Datasets:

Flownium
/

zeus-30k-trace-dataset

api_calls_json stringlengths 24.7k 1.22M	category stringclasses 7 values	completed_at stringlengths 32 32	context_json stringlengths 2 222	decision_example_count int64 1 5	family_id stringclasses 112 values	final_answer stringlengths 51 911 ⌀	final_workspace_files_json stringlengths 4 1.24k	messages_json stringlengths 755 20.7k	model_settings_json stringclasses 4 values	prompt stringlengths 138 366	split stringclasses 1 value	started_at stringlengths 32 32	task_hash stringlengths 64 64	task_id stringlengths 11 22	termination stringclasses 1 value	timing_json stringlengths 82 109	token_usage_json stringlengths 133 143	tool_definitions_json stringclasses 4 values	tool_events_json stringlengths 2 4.03k	trace_id stringlengths 11 22	verification_json stringlengths 137 5.24k
[{"attempt":1,"completed_at":"2026-10-05T06:04:59.350737+00:00","elapsed_seconds":5.262603166978806,"http_status":200,"request":{"max_tokens":2048,"messages":[{"content":"You are completing a task for an execution-trace dataset. Follow the supplied task and output format. Use only supplied facts for factual claims. Use...	clarification	2026-10-05T06:04:59.366851+00:00	{}	2	clarification/currency	{ "action": "clarify", "missing_field": "target_currency", "question": "What is the target currency for Project Juniper 109 that 228 USD should be converted to?" }	null	[{"content":"You are completing a task for an execution-trace dataset. Follow the supplied task and output format. Use only supplied facts for factual claims. Use available tools when required. Treat tool outputs as observations, follow the explicit retry policy, and ask for necessary missing information. Complete ordi...	{"max_tokens":2048,"model":"qwen3.8-27b","stream":true,"temperature":0.3,"thinking":"provider_default","top_p":0.9}	Convert 228 from USD to the other currency for Project Juniper 109. A necessary detail is missing: ask one specific clarification question instead of inventing it. Return JSON with "action": "clarify" and "missing_field" naming that detail, plus "question".	train	2026-10-05T06:04:54.088193+00:00	f28543cd05025627b3cd1dcaf51a45171d18425ceb4bc08e9f99057f0b093ae4	clarification-00000	completed	{"api_seconds":5.262603166978806,"elapsed_seconds":5.278744250070304,"tool_seconds":0}	{"cached_prompt_tokens":0,"completion_tokens":191,"prompt_tokens":169,"reasoning_tokens":136,"total_tokens":360,"usage_complete":true}	[]	[]	clarification-00000	{"checks":{"clarification_action":true,"question_present":true,"supported_missing_field_alias":true,"valid_json_object":true},"method":"explicit_missing_information_schema_with_field_aliases; question_semantics_not_fully_verified","status":"verified","task_success":true}
[{"attempt":1,"completed_at":"2026-10-05T06:05:00.530860+00:00","elapsed_seconds":6.440781916026026,"http_status":200,"request":{"max_tokens":2048,"messages":[{"content":"You are completing a task for an execution-trace dataset. Follow the supplied task and output format. Use only supplied facts for factual claims. Use...	clarification	2026-10-05T06:05:00.536641+00:00	{}	2	clarification/timezone	{ "action": "clarify", "missing_field": "local_timezone", "question": "What is your local timezone (e.g., UTC-5, America/New_York, or a specific city) so I can convert 10:00 on 2026-01-19 to UTC?" }	null	[{"content":"You are completing a task for an execution-trace dataset. Follow the supplied task and output format. Use only supplied facts for factual claims. Use available tools when required. Treat tool outputs as observations, follow the explicit retry policy, and ask for necessary missing information. Complete ordi...	{"max_tokens":2048,"model":"qwen3.8-27b","stream":true,"temperature":0.3,"thinking":"provider_default","top_p":0.9}	Schedule Project Aurora 2453 for 10:00 on 2026-01-19, converting my local time to UTC. A necessary detail is missing: ask one specific clarification question instead of inventing it. Return JSON with "action": "clarify" and "missing_field" naming that detail, plus "question".	train	2026-10-05T06:04:54.090164+00:00	f934b6b1e3aecf8f2e6cdac809ff25f69f5d565e79db294cd38103293bf166c1	clarification-00001	completed	{"api_seconds":6.440781916026026,"elapsed_seconds":6.446585917030461,"tool_seconds":0}	{"cached_prompt_tokens":0,"completion_tokens":246,"prompt_tokens":184,"reasoning_tokens":167,"total_tokens":430,"usage_complete":true}	[]	[]	clarification-00001	{"checks":{"clarification_action":true,"question_present":true,"supported_missing_field_alias":true,"valid_json_object":true},"method":"explicit_missing_information_schema_with_field_aliases; question_semantics_not_fully_verified","status":"verified","task_success":true}
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:10.369171+00:00\",\"elapsed_seconds\":16.277166(...TRUNCATED)	clarification	2026-10-05T06:05:10.387316+00:00	{}	1	clarification/date	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"time\",\n \"question\": \"What time on n(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Book the Project Quartz 7070 workshop for next Friday. A necessary detail is missing: ask one speci(...TRUNCATED)	train	2026-10-05T06:04:54.092261+00:00	6c58646fc6c2c4d44b7d56dfc0d488bb0979b643806141ce00003134f6ce1efe	clarification-00002	completed	{"api_seconds":16.277166791027412,"elapsed_seconds":16.295334749971516,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":689,\"prompt_tokens\":163,\"reasoning_tokens\":64(...TRUNCATED)	[]	[]	clarification-00002	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:01.107223+00:00\",\"elapsed_seconds\":7.0131280(...TRUNCATED)	clarification	2026-10-05T06:05:01.116237+00:00	{}	2	clarification/destination	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"destination\",\n \"question\": \"What is(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Find travel options from Sydney for Project Cedar 3208. A necessary detail is missing: ask one spec(...TRUNCATED)	train	2026-10-05T06:04:54.094189+00:00	9ed56ffc0a8289f06f73d84b35edae55b89e2c9d8a85e84f9db99473848c77cb	clarification-00003	completed	{"api_seconds":7.01312808401417,"elapsed_seconds":7.022165665985085,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":267,\"prompt_tokens\":163,\"reasoning_tokens\":21(...TRUNCATED)	[]	[]	clarification-00003	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:04.065370+00:00\",\"elapsed_seconds\":4.3886723(...TRUNCATED)	clarification	2026-10-05T06:05:04.083543+00:00	{}	2	clarification/quantity	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"quantity\",\n \"question\": \"What is th(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Calculate the total for Project Cedar 1442 at 74 per unit. A necessary detail is missing: ask one s(...TRUNCATED)	train	2026-10-05T06:04:59.676713+00:00	4c1512237fad293ab9b8cd3ca13f13c965e0b7b89181bb5d8a2141919f31ce9a	clarification-00004	completed	{"api_seconds":4.388672333909199,"elapsed_seconds":4.406902333023027,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":179,\"prompt_tokens\":167,\"reasoning_tokens\":11(...TRUNCATED)	[]	[]	clarification-00004	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:05.769020+00:00\",\"elapsed_seconds\":4.9480096(...TRUNCATED)	clarification	2026-10-05T06:05:05.777719+00:00	{}	2	clarification/recipient	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"recipient_name\",\n \"question\": \"What(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Draft a personalised invitation for Project Juniper 5508, using the recipient’s name. A necessary(...TRUNCATED)	train	2026-10-05T06:05:00.821039+00:00	6f9b4c8ff2a0848499987788ee2f5098c7f51012a66527a34239b085b1875abd	clarification-00005	completed	{"api_seconds":4.948009624960832,"elapsed_seconds":4.956756500061601,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":209,\"prompt_tokens\":169,\"reasoning_tokens\":15(...TRUNCATED)	[]	[]	clarification-00005	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:10.264357+00:00\",\"elapsed_seconds\":6.1764711(...TRUNCATED)	clarification	2026-10-05T06:05:10.287893+00:00	{}	2	clarification/format	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"target_file_format\",\n \"question\": \"(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Export the Project Aurora 1220 record to my required file format. A necessary detail is missing: as(...TRUNCATED)	train	2026-10-05T06:05:04.087933+00:00	914421da43e1f7971bd88d984698df52302e936f4a1a7be380dcf719bfd9686a	clarification-00006	completed	{"api_seconds":6.1764711670111865,"elapsed_seconds":6.200064167031087,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":274,\"prompt_tokens\":165,\"reasoning_tokens\":20(...TRUNCATED)	[]	[]	clarification-00006	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:16.797339+00:00\",\"elapsed_seconds\":9.3862932(...TRUNCATED)	clarification	2026-10-05T06:05:16.816044+00:00	{}	2	clarification/record_id	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"order_number\",\n \"question\": \"What i(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Get the current status of my Project Juniper 864 order. A necessary detail is missing: ask one spec(...TRUNCATED)	train	2026-10-05T06:05:07.411022+00:00	e74dec5595d4f8adb89f755c118c79feaac364792d61bfdde9b387b3e851ca1f	clarification-00007	completed	{"api_seconds":9.386293250019662,"elapsed_seconds":9.405182166025043,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":419,\"prompt_tokens\":164,\"reasoning_tokens\":35(...TRUNCATED)	[]	[]	clarification-00007	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:15.811570+00:00\",\"elapsed_seconds\":5.4102726(...TRUNCATED)	clarification	2026-10-05T06:05:15.823094+00:00	{}	2	clarification/unit	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"source_unit\",\n \"question\": \"What un(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Convert the measurement 266 for Project Cedar 3117 to metres. A necessary detail is missing: ask on(...TRUNCATED)	train	2026-10-05T06:05:10.401298+00:00	d6a35ee4042db010a73f04e787b8b6acc267f6a10711dbff6aa4c338a73c8d31	clarification-00008	completed	{"api_seconds":5.41027262492571,"elapsed_seconds":5.421889750054106,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":231,\"prompt_tokens\":167,\"reasoning_tokens\":16(...TRUNCATED)	[]	[]	clarification-00008	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)
"[{\"attempt\":1,\"completed_at\":\"2026-10-05T06:05:18.240497+00:00\",\"elapsed_seconds\":5.0251041(...TRUNCATED)	clarification	2026-10-05T06:05:18.256591+00:00	{}	2	clarification/time	"\n\n{\n \"action\": \"clarify\",\n \"missing_field\": \"preferred_time\",\n \"question\": \"What(...TRUNCATED)	null	"[{\"content\":\"You are completing a task for an execution-trace dataset. Follow the supplied task (...TRUNCATED)	"{\"max_tokens\":2048,\"model\":\"qwen3.8-27b\",\"stream\":true,\"temperature\":0.3,\"thinking\":\"p(...TRUNCATED)	"Schedule Project Quartz 9102 on 2026-07-17 at my preferred time. A necessary detail is missing: ask(...TRUNCATED)	train	2026-10-05T06:05:13.215397+00:00	5a8304d1808f2b288b696547a1af410b294833c2bc77f321a5ccbaf741d3fec8	clarification-00009	completed	{"api_seconds":5.02510416705627,"elapsed_seconds":5.041277499985881,"tool_seconds":0}	"{\"cached_prompt_tokens\":0,\"completion_tokens\":216,\"prompt_tokens\":174,\"reasoning_tokens\":15(...TRUNCATED)	[]	[]	clarification-00009	"{\"checks\":{\"clarification_action\":true,\"question_present\":true,\"supported_missing_field_alia(...TRUNCATED)

End of preview. Expand in Data Studio

Zeus Qwen execution dataset

Exactly 30,000 unique solver-input traces generated through the user-provided Qwen3.8-27B API. The endpoint reports this model ID; weights were not independently inspected. Thirty concurrent API workers were requested, with fallback to twenty after repeated overload errors.

Categories: {"clarification":2000,"coding":6000,"documents":5000,"failure_recovery":2000,"maths":6000,"tool_use":4000,"writing":5000}.

traces: stable-schema trace records; structured fields are lossless JSON strings.
raw_traces: full nested trace records for direct JSONL use.
api_calls: durable API attempts, including retries and interrupted attempts.
laya_decisions: question/options with supported labels and prefix-only inputs.
ordinary_responses: original final responses with context and verification eligibility.
task_manifest.jsonl.gz: task provenance, fixtures, and private grader references.
split_manifest.json: task families assigned wholly to a split.
verification_report.json: observed checks, failures, limitations, and token accounting.

Input boundaries

For decision adaptation use only input_json as input and target_json as target. Never feed verification, evidence, hidden test cases, later messages, or manifest references into a model. Related task families are kept in one split; parameter variations are not independent tasks for evaluation purposes.

Verification and limitations

Coding is executed in a restricted subprocess and checked with held-out tests. Mathematical numeric answers use independent reference computations; explanation presence is checked, while explanation semantics are unreviewed. Clarification checks cover structured field aliases and question presence, while full question semantics are unreviewed. Document extraction and database workflows are checked against controlled fixtures and explicit policies. These fixtures establish truth within the supplied synthetic environment. Writing schema, required values, and word counts are checked, but full prose semantics are unreviewed. Failures and uncertain outputs remain labeled; they are not automatically positive imitation examples. The original collection did not include semantic review; the correction revision includes scoped review documented below.

Timing and cost

Token usage includes reasoning and repeated prompts. Dollar cost is unknown because the custom provider supplied no billing rates or cost totals. API timing reflects actual calls and tools. Credentials and Authorization headers are redacted.

Reproducibility

Task generation, verifiers, tool runner, exporter, and configuration are in collector_source/. Resume state is local SQLite; public exports are sharded compressed JSONL.

Correction revision: 2026-10-06

This revision preserves all 30,000 original trace IDs and existing dataset configurations. 2,142 selected traces were regenerated through actual API calls, 127 derived final responses were edited with independent checks, 97 regenerated final responses received independently reviewed derived edits, and numeric decision options were corrected to remove the universal median-answer shortcut. Related canonical views were replaced consistently in one atomic commit.

Original Git history retains revision 6f8a6a76dd65cb8a6413f48f6506fff245635d73. Actual original API attempts remain in api_calls; new attempts and quality audit calls are appended. Derived final edits explicitly retain the original API responses and distinguish edited finals from recorded provider output. Journals and quality attempts are in revisions/2026-10-06/. Original collection code remains under collector_source/; correction code is under revisions/2026-10-06/collector_source/. Journals contain private references and must never be model inputs.

QA verifies exact IDs, stable public tabular schemas, source fingerprints, independent numeric references, actual tool/API provenance, and unchanged-record hashes. Semantic review and same-provider critique are scoped evidence, not proof of all claims. Finite coding tests and writing checks retain their documented limitations. Preserved reasoning qualifications are also present in traces.verification_json. An ordinary-response row with a preserved explicitly incorrect reasoning span is not marked as an eligible positive full-response example, even when its derived final content is correct. Component-qualified spans must be handled separately when training.

Privacy update — 2026-10-10

Identifying provider hostnames in generation metadata and saved transport errors have been replaced with redacted.invalid. Trace IDs, task content, model answers, labels, and split assignments are unchanged.