martin76ec/infrastructure-incident-synthetic

Infrastructure Incident Triage Dataset for TypeSafe AI Jev Model & DeBERTa Distillation

This repository provides a production-grade benchmark and evaluation dataset of structured payloads designed for Jev (POST https://api.typesafe.ai/v1/systemone), a "System One" decision model by TypeSafe AI.

---

1. Overview: The Jev Decision Model

Unlike generative LLMs that generate free-form text tokens sequentially, Jev is a non-autoregressive decision engine optimized for fast (70–500ms), low-cost, machine-native classifications, routing decisions, and safety guardrails.

Jev evaluates a provided context (state) against typed questions (questions) using three core primitives:

Primitive	Return Type	Description & Purpose
choice	Categorical Option + Confidence	Selects one option from a predefined set of criteria (e.g., routing target, root cause).
score	Ordered Float / Level	Places the situation on a calibrated multi-point scale (e.g., severity, risk, urgency).
noul	Boolean Probability (0.0 - 1.0)	Evaluates a truth proposition (e.g., "Is this an active outage?", "Safe to auto-remediate?").

---

2. Dataset Scenarios

The dataset covers eight critical infrastructure and site reliability engineering (SRE) incident archetypes:

File	Scenario	Key Questions / Decisions Evaluated
k8s_crashloop_oom.json	Kubernetes CrashLoopBackOff & OOMKilled Container exit code 137, memory limit exhaustion after recent commit.	• incident_severity (choice) • root_cause_classification (choice) • requires_immediate_human_oncall (noul) • safe_to_auto_increase_memory_limit (noul) • blast_radius_score (score)
database_pool_exhaustion.json	PostgreSQL Aurora Pool Saturation & Deadlock 1988/2000 connections, lock contention, HTTP 504 gateway timeouts.	• triage_escalation_target (choice) • immediate_automated_action (choice) • is_active_outage (noul) • safe_to_kill_idle_connections (noul) • urgency_score (score)
node_disk_pressure.json	Node Root Disk Saturation & Pod Evictions Root volume at 97.4%, Kubelet evicting pods including stateful Elasticsearch.	• recommended_remediation_action (choice) • primary_culprit_category (choice) • is_safe_to_cordon_and_drain (noul) • requires_immediate_node_replacement (noul) • data_loss_risk_score (score)
network_dns_failure.json	CoreDNS Saturation & VPC DNS Resolution Blackout 34.8% DNS failure rate, Route 53 resolver throttling, NodeLocal DNS missing.	• primary_failure_domain (choice) • routing_remediation_strategy (choice) • is_upstream_cloud_outage (noul) • should_initiate_sev1_incident (noul) • incident_severity_score (score)
ddos_credential_stuffing.json	Edge Ingress Surge & Credential Stuffing Attack 15.4x traffic spike on login endpoint, residential proxies, 89% 401s.	• threat_classification (choice) • recommended_mitigation_action (choice) • is_confirmed_malicious_traffic (noul) • requires_secops_engagement (noul) • threat_impact_score (score)
canary_deployment_regression.json	Canary Deployment Error Rate Spike & Rollback Canary at 20% weight exhibits 7.82% 5xx errors; NullPointerException detected.	• canary_rollout_decision (choice) • defect_origin_classification (choice) • should_automatically_rollback (noul) • requires_database_rollback (noul) • defect_confidence_score (score)
kafka_consumer_lag_storm.json	Kafka Consumer Lag & Rebalance Storm 1.28M record lag, slow batch processing causing heartbeat timeouts and rebalances.	• primary_failure_mechanism (choice) • recovery_action_plan (choice) • is_immediate_regulatory_sla_breach (noul) • is_data_loss_risk_acceptable (noul) • backlog_severity_score (score)
tls_cert_expiry.json	TLS Certificate Expiration & Renewal Failure Cert expires in 5.5 hours; Let's Encrypt HTTP-01 challenge blocked by ingress rule.	• recommended_emergency_action (choice) • renewal_blocker_classification (choice) • requires_emergency_oncall_page (noul) • is_ingress_misconfiguration_root_cause (noul) • outage_imminence_score (score)

---

3. Directory Layout

triage-dataset/
├── README.md
├── validate_payloads.py
└── payloads/
    ├── all_payloads.json
    ├── canary_deployment_regression.json
    ├── database_pool_exhaustion.json
    ├── ddos_credential_stuffing.json
    ├── k8s_crashloop_oom.json
    ├── kafka_consumer_lag_storm.json
    ├── network_dns_failure.json
    ├── node_disk_pressure.json
    └── tls_cert_expiry.json

---

4. Usage & Execution

Validating Payload Schemas

Run the included validation script to verify schema conformance:

python3 validate_payloads.py

Making an API Request (cURL)

curl -X POST https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d @payloads/k8s_crashloop_oom.json

Making an API Request (Python)

import json
import requests

with open("payloads/database_pool_exhaustion.json") as f:
    payload = json.load(f)

response = requests.post(
    "https://api.typesafe.ai/v1/systemone",
    headers={
        "Authorization": "Bearer YOUR_TYPESAFE_API_KEY",
        "Content-Type": "application/json"
    },
    json=payload
)

print(response.json())