Datasets:

observability
/

ai-observability-events

trace_id stringclasses 12 values	span_id stringclasses 8 values	parent_span_id stringclasses 6 values	event_type stringlengths 5 16	component stringlengths 3 17	operation stringlengths 6 23	status stringclasses 8 values	duration_ms int64 2 4.18k	model stringclasses 10 values	tool stringclasses 4 values	input_tokens int64 532 1.61k ⌀	output_tokens int64 118 320 ⌀	cost_usd float64 0 0.03 ⌀	verification stringclasses 2 values	message stringclasses 2 values	metadata unknown
trace-001	span-001	null	request	gateway	user_request	ok	4	null	null	null	null	null	null	null	{ "channel": "api" }
trace-001	span-002	span-001	model_call	llm	generate	ok	812	reasoning-model-a	null	842	221	0.0192	null	null	{ "prompt_version": "v3", "provider": "example-provider" }
trace-001	span-003	span-002	verification	verifier	answer_check	pass	131	null	null	null	null	0.0031	pass	null	{ "method": "independent-check" }
trace-001	span-004	span-001	response	gateway	return_response	ok	3	null	null	null	null	null	null	null	{}
trace-002	span-001	null	request	gateway	user_request	ok	3	null	null	null	null	null	null	null	{}
trace-002	span-002	span-001	retrieval	retriever	semantic_search	ok	91	null	null	null	null	null	null	null	{ "documents_found": 12, "top_k": 6 }
trace-002	span-003	span-002	rerank	reranker	rerank_chunks	ok	47	reranker-a	null	null	null	null	null	null	{ "candidates": 6, "selected": 3 }
trace-002	span-004	span-001	model_call	llm	generate_with_context	ok	744	model-b	null	1,610	287	0.0244	null	null	{ "retrieved_chunks": 3 }
trace-002	span-005	span-004	evaluation	evaluator	faithfulness	pass	109	null	null	null	null	null	pass	null	{ "score": 0.91 }
trace-002	span-006	span-001	response	gateway	return_response	ok	4	null	null	null	null	null	null	null	{}
trace-003	span-001	null	request	agent_runtime	task_received	ok	5	null	null	null	null	null	null	null	{}
trace-003	span-002	span-001	planning	agent_runtime	create_plan	ok	96	planner-model	null	null	null	null	null	null	{}
trace-003	span-003	span-002	tool_call	tool_runtime	web_tool	error	180	null	web-search	null	null	null	null	upstream timeout	{ "attempt": 1 }
trace-003	span-004	span-003	retry	tool_runtime	retry_tool	error	205	null	web-search	null	null	null	null	upstream timeout	{ "attempt": 2, "backoff_ms": 250 }
trace-003	span-005	span-002	fallback	router	select_alternate_tool	ok	26	null	null	null	null	null	null	null	{ "fallback_reason": "repeated_timeout", "selected_tool": "web-search-secondary" }
trace-003	span-006	span-005	tool_call	tool_runtime	web_tool	ok	166	null	web-search-secondary	null	null	null	null	null	{ "attempt": 1 }
trace-003	span-007	span-006	verification	verifier	tool_result_check	pass	77	null	null	null	null	null	pass	null	{}
trace-003	span-008	span-001	response	agent_runtime	complete_task	ok	6	null	null	null	null	null	null	null	{}
trace-004	span-001	null	request	agent_runtime	resume_task	ok	4	null	null	null	null	null	null	null	{}
trace-004	span-002	span-001	memory_read	memory	retrieve_context	ok	38	null	null	null	null	null	null	null	{ "memory_items": 4, "freshness_days": 2 }
trace-004	span-003	span-001	model_call	llm	reason_with_memory	ok	623	agent-model-a	null	1,342	198	0.0179	null	null	{}
trace-004	span-004	span-003	memory_write	memory	persist_decision	ok	29	null	null	null	null	null	null	null	{ "memory_type": "episodic", "items_written": 1 }
trace-004	span-005	span-001	checkpoint	agent_runtime	save_checkpoint	ok	17	null	null	null	null	null	null	null	{}
trace-004	span-006	span-001	response	agent_runtime	complete_step	ok	3	null	null	null	null	null	null	null	{}
trace-005	span-001	null	request	agent_runtime	high_impact_action	ok	5	null	null	null	null	null	null	null	{}
trace-005	span-002	span-001	risk_check	policy	evaluate_action	ok	21	null	null	null	null	null	null	null	{ "risk_level": "high", "action": "publish_change" }
trace-005	span-003	span-002	approval_request	human_oversight	request_approval	pending	7	null	null	null	null	null	null	null	{ "approval_id": "approval-005" }
trace-005	span-004	span-003	human_approval	human_oversight	approval_decision	approved	4,180	null	null	null	null	null	null	null	{ "approval_id": "approval-005", "decision": "approved" }
trace-005	span-005	span-004	tool_call	tool_runtime	publish_change	ok	142	null	publisher	null	null	null	null	null	{}
trace-005	span-006	span-005	audit_event	governance	record_action	ok	12	null	null	null	null	null	null	null	{}
trace-006	span-001	null	request	gateway	user_request	ok	3	null	null	null	null	null	null	null	{}
trace-006	span-002	span-001	model_call	llm	structured_generate	ok	548	model-c	null	721	156	0.0112	null	null	{}
trace-006	span-003	span-002	validation	validator	schema_check	fail	8	null	null	null	null	null	fail	missing required field	{}
trace-006	span-004	span-003	retry	orchestrator	regenerate	ok	19	null	null	null	null	null	null	null	{ "reason": "schema_validation_failed" }
trace-006	span-005	span-004	model_call	llm	structured_generate	ok	501	model-c	null	809	171	0.0124	null	null	{}
trace-006	span-006	span-005	validation	validator	schema_check	pass	8	null	null	null	null	null	pass	null	{}
trace-006	span-007	span-001	response	gateway	return_response	ok	3	null	null	null	null	null	null	null	{}
trace-007	span-001	null	request	supervisor	task_received	ok	5	null	null	null	null	null	null	null	{}
trace-007	span-002	span-001	delegation	supervisor	delegate_research	ok	12	null	null	null	null	null	null	null	{ "agent_id": "research-agent" }
trace-007	span-003	span-001	delegation	supervisor	delegate_analysis	ok	10	null	null	null	null	null	null	null	{ "agent_id": "analysis-agent" }
trace-007	span-004	span-002	agent_step	research-agent	gather_sources	ok	410	agent-model-r	null	null	null	null	null	null	{}
trace-007	span-005	span-003	agent_step	analysis-agent	analyze_inputs	ok	515	agent-model-a	null	null	null	null	null	null	{}
trace-007	span-006	span-001	merge	supervisor	merge_results	ok	81	null	null	null	null	null	null	null	{}
trace-007	span-007	span-006	verification	verifier	cross_check	pass	136	null	null	null	null	null	pass	null	{}
trace-007	span-008	span-001	response	supervisor	complete_task	ok	4	null	null	null	null	null	null	null	{}
trace-008	span-001	null	request	agent_runtime	task_received	ok	4	null	null	null	null	null	null	null	{ "budget_usd": 0.05 }
trace-008	span-002	span-001	router	router	select_model	ok	15	null	null	null	null	null	null	null	{ "selected_model": "economy-model", "reason": "budget_constraint" }
trace-008	span-003	span-002	model_call	llm	generate	ok	392	economy-model	null	610	143	0.0068	null	null	{}
trace-008	span-004	span-003	cost_check	cost_control	budget_check	ok	6	null	null	null	null	null	null	null	{ "running_cost_usd": 0.0068000000000000005, "budget_usd": 0.05 }
trace-008	span-005	span-001	response	agent_runtime	complete_task	ok	3	null	null	null	null	null	null	null	{}
trace-009	span-001	null	request	agent_runtime	task_received	ok	4	null	null	null	null	null	null	null	{}
trace-009	span-002	span-001	tool_selection	agent_runtime	select_tool	ok	18	null	database-write	null	null	null	null	null	{}
trace-009	span-003	span-002	permission_check	policy	authorize_tool	denied	9	null	database-write	null	null	null	null	null	{ "required_scope": "write:database", "granted_scope": "read:database" }
trace-009	span-004	span-003	human_escalation	human_oversight	request_help	ok	24	null	null	null	null	null	null	null	{ "reason": "insufficient_permission" }
trace-009	span-005	span-001	response	agent_runtime	stop_safely	ok	3	null	null	null	null	null	null	null	{}
trace-010	span-001	null	request	inference_gateway	enqueue	ok	2	null	null	null	null	null	null	null	{}
trace-010	span-002	span-001	queue	inference_gateway	queue_wait	ok	41	null	null	null	null	null	null	null	{}
trace-010	span-003	span-001	inference	inference_server	decode	ok	692	model-d	null	1,200	320	0.028	null	null	{ "batch_size": 8, "cache_hit": true, "tokens_per_second": 46.2 }
trace-010	span-004	span-003	stream	inference_gateway	stream_response	ok	118	null	null	null	null	null	null	null	{}
trace-010	span-005	span-001	response	inference_gateway	complete	ok	2	null	null	null	null	null	null	null	{}
trace-011	span-001	null	request	agent_runtime	long_horizon_task	ok	5	null	null	null	null	null	null	null	{}
trace-011	span-002	span-001	planning	agent_runtime	create_plan	ok	104	null	null	null	null	null	null	null	{}
trace-011	span-003	span-002	agent_step	agent_runtime	step_1	ok	220	null	null	null	null	null	null	null	{}
trace-011	span-004	span-002	agent_step	agent_runtime	step_2	ok	245	null	null	null	null	null	null	null	{}
trace-011	span-005	span-004	drift_detection	monitor	goal_alignment_check	warning	17	null	null	null	null	null	null	null	{ "goal_alignment_score": 0.54, "threshold": 0.7000000000000001 }
trace-011	span-006	span-005	replan	agent_runtime	replan_from_goal	ok	88	null	null	null	null	null	null	null	{}
trace-011	span-007	span-006	verification	verifier	goal_alignment_check	pass	39	null	null	null	null	null	pass	null	{}
trace-011	span-008	span-001	response	agent_runtime	continue_execution	ok	3	null	null	null	null	null	null	null	{}
trace-012	span-001	null	request	gateway	user_request	ok	3	null	null	null	null	null	null	null	{}
trace-012	span-002	span-001	redaction	privacy	redact_sensitive_fields	ok	11	null	null	null	null	null	null	null	{ "fields_redacted": 2 }
trace-012	span-003	span-002	model_call	llm	generate	ok	460	model-e	null	532	118	0.0081	null	null	{}
trace-012	span-004	span-001	logging	telemetry	store_trace	ok	7	null	null	null	null	null	null	null	{ "sensitive_content_stored": false }
trace-012	span-005	span-001	response	gateway	return_response	ok	3	null	null	null	null	null	null	null	{}

AI Observability Events

AI Observability Events is a small synthetic reference dataset for AI observability, tracing and telemetry.

It contains structured example events representing execution across modern AI systems such as:

LLM calls
retrieval and reranking
AI agents
tool calls
retries and fallbacks
memory access
verification
validation
human approval
permissions
cost controls
inference performance
goal drift
privacy redaction

The dataset is published by the Observability Hugging Face organization as an educational and prototyping resource.

Why This Dataset Exists

AI observability systems often need a common mental model for runtime telemetry.

A single AI request may involve:

User Request
   ↓
Router
   ↓
Model
   ↓
Retriever
   ↓
Tool
   ↓
Agent Step
   ↓
Memory
   ↓
Verifier
   ↓
Final Response

This dataset provides compact, inspectable event examples for those kinds of workflows.

Dataset Structure

Each row represents one observability event or span.

Core fields:

Field	Description
`trace_id`	End-to-end execution identifier
`span_id`	Identifier for the current span/event
`parent_span_id`	Parent span relationship
`event_type`	Type of telemetry event
`component`	AI system component
`operation`	Operation being performed
`status`	Runtime status
`duration_ms`	Duration in milliseconds
`model`	Model identifier when applicable
`tool`	Tool identifier when applicable
`input_tokens`	Input token count when applicable
`output_tokens`	Output token count when applicable
`cost_usd`	Example execution cost
`verification`	Verification result when applicable
`message`	Optional event/error message
`metadata`	Additional structured observability data

Example Record

{
  "trace_id": "trace-003",
  "span_id": "span-003",
  "parent_span_id": "span-002",
  "event_type": "tool_call",
  "component": "tool_runtime",
  "operation": "web_tool",
  "status": "error",
  "duration_ms": 180,
  "model": null,
  "tool": "web-search",
  "input_tokens": null,
  "output_tokens": null,
  "cost_usd": null,
  "verification": null,
  "message": "upstream timeout",
  "metadata": {
    "attempt": 1
  }
}

Included Trace Patterns

The dataset currently includes example traces for:

Simple LLM generation
Retrieval-Augmented Generation
Tool retry and fallback
Memory-aware agent execution
Human approval
Validation failure and retry
Multi-agent execution
Cost-aware routing
Permission denial
Inference performance
Goal-drift detection
Privacy redaction

Example Event Types

Included event types include:

request
model_call
retrieval
rerank
tool_call
retry
fallback
memory_read
memory_write
checkpoint
verification
validation
evaluation
planning
agent_step
delegation
merge
human_approval
permission_check
cost_check
inference
drift_detection
redaction
response

Intended Uses

This dataset can be used for:

observability demos
trace visualization prototypes
telemetry schema experiments
educational examples
event classification experiments
dashboard prototyping
debugging tutorials
agent observability research prototypes
trace correlation examples

Not a Benchmark

This dataset is not a benchmark.

It does not measure model quality, agent capability, safety or production reliability.

The values are illustrative examples.

Synthetic Data

All records in this dataset are synthetic.

They do not represent real users, real customer interactions, real production incidents or confidential telemetry.

Model names, provider names, costs, timings and identifiers are illustrative.

Limitations

This is intentionally a small reference dataset.

It does not cover:

every observability schema
every model provider
every agent framework
every failure mode
all OpenTelemetry semantic conventions
production-scale telemetry volume

The schema is designed for clarity rather than standardization.

Observability Concepts

The dataset is organized around these core concepts:

AI Observability
  USES → Traces
  USES → Spans
  USES → Logs
  USES → Metrics
  USES → Events
  OBSERVES → Models
  OBSERVES → Agents
  OBSERVES → Tools
  OBSERVES → Retrieval
  OBSERVES → Memory
  TRACKS → Cost
  TRACKS → Latency
  SUPPORTS → Evaluation
  SUPPORTS → Verification
  SUPPORTS → Reliability

Related Spaces

observability/observability-explorer
observability/agent-observability
observability/ai-trace-explorer

Collaboration & Partnerships

The Observability organization is open to collaboration with companies, research teams, universities and open-source projects working on:

AI observability
LLM observability
agent observability
tracing
telemetry
evaluation
verification
inference
production AI infrastructure

Possible collaboration formats include:

joint datasets
benchmark extensions
schema experiments
observability demos
trace visualization tools
framework integrations
clearly disclosed partnerships and sponsorships

Collaboration Contact

agenten@magenta.de

Independence

This dataset is an independent community resource.

It is not an official dataset of Hugging Face, OpenTelemetry, any AI laboratory, observability vendor, model provider or agent framework.

License

Apache-2.0

Downloads last month: 50

Number of rows:

Total file size:

34.9 kB