---
license: apache-2.0
task_categories:
- time-series-forecasting
- tabular-regression
- tabular-classification
---

# AgentFuel Benchmarks

Datasets and query sets for the three benchmark settings from the paper [Generating Expressive and Customizable Evals for Timeseries Data Analysis Agents with AgentFuel](https://arxiv.org/abs/2603.12483).

All datasets and query sets were generated using AgentFuel's data generation and question-answer generation modules.

### E-commerce
Product analytics for an e-commerce website. Browsing sessions are generated using a state machine covering browsing flows, cart abandonment, and purchase flows.

- **Datasets**: `ecommerce_users_data.csv`, `ecommerce_sessions_data.csv`
- **Queries**: 12 stateless (`ecommerce_basic.csv`) + 12 stateful (`ecommerce_stateful.csv`)

### IoT
IoT device monitoring with three sensor exemplars: temperature, pressure, and humidity, each with its own operations state machine and device health metrics.

- **Dataset**: `iot_device_data.csv`
- **Queries**: 12 stateless (`iot_basic.csv`) + 12 stateful (`iot_stateful.csv`)

### Telecom
Telecommunications network telemetry across three related entities: cell sites, transport links, and core nodes. The `_with_inc_` dataset variants include an injected cascading incident: a transport link degrades (elevated packet loss, latency, jitter), cascading to connected cell sites (higher RRC failures, lower availability), with a modest effect on core nodes (reduced attached UEs, increased CPU load).

- **Datasets**: `cell_site_data.csv`, `transport_link_data.csv`, `core_node_data.csv` (and `_with_inc_` variants)
- **Queries**: 12 stateless (`telecom_basic.csv`) + 12 incident-specific (`telecom_incident.csv`)