Datasets:

narcolepticchicken
/

agent-cost-traces

trace_id stringlengths 11 11	request stringclasses 35 values	task_type stringclasses 9 values	difficulty int64 1 5	model_tier int64 2 5	model_success bool 2 classes	optimal_tier int64 1 5	tool_calls listlengths 0 4	verifier_called bool 2 classes	retries int64 0 3	final_success bool 2 classes	total_cost float64 0.15 1.56	failure_tags listlengths 0 3	cheaper_alternative int64 1 4 ⌀	recovery_action stringclasses 3 values	verifier_needed bool 2 classes	escalation_needed bool 2 classes
trace_00000	Answer based on 50-page document.	retrieval_heavy	2	3	false	5	[]	false	2	true	0.75	[ "model_failure", "retry_loop" ]	null	retry	false	true
trace_00001	Find all payment processing mentions.	retrieval_heavy	2	3	true	2	[]	false	0	true	0.75	[]	2	none	true	false
trace_00002	Answer based on 50-page document.	retrieval_heavy	2	3	true	1	[]	false	0	true	0.75	[]	1	none	false	false
trace_00003	There's a problem with the database.	unknown_ambiguous	3	4	true	2	[]	false	0	true	1	[]	2	none	false	false
trace_00004	Implement LRU cache in Go.	coding	3	4	true	3	[ { "tool": "web_search", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false } ]	false	0	true	1.03	[]	3	none	false	false
trace_00005	Research latest transformer advances.	research	3	4	true	2	[ { "tool": "code_execute", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false } ]	false	0	true	1.03	[]	2	none	false	false
trace_00006	Just tell me what 2+2 is.	quick_answer	1	2	true	1	[]	false	0	true	0.15	[]	1	none	false	false
trace_00007	Query Q3 sales and produce chart.	tool_heavy	2	3	true	3	[ { "tool": "code_execute", "success": true, "cached": false } ]	false	0	true	0.76	[]	null	none	false	false
trace_00008	Search open issues and create summary.	tool_heavy	2	3	true	1	[ { "tool": "web_search", "success": true, "cached": false }, { "tool": "code_execute", "success": false, "cached": false } ]	false	0	true	0.77	[ "tool_failure" ]	1	none	false	false
trace_00009	Answer based on 50-page document.	retrieval_heavy	2	3	true	3	[]	true	0	true	0.77	[]	null	none	true	false
trace_00010	Answer based on 50-page document.	retrieval_heavy	2	3	true	1	[]	false	0	true	0.75	[]	1	none	false	false
trace_00011	Search open issues and create summary.	tool_heavy	2	3	true	1	[ { "tool": "code_execute", "success": false, "cached": false }, { "tool": "file_read", "success": true, "cached": false } ]	false	0	true	0.77	[ "tool_failure" ]	1	none	false	false
trace_00012	What is the capital of France?	quick_answer	1	2	true	2	[]	false	0	true	0.15	[]	null	none	false	false
trace_00013	Investigate data center climate impact.	research	3	4	true	3	[ { "tool": "file_write", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": true } ]	false	0	true	1.03	[]	3	none	true	false
trace_00014	Write a Python function to reverse a linked list.	coding	3	4	true	1	[ { "tool": "file_read", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false } ]	false	0	true	1.02	[]	1	none	false	false
trace_00015	Answer based on 50-page document.	retrieval_heavy	2	3	true	1	[]	false	0	true	0.75	[]	1	none	false	false
trace_00016	Help me with this thing.	unknown_ambiguous	3	4	true	3	[]	false	0	true	1	[]	3	none	false	false
trace_00017	Write email to team about deployment.	document_drafting	2	3	false	2	[]	true	3	false	0.77	[ "model_failure", "retry_loop" ]	2	retry	true	false
trace_00018	Small clarification on this formula.	quick_answer	1	2	true	1	[]	false	0	true	0.15	[]	1	none	false	false
trace_00019	What is the capital of France?	quick_answer	1	2	false	3	[]	true	0	false	0.17	[ "model_failure", "false_done" ]	null	switch_model	true	true
trace_00020	Fetch API docs and generate client code.	tool_heavy	2	3	true	2	[ { "tool": "web_search", "success": true, "cached": true }, { "tool": "web_search", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	0	true	0.79	[]	2	none	false	false
trace_00021	Survey recent RLHF techniques.	research	3	4	false	2	[ { "tool": "code_execute", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false } ]	false	0	false	1.02	[ "model_failure", "false_done" ]	2	switch_model	false	false
trace_00022	Draft project proposal for ML pipeline.	document_drafting	2	3	true	3	[]	false	0	true	0.75	[]	null	none	false	false
trace_00023	Check GDPR compliance for data pipeline urgently.	legal_regulated	5	5	true	3	[]	true	0	true	1.52	[]	3	none	true	false
trace_00024	Help me with this thing.	unknown_ambiguous	3	4	true	3	[]	false	0	true	1	[]	3	none	false	false
trace_00025	Just fix the typo in line 42.	coding	2	3	true	2	[ { "tool": "code_execute", "success": false, "cached": false }, { "tool": "code_execute", "success": false, "cached": false } ]	false	0	true	0.77	[ "tool_failure" ]	2	none	false	false
trace_00026	Small clarification on this formula.	quick_answer	1	2	true	2	[]	false	0	true	0.15	[]	null	none	false	false
trace_00027	Review this contract for liability clauses.	legal_regulated	5	5	true	5	[]	false	0	true	1.5	[]	null	none	true	false
trace_00028	Write email to team about deployment.	document_drafting	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00029	Draft project proposal for ML pipeline.	document_drafting	2	3	true	3	[]	false	0	true	0.75	[]	null	none	false	false
trace_00030	Find all payment processing mentions.	retrieval_heavy	2	3	true	3	[]	false	0	true	0.75	[]	null	none	false	false
trace_00031	Redesign data architecture end-to-end.	long_horizon	4	5	true	4	[]	true	0	true	1.52	[]	4	none	true	false
trace_00032	Fetch API docs and generate client code.	tool_heavy	2	3	false	2	[ { "tool": "code_search", "success": false, "cached": true }, { "tool": "file_write", "success": true, "cached": true }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false } ]	false	2	true	0.79	[ "model_failure", "tool_failure", "retry_loop" ]	2	retry	false	false
trace_00033	Search open issues and create summary.	tool_heavy	2	3	true	3	[ { "tool": "code_execute", "success": true, "cached": false }, { "tool": "code_execute", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false } ]	false	0	true	0.78	[]	null	none	false	false
trace_00034	Plan 3-month roadmap.	long_horizon	4	5	true	3	[]	true	0	true	1.52	[]	3	none	true	false
trace_00035	Fix a typo in the README.	coding	2	3	true	1	[ { "tool": "code_search", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "code_execute", "success": true, "cached": false }, { "tool": "file_read", "success": false, "cached": false } ]	false	0	true	0.79	[ "tool_failure" ]	1	none	false	false
trace_00036	Fetch API docs and generate client code.	tool_heavy	2	3	true	1	[ { "tool": "code_search", "success": true, "cached": false } ]	false	0	true	0.76	[]	1	none	false	false
trace_00037	Find sources comparing LoRA and full FT briefly.	research	2	3	true	2	[ { "tool": "code_search", "success": true, "cached": true }, { "tool": "code_search", "success": true, "cached": true } ]	false	0	true	0.77	[]	2	none	false	false
trace_00038	Answer based on 50-page document.	retrieval_heavy	2	3	true	1	[]	false	0	true	0.75	[]	1	none	false	false
trace_00039	Implement LRU cache in Go.	coding	3	4	true	2	[ { "tool": "file_read", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false } ]	false	0	true	1.02	[]	2	none	false	false
trace_00040	Find sources comparing LoRA and full FT briefly.	research	2	3	false	2	[ { "tool": "code_execute", "success": true, "cached": false }, { "tool": "file_read", "success": true, "cached": true } ]	false	3	false	0.77	[ "model_failure", "retry_loop" ]	2	retry	false	false
trace_00041	Answer based on 50-page document.	retrieval_heavy	2	3	true	2	[]	false	0	true	0.75	[]	2	none	true	false
trace_00042	Write a test for the parser module.	coding	3	4	true	1	[ { "tool": "file_read", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false } ]	false	0	true	1.04	[]	1	none	false	false
trace_00043	Check GDPR compliance for data pipeline urgently.	legal_regulated	5	5	true	3	[]	true	0	true	1.52	[]	3	none	true	false
trace_00044	Orchestrate complete multi-region deployment.	long_horizon	4	5	true	2	[]	true	0	true	1.52	[]	2	none	true	false
trace_00045	Draft project proposal for ML pipeline.	document_drafting	2	3	true	3	[]	false	0	true	0.75	[]	null	none	false	false
trace_00046	Research latest transformer advances.	research	3	4	false	2	[ { "tool": "web_search", "success": true, "cached": false }, { "tool": "web_search", "success": true, "cached": false }, { "tool": "file_read", "success": true, "cached": true } ]	false	0	false	1.03	[ "model_failure", "false_done" ]	2	switch_model	false	false
trace_00047	Verify regulatory compliance for medical data.	legal_regulated	5	5	false	2	[]	false	1	false	1.5	[ "model_failure" ]	2	retry	true	false
trace_00048	I need something about the server.	unknown_ambiguous	3	4	true	3	[]	false	0	true	1	[]	3	none	false	false
trace_00049	Answer based on 50-page document.	retrieval_heavy	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00050	Review this contract for liability clauses.	legal_regulated	5	5	false	2	[]	true	1	true	1.52	[ "model_failure" ]	2	retry	true	false
trace_00051	Verify regulatory compliance for medical data.	legal_regulated	5	5	true	1	[]	true	0	true	1.52	[]	1	none	true	false
trace_00052	Explain quantum computing briefly.	quick_answer	1	2	true	2	[]	false	0	true	0.15	[]	null	none	false	false
trace_00053	Research latest transformer advances.	research	3	4	true	2	[ { "tool": "code_execute", "success": true, "cached": false } ]	false	0	true	1.01	[]	2	none	false	false
trace_00054	Fix a typo in the README.	coding	2	3	false	2	[ { "tool": "file_read", "success": true, "cached": false }, { "tool": "code_execute", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	2	true	0.79	[ "model_failure", "retry_loop" ]	2	retry	false	false
trace_00055	Fetch API docs and generate client code.	tool_heavy	2	3	false	4	[ { "tool": "file_read", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	1	true	0.77	[ "model_failure" ]	null	retry	false	true
trace_00056	Write email to team about deployment.	document_drafting	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00057	Plan 3-month roadmap.	long_horizon	4	5	true	4	[]	true	0	true	1.52	[]	4	none	true	false
trace_00058	Draft project proposal for ML pipeline.	document_drafting	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00059	Fetch API docs and generate client code.	tool_heavy	2	3	false	4	[ { "tool": "code_execute", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	3	false	0.77	[ "model_failure", "retry_loop" ]	null	retry	false	true
trace_00060	Redesign data architecture end-to-end.	long_horizon	4	5	true	3	[]	false	0	true	1.5	[]	3	none	true	false
trace_00061	Debug this critical production segfault NOW.	coding	4	5	true	4	[ { "tool": "file_write", "success": true, "cached": true }, { "tool": "code_search", "success": true, "cached": false } ]	false	0	true	1.52	[]	4	none	false	false
trace_00062	Write a test for the parser module.	coding	3	4	true	3	[ { "tool": "web_search", "success": true, "cached": false }, { "tool": "web_search", "success": false, "cached": false }, { "tool": "code_execute", "success": true, "cached": true }, { "tool": "file_read", "success": true, "cached": false } ]	false	0	true	1.04	[ "tool_failure" ]	3	none	true	false
trace_00063	Find sources comparing LoRA and full FT briefly.	research	2	3	true	3	[ { "tool": "code_execute", "success": true, "cached": false } ]	false	0	true	0.76	[]	null	none	false	false
trace_00064	Find all payment processing mentions.	retrieval_heavy	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00065	Survey recent RLHF techniques.	research	3	4	true	4	[ { "tool": "file_read", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false }, { "tool": "file_read", "success": false, "cached": false } ]	false	0	true	1.04	[ "tool_failure" ]	null	none	false	false
trace_00066	Answer based on 50-page document.	retrieval_heavy	2	3	true	3	[]	true	0	true	0.77	[]	null	none	true	false
trace_00067	Verify regulatory compliance for medical data.	legal_regulated	5	5	true	4	[]	false	0	true	1.5	[]	4	none	true	false
trace_00068	Just tell me what 2+2 is.	quick_answer	1	2	true	2	[]	false	0	true	0.15	[]	null	none	false	false
trace_00069	Just fix the typo in line 42.	coding	2	3	true	2	[ { "tool": "code_execute", "success": true, "cached": false } ]	false	0	true	0.76	[]	2	none	false	false
trace_00070	Help me with this thing.	unknown_ambiguous	3	4	true	3	[]	false	0	true	1	[]	3	none	false	false
trace_00071	Create technical report on performance.	document_drafting	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00072	Find all payment processing mentions.	retrieval_heavy	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00073	Query Q3 sales and produce chart.	tool_heavy	2	3	false	2	[ { "tool": "web_search", "success": true, "cached": false }, { "tool": "code_execute", "success": true, "cached": true }, { "tool": "file_write", "success": true, "cached": false }, { "tool": "file_read", "success": true, "cached": true } ]	false	3	false	0.79	[ "model_failure", "retry_loop" ]	2	retry	false	false
trace_00074	Query Q3 sales and produce chart.	tool_heavy	2	3	true	2	[ { "tool": "file_write", "success": true, "cached": true }, { "tool": "code_execute", "success": true, "cached": false } ]	false	0	true	0.77	[]	2	none	false	false
trace_00075	Review this contract for liability clauses.	legal_regulated	5	5	true	3	[]	true	0	true	1.52	[]	3	none	true	false
trace_00076	Write a test for the parser module.	coding	3	4	true	2	[ { "tool": "web_search", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false }, { "tool": "web_search", "success": false, "cached": false }, { "tool": "web_search", "success": true, "cached": true } ]	false	0	true	1.04	[ "tool_failure" ]	2	none	false	false
trace_00077	Draft project proposal for ML pipeline.	document_drafting	2	3	true	3	[]	false	0	true	0.75	[]	null	none	false	false
trace_00078	Just fix the typo in line 42.	coding	2	3	true	1	[ { "tool": "file_write", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	0	true	0.77	[]	1	none	false	false
trace_00079	Verify regulatory compliance for medical data.	legal_regulated	5	5	true	3	[]	true	0	true	1.52	[]	3	none	true	false
trace_00080	Draft privacy policy section.	legal_regulated	5	5	true	4	[]	false	0	true	1.5	[]	4	none	true	false
trace_00081	Just fix the typo in line 42.	coding	2	3	true	1	[ { "tool": "code_search", "success": true, "cached": false }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "file_read", "success": true, "cached": false } ]	false	0	true	0.78	[]	1	none	false	false
trace_00082	Investigate data center climate impact.	research	3	4	true	3	[ { "tool": "file_write", "success": true, "cached": false } ]	false	0	true	1.01	[]	3	none	false	false
trace_00083	Investigate data center climate impact.	research	3	4	true	4	[ { "tool": "code_execute", "success": true, "cached": false }, { "tool": "web_search", "success": false, "cached": true } ]	false	0	true	1.02	[ "tool_failure" ]	null	none	false	false
trace_00084	Find sources comparing LoRA and full FT briefly.	research	2	3	true	2	[ { "tool": "web_search", "success": false, "cached": true }, { "tool": "file_read", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	0	true	0.78	[ "tool_failure" ]	2	none	false	false
trace_00085	Implement LRU cache in Go.	coding	3	4	true	3	[ { "tool": "file_read", "success": true, "cached": false } ]	true	0	true	1.03	[]	3	none	true	false
trace_00086	Find all payment processing mentions.	retrieval_heavy	2	3	false	1	[]	false	3	false	0.75	[ "model_failure", "retry_loop" ]	1	retry	false	false
trace_00087	Draft project proposal for ML pipeline.	document_drafting	2	3	true	2	[]	true	0	true	0.77	[]	2	none	true	false
trace_00088	Debug this critical production segfault NOW.	coding	4	5	true	4	[ { "tool": "file_write", "success": false, "cached": true }, { "tool": "code_search", "success": false, "cached": true } ]	false	0	true	1.52	[ "tool_failure" ]	4	none	false	false
trace_00089	Draft privacy policy section.	legal_regulated	5	5	true	5	[]	false	0	true	1.5	[]	null	none	true	false
trace_00090	Draft privacy policy section.	legal_regulated	5	5	true	4	[]	true	0	true	1.52	[]	4	none	true	false
trace_00091	There's a problem with the database.	unknown_ambiguous	3	4	true	2	[]	true	0	true	1.02	[]	2	none	true	false
trace_00092	Investigate data center climate impact.	research	3	4	false	5	[ { "tool": "code_search", "success": true, "cached": false }, { "tool": "code_execute", "success": false, "cached": false }, { "tool": "code_search", "success": true, "cached": false }, { "tool": "file_write", "success": true, "cached": false } ]	false	2	true	1.04	[ "model_failure", "tool_failure", "retry_loop" ]	null	retry	false	true
trace_00093	Just tell me what 2+2 is.	quick_answer	1	2	true	1	[]	false	0	true	0.15	[]	1	none	false	false
trace_00094	Answer based on 50-page document.	retrieval_heavy	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false
trace_00095	What is 2+2?	quick_answer	1	2	false	3	[]	false	1	false	0.15	[ "model_failure" ]	null	retry	false	true
trace_00096	Small clarification on this formula.	quick_answer	1	2	true	2	[]	false	0	true	0.15	[]	null	none	false	false
trace_00097	Research latest transformer advances.	research	3	4	false	3	[ { "tool": "code_execute", "success": true, "cached": false } ]	false	1	false	1.01	[ "model_failure" ]	3	retry	false	false
trace_00098	Fix a typo in the README.	coding	2	3	true	3	[ { "tool": "file_write", "success": false, "cached": false }, { "tool": "file_read", "success": true, "cached": true }, { "tool": "code_search", "success": true, "cached": false } ]	false	0	true	0.78	[ "tool_failure" ]	null	none	true	false
trace_00099	Write email to team about deployment.	document_drafting	2	3	true	2	[]	false	0	true	0.75	[]	2	none	false	false

End of preview. Expand in Data Studio

Agent Cost Traces: Synthetic Training Data

10,000 synthetic agent traces for training cost-aware model routers and agent optimizers.

Schema

Field	Type	Description
trace_id	string	Unique identifier
request	string	User request text
task_type	string	One of 9 task categories
difficulty	int	Estimated difficulty (1-5)
model_tier	int	Model tier used (1-5)
model_success	bool	Whether the model succeeded
optimal_tier	int	Minimum tier that would succeed
tool_calls	list	Tool calls made
verifier_called	bool	Whether a verifier was called
retries	int	Number of retries
final_success	bool	Whether the task ultimately succeeded
total_cost	float	Total cost of the run
failure_tags	list	Tags describing failures
cheaper_alternative	int/null	Cheaper tier that would also work
recovery_action	string	What recovery was attempted
verifier_needed	bool	Whether verification was needed
escalation_needed	bool	Whether a higher tier was needed

Task Types and Difficulty

Task Type	Base Difficulty	Floor Tier
quick_answer	1	1
document_drafting	2	2
tool_heavy	2	2
retrieval_heavy	2	2
research	3	3
coding	3	3
unknown_ambiguous	3	3
long_horizon	4	3
legal_regulated	5	4

Success Model

P(success) = tier_strength^(difficulty × 0.6)

Tier strengths: tiny=0.35, cheap=0.55, medium=0.80, frontier=0.93, specialist=0.97

Use Cases

Train model cascade routers
Evaluate cost-quality frontiers
Benchmark cost-aware agent optimizers
Study failure recovery policies
Research selective verification strategies

⚠️ Limitations

This is synthetic data. Real execution data produces dramatically better routers. On SWE-bench, the v8 router trained on this synthetic data increased cost by 11.6%, while v10 trained on just 500 real outcomes saved 23.3%. Use this data for prototyping and sanity checks, not for production routing.

For real execution data, see:

License

MIT

Downloads last month: 32