Datasets:
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: ValueError
Message: Expected object or value
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1816, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 281, in _generate_tables
examples = [ujson_loads(line) for line in batch.splitlines()]
~~~~~~~~~~~^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Expected object or value
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
trajectory_id string | path string | seed_input_id string | task_type string | difficulty float64 | language string | system_prompt string | user_input dict | trajectory list | meta dict |
|---|---|---|---|---|---|---|---|---|---|
incident_t_000000 | A | dns_failure_cart_prod_0 | dns_failure | 0.43 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "cart",
"namespace": "prod",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "cart",
"namespace": "prod",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV3"
},
{
"metric": "dns_error_rate_pct",
... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CoreDNSResolutionErrors firing on cart in prod. intermittent DNS resolution failures cluster-wide. Let me check the dns_error_rate_pct before acting.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "cart",
"nam... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000001 | A | oom_kill_orders_edge_1 | oom_kill | 0.62 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "orders",
"namespace": "edge",
"incident_context": "container memory exceeded limit; pod OOMKilled and restarting",
"system_state": [
{
"service": "orders",
"namespace": "edge",
"alert": "PodOOMKilled",
"severity": "SEV2"
},
{
"metric": "memory_util_pct"... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert PodOOMKilled firing on orders in edge. container memory exceeded limit; pod OOMKilled and restarting. Let me check the memory_util_pct before acting.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "orders",
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "oom_kill",
"alert": "PodOOMKilled",
"tools_used": [
... |
incident_t_000002 | A | crashloop_shipping_prod_2 | crashloop | 0.57 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "shipping",
"namespace": "prod",
"incident_context": "pod crash-looping after a bad config/env change",
"system_state": [
{
"service": "shipping",
"namespace": "prod",
"alert": "CrashLoopBackOff",
"severity": "SEV2"
},
{
"metric": "ready_replicas_pct",
... | [
{
"step": 1,
"role": "assistant",
"thought": "CrashLoopBackOff on shipping/prod. I should verify ready_replicas_pct rather than assume the cause.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "shipping",
"namespace": "prod"
}
},
"label": {
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "crashloop",
"alert": "CrashLoopBackOff",
"tools_used"... |
incident_t_000003 | A | bad_deploy_errors_auth_prod_3 | bad_deploy_errors | 0.45 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "prod",
"incident_context": "5xx error rate spiked right after a deploy",
"system_state": [
{
"service": "auth",
"namespace": "prod",
"alert": "ErrorRateSLOburn",
"severity": "SEV3"
},
{
"metric": "error_rate_pct",
"value": 30,... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert ErrorRateSLOburn firing on auth in prod. 5xx error rate spiked right after a deploy. Let me check the error_rate_pct before acting.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "prod"... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "bad_deploy_errors",
"alert": "ErrorRateSLOburn",
"too... |
incident_t_000004 | A | cache_stampede_auth_platform_4 | cache_stampede | 0.4 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "platform",
"incident_context": "cache hit ratio collapsed, origin overloaded",
"system_state": [
{
"service": "auth",
"namespace": "platform",
"alert": "CacheHitRatioDrop",
"severity": "SEV3"
},
{
"metric": "cache_hit_pct",
"v... | [
{
"step": 1,
"role": "assistant",
"thought": "auth is paging: cache hit ratio collapsed, origin overloaded. First confirm scope by reading get_metrics.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "platform"
}
},
"label":... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cache_stampede",
"alert": "CacheHitRatioDrop",
"tools... |
incident_t_000005 | A | cpu_saturation_payments_prod-us-east_5 | cpu_saturation | 0.43 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "payments",
"namespace": "prod-us-east",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "payments",
"namespace": "prod-us-east",
"alert": "CPUThrottlingHigh",
"severity": "SEV2"
},
{
"metric": "cpu_throttle_pct",
... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CPUThrottlingHigh firing on payments in prod-us-east. CPU throttled, requests queueing. Let me check the cpu_throttle_pct before acting.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "payments",
"namespace... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000006 | A | db_pool_exhaustion_search_prod-eu_6 | db_pool_exhaustion | 0.52 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "search",
"namespace": "prod-eu",
"incident_context": "all DB connections in use, requests timing out",
"system_state": [
{
"service": "search",
"namespace": "prod-eu",
"alert": "DBConnectionPoolExhausted",
"severity": "SEV2"
},
{
"metric": "db_conn_wait... | [
{
"step": 1,
"role": "assistant",
"thought": "DBConnectionPoolExhausted on search/prod-eu. I should verify db_conn_wait_ms rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "search",
"namespace": "prod-eu"
}
},
"label... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "db_pool_exhaustion",
"alert": "DBConnectionPoolExhauste... |
incident_t_000007 | A | latency_spike_inventory_edge_7 | latency_spike | 0.57 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "edge",
"incident_context": "p99 latency above SLO under a traffic surge",
"system_state": [
{
"service": "inventory",
"namespace": "edge",
"alert": "HighP99Latency",
"severity": "SEV1"
},
{
"metric": "p99_latency_ms",
"va... | [
{
"step": 1,
"role": "assistant",
"thought": "HighP99Latency on inventory/edge. I should verify p99_latency_ms rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "inventory",
"namespace": "edge"
}
},
"label": {
"... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "latency_spike",
"alert": "HighP99Latency",
"tools_use... |
incident_t_000008 | A | cache_stampede_gateway_edge_8 | cache_stampede | 0.4 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "gateway",
"namespace": "edge",
"incident_context": "cache hit ratio collapsed, origin overloaded",
"system_state": [
{
"service": "gateway",
"namespace": "edge",
"alert": "CacheHitRatioDrop",
"severity": "SEV1"
},
{
"metric": "cache_hit_pct",
"val... | [
{
"step": 1,
"role": "assistant",
"thought": "gateway is paging: cache hit ratio collapsed, origin overloaded. First confirm scope by reading get_metrics.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "gateway",
"namespace": "edge"
}
},
"label... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cache_stampede",
"alert": "CacheHitRatioDrop",
"tools... |
incident_t_000009 | A | dns_failure_user-profile_edge_9 | dns_failure | 0.33 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "user-profile",
"namespace": "edge",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "user-profile",
"namespace": "edge",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV3"
},
{
"metric": "dns_e... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CoreDNSResolutionErrors firing on user-profile in edge. intermittent DNS resolution failures cluster-wide. Let me check the dns_error_rate_pct before acting.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "user-profil... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000010 | A | consumer_lag_auth_prod_10 | consumer_lag | 0.79 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "prod",
"incident_context": "consumer group lag growing, backlog not draining",
"system_state": [
{
"service": "auth",
"namespace": "prod",
"alert": "KafkaConsumerLagHigh",
"severity": "SEV3"
},
{
"metric": "consumer_lag_k",
"v... | [
{
"step": 1,
"role": "assistant",
"thought": "auth is paging: consumer group lag growing, backlog not draining. First confirm scope by reading get_metrics.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "prod"
}
},
"label":... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "consumer_lag",
"alert": "KafkaConsumerLagHigh",
"tool... |
incident_t_000011 | A | cpu_saturation_orders_edge_11 | cpu_saturation | 0.45 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "orders",
"namespace": "edge",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "orders",
"namespace": "edge",
"alert": "CPUThrottlingHigh",
"severity": "SEV1"
},
{
"metric": "cpu_throttle_pct",
"value": 85,
... | [
{
"step": 1,
"role": "assistant",
"thought": "CPUThrottlingHigh on orders/edge. I should verify cpu_throttle_pct rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "orders",
"namespace": "edge"
}
},
"label": {
"a... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000012 | A | upstream_5xx_payments_prod-us-east_12 | upstream_5xx | 0.35 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "payments",
"namespace": "prod-us-east",
"incident_context": "a dependency is returning 5xx, cascading failures",
"system_state": [
{
"service": "payments",
"namespace": "prod-us-east",
"alert": "UpstreamDependencyDown",
"severity": "SEV2"
},
{
"metric":... | [
{
"step": 1,
"role": "assistant",
"thought": "payments is paging: a dependency is returning 5xx, cascading failures. First confirm scope by reading query_traces.",
"action": {
"tool": "query_traces",
"args": {
"service_id": "payments",
"namespace": "prod-us-east"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "upstream_5xx",
"alert": "UpstreamDependencyDown",
"to... |
incident_t_000013 | A | db_pool_exhaustion_catalog_prod-us-east_13 | db_pool_exhaustion | 0.38 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "catalog",
"namespace": "prod-us-east",
"incident_context": "all DB connections in use, requests timing out",
"system_state": [
{
"service": "catalog",
"namespace": "prod-us-east",
"alert": "DBConnectionPoolExhausted",
"severity": "SEV1"
},
{
"metric": "... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like all DB connections in use, requests timing out on catalog. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "catalog",
"namespace": "prod-us-east"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "db_pool_exhaustion",
"alert": "DBConnectionPoolExhauste... |
incident_t_000014 | A | cert_expiry_cart_prod-us-east_14 | cert_expiry | 0.53 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "cart",
"namespace": "prod-us-east",
"incident_context": "TLS handshake failures from an expired certificate",
"system_state": [
{
"service": "cart",
"namespace": "prod-us-east",
"alert": "TLSCertExpiringSoon",
"severity": "SEV3"
},
{
"metric": "tls_hand... | [
{
"step": 1,
"role": "assistant",
"thought": "TLSCertExpiringSoon on cart/prod-us-east. I should verify tls_handshake_success_pct rather than assume the cause.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "cart",
"namespace": "prod-us-east"
}
},
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cert_expiry",
"alert": "TLSCertExpiringSoon",
"tools_... |
incident_t_000015 | A | node_not_ready_checkout_prod_15 | node_not_ready | 0.68 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "checkout",
"namespace": "prod",
"incident_context": "a node went NotReady (kubelet/network), pods stuck",
"system_state": [
{
"service": "checkout",
"namespace": "prod",
"alert": "NodeNotReady",
"severity": "SEV1"
},
{
"metric": "ready_nodes_pct",
... | [
{
"step": 1,
"role": "assistant",
"thought": "NodeNotReady on checkout/prod. I should verify ready_nodes_pct rather than assume the cause.",
"action": {
"tool": "get_node_status",
"args": {
"service_id": "checkout",
"namespace": "prod"
}
},
"label": {
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "node_not_ready",
"alert": "NodeNotReady",
"tools_used... |
incident_t_000016 | A | db_pool_exhaustion_user-profile_prod-apac_16 | db_pool_exhaustion | 0.7 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "user-profile",
"namespace": "prod-apac",
"incident_context": "all DB connections in use, requests timing out",
"system_state": [
{
"service": "user-profile",
"namespace": "prod-apac",
"alert": "DBConnectionPoolExhausted",
"severity": "SEV2"
},
{
"metric... | [
{
"step": 1,
"role": "assistant",
"thought": "user-profile is paging: all DB connections in use, requests timing out. First confirm scope by reading get_metrics.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "user-profile",
"namespace": "prod-apac"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "db_pool_exhaustion",
"alert": "DBConnectionPoolExhauste... |
incident_t_000017 | A | cache_stampede_inventory_prod_17 | cache_stampede | 0.2 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "prod",
"incident_context": "cache hit ratio collapsed, origin overloaded",
"system_state": [
{
"service": "inventory",
"namespace": "prod",
"alert": "CacheHitRatioDrop",
"severity": "SEV2"
},
{
"metric": "cache_hit_pct",
... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CacheHitRatioDrop firing on inventory in prod. cache hit ratio collapsed, origin overloaded. Let me check the cache_hit_pct before acting.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "inventory",
"namesp... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cache_stampede",
"alert": "CacheHitRatioDrop",
"tools... |
incident_t_000018 | A | dns_failure_recommendations_prod-eu_18 | dns_failure | 0.4 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "recommendations",
"namespace": "prod-eu",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "recommendations",
"namespace": "prod-eu",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV3"
},
{
"met... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like intermittent DNS resolution failures cluster-wide on recommendations. Don't act blind — pull get_logs to confirm.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "recommendations",
"namespace": "prod-eu"
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000019 | A | consumer_lag_cart_prod_19 | consumer_lag | 0.52 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "cart",
"namespace": "prod",
"incident_context": "consumer group lag growing, backlog not draining",
"system_state": [
{
"service": "cart",
"namespace": "prod",
"alert": "KafkaConsumerLagHigh",
"severity": "SEV3"
},
{
"metric": "consumer_lag_k",
"v... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like consumer group lag growing, backlog not draining on cart. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "cart",
"namespace": "prod"
}
},
"l... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "consumer_lag",
"alert": "KafkaConsumerLagHigh",
"tool... |
incident_t_000020 | A | cpu_saturation_checkout_edge_20 | cpu_saturation | 0.49 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "checkout",
"namespace": "edge",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "checkout",
"namespace": "edge",
"alert": "CPUThrottlingHigh",
"severity": "SEV2"
},
{
"metric": "cpu_throttle_pct",
"value": 66... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CPUThrottlingHigh firing on checkout in edge. CPU throttled, requests queueing. Let me check the cpu_throttle_pct before acting.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "checkout",
"namespace": "edge... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000021 | A | dns_failure_gateway_platform_21 | dns_failure | 0.57 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "gateway",
"namespace": "platform",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "gateway",
"namespace": "platform",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV2"
},
{
"metric": "dns_err... | [
{
"step": 1,
"role": "assistant",
"thought": "CoreDNSResolutionErrors on gateway/platform. I should verify dns_error_rate_pct rather than assume the cause.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "gateway",
"namespace": "platform"
}
},
"lab... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000022 | A | upstream_5xx_catalog_platform_22 | upstream_5xx | 0.44 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "catalog",
"namespace": "platform",
"incident_context": "a dependency is returning 5xx, cascading failures",
"system_state": [
{
"service": "catalog",
"namespace": "platform",
"alert": "UpstreamDependencyDown",
"severity": "SEV3"
},
{
"metric": "upstream... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert UpstreamDependencyDown firing on catalog in platform. a dependency is returning 5xx, cascading failures. Let me check the upstream_success_pct before acting.",
"action": {
"tool": "query_traces",
"args": {
"service_id": "catalog... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "upstream_5xx",
"alert": "UpstreamDependencyDown",
"to... |
incident_t_000023 | A | stuck_rollout_payments_prod-eu_23 | stuck_rollout | 0.4 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "payments",
"namespace": "prod-eu",
"incident_context": "deployment stuck — readiness probe never passes",
"system_state": [
{
"service": "payments",
"namespace": "prod-eu",
"alert": "RolloutStuck",
"severity": "SEV2"
},
{
"metric": "available_replicas_p... | [
{
"step": 1,
"role": "assistant",
"thought": "RolloutStuck on payments/prod-eu. I should verify available_replicas_pct rather than assume the cause.",
"action": {
"tool": "get_deployment_status",
"args": {
"service_id": "payments",
"namespace": "prod-eu"
}
},
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "stuck_rollout",
"alert": "RolloutStuck",
"tools_used"... |
incident_t_000024 | A | upstream_5xx_recommendations_prod-apac_24 | upstream_5xx | 0.53 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "recommendations",
"namespace": "prod-apac",
"incident_context": "a dependency is returning 5xx, cascading failures",
"system_state": [
{
"service": "recommendations",
"namespace": "prod-apac",
"alert": "UpstreamDependencyDown",
"severity": "SEV2"
},
{
"... | [
{
"step": 1,
"role": "assistant",
"thought": "UpstreamDependencyDown on recommendations/prod-apac. I should verify upstream_success_pct rather than assume the cause.",
"action": {
"tool": "query_traces",
"args": {
"service_id": "recommendations",
"namespace": "prod-apac"
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "upstream_5xx",
"alert": "UpstreamDependencyDown",
"to... |
incident_t_000025 | A | memory_leak_recommendations_prod-eu_25 | memory_leak | 0.42 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "recommendations",
"namespace": "prod-eu",
"incident_context": "memory creeping up over hours, GC not reclaiming",
"system_state": [
{
"service": "recommendations",
"namespace": "prod-eu",
"alert": "MemoryLeakSuspected",
"severity": "SEV1"
},
{
"metric":... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like memory creeping up over hours, GC not reclaiming on recommendations. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "recommendations",
"namespace": "prod-... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "memory_leak",
"alert": "MemoryLeakSuspected",
"tools_... |
incident_t_000026 | A | cert_expiry_recommendations_prod-apac_26 | cert_expiry | 0.34 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "recommendations",
"namespace": "prod-apac",
"incident_context": "TLS handshake failures from an expired certificate",
"system_state": [
{
"service": "recommendations",
"namespace": "prod-apac",
"alert": "TLSCertExpiringSoon",
"severity": "SEV2"
},
{
"me... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert TLSCertExpiringSoon firing on recommendations in prod-apac. TLS handshake failures from an expired certificate. Let me check the tls_handshake_success_pct before acting.",
"action": {
"tool": "get_logs",
"args": {
"service_id": ... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cert_expiry",
"alert": "TLSCertExpiringSoon",
"tools_... |
incident_t_000027 | A | memory_leak_auth_edge_27 | memory_leak | 0.69 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "edge",
"incident_context": "memory creeping up over hours, GC not reclaiming",
"system_state": [
{
"service": "auth",
"namespace": "edge",
"alert": "MemoryLeakSuspected",
"severity": "SEV2"
},
{
"metric": "memory_util_pct",
"v... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like memory creeping up over hours, GC not reclaiming on auth. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "edge"
}
},
"l... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "memory_leak",
"alert": "MemoryLeakSuspected",
"tools_... |
incident_t_000028 | A | dns_failure_inventory_platform_28 | dns_failure | 0.43 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "platform",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "inventory",
"namespace": "platform",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV2"
},
{
"metric": "dns... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like intermittent DNS resolution failures cluster-wide on inventory. Don't act blind — pull get_logs to confirm.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "inventory",
"namespace": "platform"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000029 | A | cpu_saturation_inventory_edge_29 | cpu_saturation | 0.35 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "edge",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "inventory",
"namespace": "edge",
"alert": "CPUThrottlingHigh",
"severity": "SEV2"
},
{
"metric": "cpu_throttle_pct",
"value": ... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CPUThrottlingHigh firing on inventory in edge. CPU throttled, requests queueing. Let me check the cpu_throttle_pct before acting.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "inventory",
"namespace": "ed... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000030 | A | db_pool_exhaustion_auth_prod-apac_30 | db_pool_exhaustion | 0.59 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "prod-apac",
"incident_context": "all DB connections in use, requests timing out",
"system_state": [
{
"service": "auth",
"namespace": "prod-apac",
"alert": "DBConnectionPoolExhausted",
"severity": "SEV1"
},
{
"metric": "db_conn_wait... | [
{
"step": 1,
"role": "assistant",
"thought": "DBConnectionPoolExhausted on auth/prod-apac. I should verify db_conn_wait_ms rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "prod-apac"
}
},
"label... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "db_pool_exhaustion",
"alert": "DBConnectionPoolExhauste... |
incident_t_000031 | A | cpu_saturation_sessions_prod_31 | cpu_saturation | 0.49 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "sessions",
"namespace": "prod",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "sessions",
"namespace": "prod",
"alert": "CPUThrottlingHigh",
"severity": "SEV3"
},
{
"metric": "cpu_throttle_pct",
"value": 67... | [
{
"step": 1,
"role": "assistant",
"thought": "CPUThrottlingHigh on sessions/prod. I should verify cpu_throttle_pct rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "sessions",
"namespace": "prod"
}
},
"label": {
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000032 | A | cache_stampede_sessions_prod-us-east_32 | cache_stampede | 0.53 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "sessions",
"namespace": "prod-us-east",
"incident_context": "cache hit ratio collapsed, origin overloaded",
"system_state": [
{
"service": "sessions",
"namespace": "prod-us-east",
"alert": "CacheHitRatioDrop",
"severity": "SEV3"
},
{
"metric": "cache_hi... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like cache hit ratio collapsed, origin overloaded on sessions. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "sessions",
"namespace": "prod-us-east"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cache_stampede",
"alert": "CacheHitRatioDrop",
"tools... |
incident_t_000033 | A | cert_expiry_recommendations_platform_33 | cert_expiry | 0.31 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "recommendations",
"namespace": "platform",
"incident_context": "TLS handshake failures from an expired certificate",
"system_state": [
{
"service": "recommendations",
"namespace": "platform",
"alert": "TLSCertExpiringSoon",
"severity": "SEV3"
},
{
"metr... | [
{
"step": 1,
"role": "assistant",
"thought": "TLSCertExpiringSoon on recommendations/platform. I should verify tls_handshake_success_pct rather than assume the cause.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "recommendations",
"namespace": "platform"
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cert_expiry",
"alert": "TLSCertExpiringSoon",
"tools_... |
incident_t_000034 | A | stuck_rollout_catalog_prod-apac_34 | stuck_rollout | 0.37 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "catalog",
"namespace": "prod-apac",
"incident_context": "deployment stuck — readiness probe never passes",
"system_state": [
{
"service": "catalog",
"namespace": "prod-apac",
"alert": "RolloutStuck",
"severity": "SEV1"
},
{
"metric": "available_replicas... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert RolloutStuck firing on catalog in prod-apac. deployment stuck — readiness probe never passes. Let me check the available_replicas_pct before acting.",
"action": {
"tool": "get_deployment_status",
"args": {
"service_id": "catalog... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "stuck_rollout",
"alert": "RolloutStuck",
"tools_used"... |
incident_t_000035 | A | bad_deploy_errors_media-upload_prod-us-east_35 | bad_deploy_errors | 0.65 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "media-upload",
"namespace": "prod-us-east",
"incident_context": "5xx error rate spiked right after a deploy",
"system_state": [
{
"service": "media-upload",
"namespace": "prod-us-east",
"alert": "ErrorRateSLOburn",
"severity": "SEV2"
},
{
"metric": "err... | [
{
"step": 1,
"role": "assistant",
"thought": "ErrorRateSLOburn on media-upload/prod-us-east. I should verify error_rate_pct rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "media-upload",
"namespace": "prod-us-east"
}
}... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "bad_deploy_errors",
"alert": "ErrorRateSLOburn",
"too... |
incident_t_000036 | A | node_not_ready_checkout_prod-apac_36 | node_not_ready | 0.45 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "checkout",
"namespace": "prod-apac",
"incident_context": "a node went NotReady (kubelet/network), pods stuck",
"system_state": [
{
"service": "checkout",
"namespace": "prod-apac",
"alert": "NodeNotReady",
"severity": "SEV3"
},
{
"metric": "ready_nodes_p... | [
{
"step": 1,
"role": "assistant",
"thought": "NodeNotReady on checkout/prod-apac. I should verify ready_nodes_pct rather than assume the cause.",
"action": {
"tool": "get_node_status",
"args": {
"service_id": "checkout",
"namespace": "prod-apac"
}
},
"label"... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "node_not_ready",
"alert": "NodeNotReady",
"tools_used... |
incident_t_000037 | A | cpu_saturation_catalog_prod-eu_37 | cpu_saturation | 0.34 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "catalog",
"namespace": "prod-eu",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "catalog",
"namespace": "prod-eu",
"alert": "CPUThrottlingHigh",
"severity": "SEV2"
},
{
"metric": "cpu_throttle_pct",
"value"... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like CPU throttled, requests queueing on catalog. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "catalog",
"namespace": "prod-eu"
}
},
"label": ... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000038 | A | oom_kill_search_prod-us-east_38 | oom_kill | 0.51 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "search",
"namespace": "prod-us-east",
"incident_context": "container memory exceeded limit; pod OOMKilled and restarting",
"system_state": [
{
"service": "search",
"namespace": "prod-us-east",
"alert": "PodOOMKilled",
"severity": "SEV3"
},
{
"metric": "... | [
{
"step": 1,
"role": "assistant",
"thought": "PodOOMKilled on search/prod-us-east. I should verify memory_util_pct rather than assume the cause.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "search",
"namespace": "prod-us-east"
}
},
"label":... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "oom_kill",
"alert": "PodOOMKilled",
"tools_used": [
... |
incident_t_000039 | A | disk_pressure_cart_platform_39 | disk_pressure | 0.28 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "cart",
"namespace": "platform",
"incident_context": "node disk almost full, evictions imminent",
"system_state": [
{
"service": "cart",
"namespace": "platform",
"alert": "NodeDiskPressure",
"severity": "SEV1"
},
{
"metric": "disk_used_pct",
"value... | [
{
"step": 1,
"role": "assistant",
"thought": "NodeDiskPressure on cart/platform. I should verify disk_used_pct rather than assume the cause.",
"action": {
"tool": "get_node_status",
"args": {
"service_id": "cart",
"namespace": "platform"
}
},
"label": {
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "disk_pressure",
"alert": "NodeDiskPressure",
"tools_u... |
incident_t_000040 | A | stuck_rollout_cart_platform_40 | stuck_rollout | 0.32 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "cart",
"namespace": "platform",
"incident_context": "deployment stuck — readiness probe never passes",
"system_state": [
{
"service": "cart",
"namespace": "platform",
"alert": "RolloutStuck",
"severity": "SEV2"
},
{
"metric": "available_replicas_pct",
... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert RolloutStuck firing on cart in platform. deployment stuck — readiness probe never passes. Let me check the available_replicas_pct before acting.",
"action": {
"tool": "get_deployment_status",
"args": {
"service_id": "cart",
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "stuck_rollout",
"alert": "RolloutStuck",
"tools_used"... |
incident_t_000041 | A | cache_stampede_sessions_edge_41 | cache_stampede | 0.32 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "sessions",
"namespace": "edge",
"incident_context": "cache hit ratio collapsed, origin overloaded",
"system_state": [
{
"service": "sessions",
"namespace": "edge",
"alert": "CacheHitRatioDrop",
"severity": "SEV2"
},
{
"metric": "cache_hit_pct",
"v... | [
{
"step": 1,
"role": "assistant",
"thought": "CacheHitRatioDrop on sessions/edge. I should verify cache_hit_pct rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "sessions",
"namespace": "edge"
}
},
"label": {
"... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cache_stampede",
"alert": "CacheHitRatioDrop",
"tools... |
incident_t_000042 | A | cpu_saturation_sessions_platform_42 | cpu_saturation | 0.61 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "sessions",
"namespace": "platform",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "sessions",
"namespace": "platform",
"alert": "CPUThrottlingHigh",
"severity": "SEV1"
},
{
"metric": "cpu_throttle_pct",
"va... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like CPU throttled, requests queueing on sessions. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "sessions",
"namespace": "platform"
}
},
"label... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000043 | A | oom_kill_orders_prod_43 | oom_kill | 0.45 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "orders",
"namespace": "prod",
"incident_context": "container memory exceeded limit; pod OOMKilled and restarting",
"system_state": [
{
"service": "orders",
"namespace": "prod",
"alert": "PodOOMKilled",
"severity": "SEV3"
},
{
"metric": "memory_util_pct"... | [
{
"step": 1,
"role": "assistant",
"thought": "PodOOMKilled on orders/prod. I should verify memory_util_pct rather than assume the cause.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "orders",
"namespace": "prod"
}
},
"label": {
"action... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "oom_kill",
"alert": "PodOOMKilled",
"tools_used": [
... |
incident_t_000044 | A | dns_failure_billing_prod-apac_44 | dns_failure | 0.57 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "billing",
"namespace": "prod-apac",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "billing",
"namespace": "prod-apac",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV2"
},
{
"metric": "dns_e... | [
{
"step": 1,
"role": "assistant",
"thought": "CoreDNSResolutionErrors on billing/prod-apac. I should verify dns_error_rate_pct rather than assume the cause.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "billing",
"namespace": "prod-apac"
}
},
"l... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000045 | A | upstream_5xx_inventory_prod-us-east_45 | upstream_5xx | 0.43 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "prod-us-east",
"incident_context": "a dependency is returning 5xx, cascading failures",
"system_state": [
{
"service": "inventory",
"namespace": "prod-us-east",
"alert": "UpstreamDependencyDown",
"severity": "SEV2"
},
{
"metric... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like a dependency is returning 5xx, cascading failures on inventory. Don't act blind — pull query_traces to confirm.",
"action": {
"tool": "query_traces",
"args": {
"service_id": "inventory",
"namespace": "prod-us-east"
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "upstream_5xx",
"alert": "UpstreamDependencyDown",
"to... |
incident_t_000046 | A | oom_kill_payments_platform_46 | oom_kill | 0.5 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "payments",
"namespace": "platform",
"incident_context": "container memory exceeded limit; pod OOMKilled and restarting",
"system_state": [
{
"service": "payments",
"namespace": "platform",
"alert": "PodOOMKilled",
"severity": "SEV1"
},
{
"metric": "memo... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert PodOOMKilled firing on payments in platform. container memory exceeded limit; pod OOMKilled and restarting. Let me check the memory_util_pct before acting.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "payments"... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "oom_kill",
"alert": "PodOOMKilled",
"tools_used": [
... |
incident_t_000047 | A | stuck_rollout_media-upload_prod_47 | stuck_rollout | 0.38 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "media-upload",
"namespace": "prod",
"incident_context": "deployment stuck — readiness probe never passes",
"system_state": [
{
"service": "media-upload",
"namespace": "prod",
"alert": "RolloutStuck",
"severity": "SEV2"
},
{
"metric": "available_replicas... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert RolloutStuck firing on media-upload in prod. deployment stuck — readiness probe never passes. Let me check the available_replicas_pct before acting.",
"action": {
"tool": "get_deployment_status",
"args": {
"service_id": "media-u... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "stuck_rollout",
"alert": "RolloutStuck",
"tools_used"... |
incident_t_000048 | A | consumer_lag_notifications_platform_48 | consumer_lag | 0.6 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "notifications",
"namespace": "platform",
"incident_context": "consumer group lag growing, backlog not draining",
"system_state": [
{
"service": "notifications",
"namespace": "platform",
"alert": "KafkaConsumerLagHigh",
"severity": "SEV2"
},
{
"metric": ... | [
{
"step": 1,
"role": "assistant",
"thought": "KafkaConsumerLagHigh on notifications/platform. I should verify consumer_lag_k rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "notifications",
"namespace": "platform"
}
},
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "consumer_lag",
"alert": "KafkaConsumerLagHigh",
"tool... |
incident_t_000049 | A | crashloop_inventory_prod-us-east_49 | crashloop | 0.49 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "prod-us-east",
"incident_context": "pod crash-looping after a bad config/env change",
"system_state": [
{
"service": "inventory",
"namespace": "prod-us-east",
"alert": "CrashLoopBackOff",
"severity": "SEV3"
},
{
"metric": "read... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CrashLoopBackOff firing on inventory in prod-us-east. pod crash-looping after a bad config/env change. Let me check the ready_replicas_pct before acting.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "inventory",... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "crashloop",
"alert": "CrashLoopBackOff",
"tools_used"... |
incident_t_000050 | A | db_pool_exhaustion_auth_edge_50 | db_pool_exhaustion | 0.56 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "edge",
"incident_context": "all DB connections in use, requests timing out",
"system_state": [
{
"service": "auth",
"namespace": "edge",
"alert": "DBConnectionPoolExhausted",
"severity": "SEV3"
},
{
"metric": "db_conn_wait_ms",
... | [
{
"step": 1,
"role": "assistant",
"thought": "DBConnectionPoolExhausted on auth/edge. I should verify db_conn_wait_ms rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "edge"
}
},
"label": {
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "db_pool_exhaustion",
"alert": "DBConnectionPoolExhauste... |
incident_t_000051 | A | consumer_lag_user-profile_prod-apac_51 | consumer_lag | 0.45 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "user-profile",
"namespace": "prod-apac",
"incident_context": "consumer group lag growing, backlog not draining",
"system_state": [
{
"service": "user-profile",
"namespace": "prod-apac",
"alert": "KafkaConsumerLagHigh",
"severity": "SEV2"
},
{
"metric": ... | [
{
"step": 1,
"role": "assistant",
"thought": "KafkaConsumerLagHigh on user-profile/prod-apac. I should verify consumer_lag_k rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "user-profile",
"namespace": "prod-apac"
}
},
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "consumer_lag",
"alert": "KafkaConsumerLagHigh",
"tool... |
incident_t_000052 | A | consumer_lag_gateway_prod_52 | consumer_lag | 0.42 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "gateway",
"namespace": "prod",
"incident_context": "consumer group lag growing, backlog not draining",
"system_state": [
{
"service": "gateway",
"namespace": "prod",
"alert": "KafkaConsumerLagHigh",
"severity": "SEV2"
},
{
"metric": "consumer_lag_k",
... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like consumer group lag growing, backlog not draining on gateway. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "gateway",
"namespace": "prod"
}
},
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "consumer_lag",
"alert": "KafkaConsumerLagHigh",
"tool... |
incident_t_000053 | A | dns_failure_media-upload_prod_53 | dns_failure | 0.64 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "media-upload",
"namespace": "prod",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "media-upload",
"namespace": "prod",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV2"
},
{
"metric": "dns_e... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like intermittent DNS resolution failures cluster-wide on media-upload. Don't act blind — pull get_logs to confirm.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "media-upload",
"namespace": "prod"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
incident_t_000054 | A | node_not_ready_payments_prod-apac_54 | node_not_ready | 0.43 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "payments",
"namespace": "prod-apac",
"incident_context": "a node went NotReady (kubelet/network), pods stuck",
"system_state": [
{
"service": "payments",
"namespace": "prod-apac",
"alert": "NodeNotReady",
"severity": "SEV2"
},
{
"metric": "ready_nodes_p... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert NodeNotReady firing on payments in prod-apac. a node went NotReady (kubelet/network), pods stuck. Let me check the ready_nodes_pct before acting.",
"action": {
"tool": "get_node_status",
"args": {
"service_id": "payments",
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "node_not_ready",
"alert": "NodeNotReady",
"tools_used... |
incident_t_000055 | A | node_not_ready_billing_platform_55 | node_not_ready | 0.91 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "billing",
"namespace": "platform",
"incident_context": "a node went NotReady (kubelet/network), pods stuck",
"system_state": [
{
"service": "billing",
"namespace": "platform",
"alert": "NodeNotReady",
"severity": "SEV2"
},
{
"metric": "ready_nodes_pct",... | [
{
"step": 1,
"role": "assistant",
"thought": "billing is paging: a node went NotReady (kubelet/network), pods stuck. First confirm scope by reading get_node_status.",
"action": {
"tool": "get_node_status",
"args": {
"service_id": "billing",
"namespace": "platform"
}... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "node_not_ready",
"alert": "NodeNotReady",
"tools_used... |
incident_t_000056 | A | consumer_lag_notifications_edge_56 | consumer_lag | 0.49 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "notifications",
"namespace": "edge",
"incident_context": "consumer group lag growing, backlog not draining",
"system_state": [
{
"service": "notifications",
"namespace": "edge",
"alert": "KafkaConsumerLagHigh",
"severity": "SEV2"
},
{
"metric": "consume... | [
{
"step": 1,
"role": "assistant",
"thought": "KafkaConsumerLagHigh on notifications/edge. I should verify consumer_lag_k rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "notifications",
"namespace": "edge"
}
},
"lab... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "consumer_lag",
"alert": "KafkaConsumerLagHigh",
"tool... |
incident_t_000057 | A | node_not_ready_gateway_prod-apac_57 | node_not_ready | 0.89 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "gateway",
"namespace": "prod-apac",
"incident_context": "a node went NotReady (kubelet/network), pods stuck",
"system_state": [
{
"service": "gateway",
"namespace": "prod-apac",
"alert": "NodeNotReady",
"severity": "SEV1"
},
{
"metric": "ready_nodes_pct... | [
{
"step": 1,
"role": "assistant",
"thought": "gateway is paging: a node went NotReady (kubelet/network), pods stuck. First confirm scope by reading get_node_status.",
"action": {
"tool": "get_node_status",
"args": {
"service_id": "gateway",
"namespace": "prod-apac"
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "node_not_ready",
"alert": "NodeNotReady",
"tools_used... |
incident_t_000058 | A | cert_expiry_gateway_prod_58 | cert_expiry | 0.51 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "gateway",
"namespace": "prod",
"incident_context": "TLS handshake failures from an expired certificate",
"system_state": [
{
"service": "gateway",
"namespace": "prod",
"alert": "TLSCertExpiringSoon",
"severity": "SEV1"
},
{
"metric": "tls_handshake_succ... | [
{
"step": 1,
"role": "assistant",
"thought": "gateway is paging: TLS handshake failures from an expired certificate. First confirm scope by reading get_logs.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "gateway",
"namespace": "prod"
}
},
"label... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cert_expiry",
"alert": "TLSCertExpiringSoon",
"tools_... |
incident_t_000059 | A | bad_deploy_errors_notifications_platform_59 | bad_deploy_errors | 0.68 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "notifications",
"namespace": "platform",
"incident_context": "5xx error rate spiked right after a deploy",
"system_state": [
{
"service": "notifications",
"namespace": "platform",
"alert": "ErrorRateSLOburn",
"severity": "SEV2"
},
{
"metric": "error_rat... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like 5xx error rate spiked right after a deploy on notifications. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "notifications",
"namespace": "platform"
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "bad_deploy_errors",
"alert": "ErrorRateSLOburn",
"too... |
incident_t_000060 | A | latency_spike_media-upload_prod_60 | latency_spike | 0.36 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "media-upload",
"namespace": "prod",
"incident_context": "p99 latency above SLO under a traffic surge",
"system_state": [
{
"service": "media-upload",
"namespace": "prod",
"alert": "HighP99Latency",
"severity": "SEV2"
},
{
"metric": "p99_latency_ms",
... | [
{
"step": 1,
"role": "assistant",
"thought": "HighP99Latency on media-upload/prod. I should verify p99_latency_ms rather than assume the cause.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "media-upload",
"namespace": "prod"
}
},
"label": {
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "latency_spike",
"alert": "HighP99Latency",
"tools_use... |
incident_t_000061 | A | cpu_saturation_gateway_prod-eu_61 | cpu_saturation | 0.73 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "gateway",
"namespace": "prod-eu",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "gateway",
"namespace": "prod-eu",
"alert": "CPUThrottlingHigh",
"severity": "SEV1"
},
{
"metric": "cpu_throttle_pct",
"value"... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like CPU throttled, requests queueing on gateway. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "gateway",
"namespace": "prod-eu"
}
},
"label": ... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000062 | A | bad_deploy_errors_auth_prod-apac_62 | bad_deploy_errors | 0.35 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "auth",
"namespace": "prod-apac",
"incident_context": "5xx error rate spiked right after a deploy",
"system_state": [
{
"service": "auth",
"namespace": "prod-apac",
"alert": "ErrorRateSLOburn",
"severity": "SEV3"
},
{
"metric": "error_rate_pct",
"v... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like 5xx error rate spiked right after a deploy on auth. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "auth",
"namespace": "prod-apac"
}
},
"la... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "bad_deploy_errors",
"alert": "ErrorRateSLOburn",
"too... |
incident_t_000063 | A | memory_leak_inventory_prod-us-east_63 | memory_leak | 0.35 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "inventory",
"namespace": "prod-us-east",
"incident_context": "memory creeping up over hours, GC not reclaiming",
"system_state": [
{
"service": "inventory",
"namespace": "prod-us-east",
"alert": "MemoryLeakSuspected",
"severity": "SEV1"
},
{
"metric": "... | [
{
"step": 1,
"role": "assistant",
"thought": "inventory is paging: memory creeping up over hours, GC not reclaiming. First confirm scope by reading get_metrics.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "inventory",
"namespace": "prod-us-east"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "memory_leak",
"alert": "MemoryLeakSuspected",
"tools_... |
incident_t_000064 | A | cpu_saturation_sessions_prod-eu_64 | cpu_saturation | 0.52 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "sessions",
"namespace": "prod-eu",
"incident_context": "CPU throttled, requests queueing",
"system_state": [
{
"service": "sessions",
"namespace": "prod-eu",
"alert": "CPUThrottlingHigh",
"severity": "SEV2"
},
{
"metric": "cpu_throttle_pct",
"valu... | [
{
"step": 1,
"role": "assistant",
"thought": "Looks like CPU throttled, requests queueing on sessions. Don't act blind — pull get_metrics to confirm.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "sessions",
"namespace": "prod-eu"
}
},
"label"... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cpu_saturation",
"alert": "CPUThrottlingHigh",
"tools... |
incident_t_000065 | A | crashloop_notifications_prod-apac_65 | crashloop | 0.69 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "notifications",
"namespace": "prod-apac",
"incident_context": "pod crash-looping after a bad config/env change",
"system_state": [
{
"service": "notifications",
"namespace": "prod-apac",
"alert": "CrashLoopBackOff",
"severity": "SEV2"
},
{
"metric": "re... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CrashLoopBackOff firing on notifications in prod-apac. pod crash-looping after a bad config/env change. Let me check the ready_replicas_pct before acting.",
"action": {
"tool": "describe_pod",
"args": {
"service_id": "notificati... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "crashloop",
"alert": "CrashLoopBackOff",
"tools_used"... |
incident_t_000066 | A | cache_stampede_media-upload_prod-eu_66 | cache_stampede | 0.46 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "media-upload",
"namespace": "prod-eu",
"incident_context": "cache hit ratio collapsed, origin overloaded",
"system_state": [
{
"service": "media-upload",
"namespace": "prod-eu",
"alert": "CacheHitRatioDrop",
"severity": "SEV3"
},
{
"metric": "cache_hit_... | [
{
"step": 1,
"role": "assistant",
"thought": "media-upload is paging: cache hit ratio collapsed, origin overloaded. First confirm scope by reading get_metrics.",
"action": {
"tool": "get_metrics",
"args": {
"service_id": "media-upload",
"namespace": "prod-eu"
}
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "cache_stampede",
"alert": "CacheHitRatioDrop",
"tools... |
incident_t_000067 | A | dns_failure_payments_prod-eu_67 | dns_failure | 0.55 | en | You are a platform-engineering (SRE) agent on call. An incident is firing. Assess the system state by reading metrics/logs/events, then decide and execute remediation using the available tools. Respect tool guardrails (some actions need human approval). End with end_incident once the SLO is restored. | {
"service_id": "payments",
"namespace": "prod-eu",
"incident_context": "intermittent DNS resolution failures cluster-wide",
"system_state": [
{
"service": "payments",
"namespace": "prod-eu",
"alert": "CoreDNSResolutionErrors",
"severity": "SEV3"
},
{
"metric": "dns_err... | [
{
"step": 1,
"role": "assistant",
"thought": "Alert CoreDNSResolutionErrors firing on payments in prod-eu. intermittent DNS resolution failures cluster-wide. Let me check the dns_error_rate_pct before acting.",
"action": {
"tool": "get_logs",
"args": {
"service_id": "payments",
... | {
"construction_path": "synthetic_templated",
"usefulness_gain_per_step": [
1.4,
null,
1.4,
null,
2,
null,
0.3
],
"alternative_paths_count": 1,
"outcome_correct": true,
"outcome_match_to_seed": "exact",
"incident_type": "dns_failure",
"alert": "CoreDNSResolutionErrors",
"to... |
End of preview.
Infra-Ops Agent — incident-resolution trajectories
Training data for a platform-engineering / SRE agent that resolves production incidents as a verifiable state-transition task:
system state (metrics / logs / alerts) → agent picks a tool → new state → resolved?
Schema mirrors the FIREBALL D&D dataset (structured before/after state), so the same training pipeline applies to a real domain.
Code (generators, verifier, Unsloth SFT/DPO scripts): https://github.com/ashishranjan2404/infra-ops-agent
Files
| File | Role | Rows |
|---|---|---|
incidents.jsonl |
SFT — correct incident-resolution trajectories | 150,000 |
pairs.jsonl |
DPO — winner/loser preference pairs | 30,000 |
rejections.jsonl |
labeled failure trajectories (failure_mode) |
20,000 |
incidents_stats.json |
distribution summary | — |
Contents
- 15 incident types: oom_kill, crashloop, latency_spike, bad_deploy_errors, disk_pressure, cert_expiry, memory_leak, db_pool_exhaustion, node_not_ready, consumer_lag, dns_failure, upstream_5xx, cpu_saturation, stuck_rollout, cache_stampede.
- 25-tool registry with trust tiers (
autonomous/approval/blocked). - 7 failure modes across two families: outcome (caught by the verifiable reward) and process (still resolves — needs the DPO preference signal).
Verifiable reward
Resolution = the affected metric returns under its slo_threshold (deterministic,
path-independent). See verify.py in the code repo.
Note
Data is synthetic + templated (like FIREBALL's Path-A converter) and fully reproducible from the seeded scripts. No real/production data. License: MIT.
- Downloads last month
- 44