/audit_record.json", "name": "data/audit_record.json", "contentUrl": "data/audit_record.json", "contentSize": "25373 B", "encodingFormat": "application/json", "sha256": "1751e3a0b4f340dc742eec6d82e200173617d513a610c0d95dc19ca3bfd7224a" }, { "@type": "cr:FileObject", "@id": "data/audit_record_excerpt.json", "name": "data/audit_record_excerpt.json", "contentUrl": "data/audit_record_excerpt.json", "contentSize": "3294 B", "encodingFormat": "application/json", "sha256": "f6ba08648dd8a6bcd38652c4212606a1c293ba88a3557c94be7f6aaefa9e2fc9" }, { "@type": "cr:FileObject", "@id": "data/audit_record_schema.json", "name": "data/audit_record_schema.json", "contentUrl": "data/audit_record_schema.json", "contentSize": "1537 B", "encodingFormat": "application/json", "sha256": "343db2a2269781552ed61db235b983d720e085458a8ca8c0731d9aaf8c186f51" }, { "@type": "cr:FileObject", "@id": "data/study1_trial_results.json", "name": "data/study1_trial_results.json", "contentUrl": "data/study1_trial_results.json", "contentSize": "189501 B", "encodingFormat": "application/json", "sha256": "46ca7c2b885e70f362f450c8d0d7dbad150bb59030f0705a155f5ab40a527c17" }, { "@type": "cr:FileObject", "@id": "data/study2_trial_results.json", "name": "data/study2_trial_results.json", "contentUrl": "data/study2_trial_results.json", "contentSize": "51191 B", "encodingFormat": "application/json", "sha256": "5b7374845b0a0be91085ac71803488a83cfb6a7d1ed403825480f7175abf1840" }, { "@type": "cr:FileObject", "@id": "data/analysis/study2_gold_check.json", "name": "data/analysis/study2_gold_check.json", "contentUrl": "data/analysis/study2_gold_check.json", "contentSize": "678 B", "encodingFormat": "application/json", "sha256": "da70042a0497627dd898ccd1fd5888c79ce4e844c7a2b581f940160bb359c870" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_items.json", "name": "data/study3/study3_items.json", "contentUrl": "data/study3/study3_items.json", "contentSize": "63486 B", "encodingFormat": "application/json", "sha256": "e761b57d1bd3c630e468c566aea4848915226f18691b08740a47f1421f91e839" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_main_calls.jsonl", "name": "data/study3/study3_main_calls.jsonl", "contentUrl": "data/study3/study3_main_calls.jsonl", "contentSize": "6792288 B", "encodingFormat": "application/jsonlines", "sha256": "a0ced59fd11fca3a0f2bd5e798a867911198ef71d06d0a9c499e8ae8b9fadd07" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_manifest.json", "name": "data/study3/study3_manifest.json", "contentUrl": "data/study3/study3_manifest.json", "contentSize": "2933 B", "encodingFormat": "application/json", "sha256": "ed40050a2c5716d48fce380297e36c5dd05e8fcc6e722b483e170a5e1d7922ba" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_pilot_calls.jsonl", "name": "data/study3/study3_pilot_calls.jsonl", "contentUrl": "data/study3/study3_pilot_calls.jsonl", "contentSize": "829677 B", "encodingFormat": "application/jsonlines", "sha256": "bc952380d7290021f8efee995c8ad3601a9d92a1c390ac0641ec001ed8443068" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl", "name": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl", "contentUrl": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl", "contentSize": "84147 B", "encodingFormat": "application/jsonlines", "sha256": "ba4e9ae11df44a78ff49711745561a7cfb3e1d15a59461e25af2d1d5929d6f9c" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_pilot_decision.json", "name": "data/study3/study3_pilot_decision.json", "contentUrl": "data/study3/study3_pilot_decision.json", "contentSize": "323 B", "encodingFormat": "application/json", "sha256": "d679e45dc1bf3a905e63f33fcce8479cf50aba0cf7679840574561eed5dd2586" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_pilot_items.json", "name": "data/study3/study3_pilot_items.json", "contentUrl": "data/study3/study3_pilot_items.json", "contentSize": "8139 B", "encodingFormat": "application/json", "sha256": "30d50be382e488d3004ff2bfdde1a9cd025e6a4292aab03330d566ff19269ab3" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_protocol.json", "name": "data/study3/study3_protocol.json", "contentUrl": "data/study3/study3_protocol.json", "contentSize": "5485 B", "encodingFormat": "application/json", "sha256": "a49e9eef4c71f4ad5342223ab93def304b2c1fe748e80c4789fa329c5e605a5d" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_protocol_v1.0_superseded.json", "name": "data/study3/study3_protocol_v1.0_superseded.json", "contentUrl": "data/study3/study3_protocol_v1.0_superseded.json", "contentSize": "4212 B", "encodingFormat": "application/json", "sha256": "964e2222acf720e9cbed5c3cc9473d180497a7d63bcde76190c10d91e8831971" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_protocol_v1.1_superseded.json", "name": "data/study3/study3_protocol_v1.1_superseded.json", "contentUrl": "data/study3/study3_protocol_v1.1_superseded.json", "contentSize": "4773 B", "encodingFormat": "application/json", "sha256": "fc7b4dd1febebe75ab39c0a094e100a5825b455db7f0871532874e524f53b53d" }, { "@type": "cr:FileObject", "@id": "data/study3/study3_summary.json", "name": "data/study3/study3_summary.json", "contentUrl": "data/study3/study3_summary.json", "contentSize": "30896 B", "encodingFormat": "application/json", "sha256": "883cc5c331b1cafe97935e9b379bb14fc8a6dce4cf6d8af9da741c32a7efb1ad" }, { "@type": "cr:FileObject", "@id": "docs/SUPPLEMENT_README.md", "name": "docs/SUPPLEMENT_README.md", "contentUrl": "docs/SUPPLEMENT_README.md", "contentSize": "2400 B", "encodingFormat": "text/markdown", "sha256": "21f87ede632868d368fe02bdb35d6d7090bcf22c5482465e1c6df410d29e606b" } ], "recordSet": [ { "@type": "cr:RecordSet", "@id": "data_study3_study3_main_calls", "name": "data_study3_study3_main_calls", "description": "Records from data/study3/study3_main_calls.jsonl (1,800 rows). Nested fields not described here: request, usage.", "field": [ { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/phase", "name": "phase", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "phase" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/item_id", "name": "item_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "item_id" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/model", "name": "model", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "model" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/family", "name": "family", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "family" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/level", "name": "level", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "level" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/split", "name": "split", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "split" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/gold", "name": "gold", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "gold" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/response_model", "name": "response_model", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "response_model" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/response_id", "name": "response_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "response_id" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/created", "name": "created", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "created" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/finish_reason", "name": "finish_reason", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "finish_reason" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/latency_s", "name": "latency_s", "dataType": "sc:Float", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "latency_s" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/attempts", "name": "attempts", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "attempts" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/content", "name": "content", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "content" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/reasoning_content", "name": "reasoning_content", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "reasoning_content" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/protocol_sha256", "name": "protocol_sha256", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "protocol_sha256" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/utc", "name": "utc", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "utc" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/content_nonempty", "name": "content_nonempty", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "content_nonempty" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/answer_parsed", "name": "answer_parsed", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "answer_parsed" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/answer", "name": "answer", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "answer" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/confidence_parsed", "name": "confidence_parsed", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "confidence_parsed" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/confidence_raw", "name": "confidence_raw", "dataType": "sc:Float", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "confidence_raw" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/confidence_in_range", "name": "confidence_in_range", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "confidence_in_range" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/jointly_usable", "name": "jointly_usable", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "jointly_usable" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_main_calls/correct", "name": "correct", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_main_calls.jsonl" }, "extract": { "column": "correct" } } } ] }, { "@type": "cr:RecordSet", "@id": "data_study3_study3_pilot_calls", "name": "data_study3_study3_pilot_calls", "description": "Records from data/study3/study3_pilot_calls.jsonl (240 rows). Nested fields not described here: request, usage.", "field": [ { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/phase", "name": "phase", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "phase" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/item_id", "name": "item_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "item_id" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/model", "name": "model", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "model" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/family", "name": "family", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "family" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/level", "name": "level", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "level" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/split", "name": "split", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "split" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/gold", "name": "gold", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "gold" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/response_model", "name": "response_model", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "response_model" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/response_id", "name": "response_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "response_id" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/created", "name": "created", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "created" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/finish_reason", "name": "finish_reason", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "finish_reason" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/latency_s", "name": "latency_s", "dataType": "sc:Float", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "latency_s" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/attempts", "name": "attempts", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "attempts" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/content", "name": "content", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "content" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/reasoning_content", "name": "reasoning_content", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "reasoning_content" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/protocol_sha256", "name": "protocol_sha256", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "protocol_sha256" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/utc", "name": "utc", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "utc" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/content_nonempty", "name": "content_nonempty", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "content_nonempty" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/answer_parsed", "name": "answer_parsed", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "answer_parsed" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/answer", "name": "answer", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "answer" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/confidence_parsed", "name": "confidence_parsed", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "confidence_parsed" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/confidence_raw", "name": "confidence_raw", "dataType": "sc:Float", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "confidence_raw" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/confidence_in_range", "name": "confidence_in_range", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "confidence_in_range" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/jointly_usable", "name": "jointly_usable", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "jointly_usable" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls/correct", "name": "correct", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls.jsonl" }, "extract": { "column": "correct" } } } ] }, { "@type": "cr:RecordSet", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside", "name": "data_study3_study3_pilot_calls_v1.0_set_aside", "description": "Records from data/study3/study3_pilot_calls_v1.0_set_aside.jsonl (32 rows). Nested fields not described here: request, usage.", "field": [ { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/phase", "name": "phase", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "phase" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/item_id", "name": "item_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "item_id" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/model", "name": "model", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "model" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/family", "name": "family", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "family" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/level", "name": "level", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "level" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/split", "name": "split", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "split" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/gold", "name": "gold", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "gold" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/response_model", "name": "response_model", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "response_model" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/response_id", "name": "response_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "response_id" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/created", "name": "created", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "created" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/finish_reason", "name": "finish_reason", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "finish_reason" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/latency_s", "name": "latency_s", "dataType": "sc:Float", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "latency_s" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/attempts", "name": "attempts", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "attempts" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/content", "name": "content", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "content" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/reasoning_content", "name": "reasoning_content", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "reasoning_content" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/protocol_sha256", "name": "protocol_sha256", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "protocol_sha256" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/utc", "name": "utc", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "utc" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/content_nonempty", "name": "content_nonempty", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "content_nonempty" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/answer_parsed", "name": "answer_parsed", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "answer_parsed" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/answer", "name": "answer", "dataType": "sc:Text", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "answer" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/confidence_parsed", "name": "confidence_parsed", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "confidence_parsed" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/confidence_raw", "name": "confidence_raw", "dataType": "sc:Integer", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "confidence_raw" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/confidence_in_range", "name": "confidence_in_range", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "confidence_in_range" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/jointly_usable", "name": "jointly_usable", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "jointly_usable" } } }, { "@type": "cr:Field", "@id": "data_study3_study3_pilot_calls_v1.0_set_aside/correct", "name": "correct", "dataType": "sc:Boolean", "source": { "fileObject": { "@id": "data/study3/study3_pilot_calls_v1.0_set_aside.jsonl" }, "extract": { "column": "correct" } } } ] } ] }