Datasets:

kavirubc
/

AgentDisruptBench

@context dict	@type string	name string	description string	conformsTo string	url string	license string	version string	datePublished timestamp[s]	creator dict	keywords list	isLiveDataset bool	distribution list	recordSet list
{ "@language": "en", "@vocab": "https://schema.org/", "citeAs": "cr:citeAs", "column": "cr:column", "conformsTo": "dct:conformsTo", "cr": "http://mlcommons.org/croissant/", "data": { "@id": "cr:data", "@type": "@json" }, "dataBiases": "cr:dataBiases", "dataCollection": "cr:dataCollection", "dataType": { "@id": "cr:dataType", "@type": "@vocab" }, "dct": "http://purl.org/dc/terms/", "extract": "cr:extract", "field": "cr:field", "fileProperty": "cr:fileProperty", "fileSet": "cr:fileSet", "format": "cr:format", "includes": "cr:includes", "isLiveDataset": "cr:isLiveDataset", "jsonPath": "cr:jsonPath", "key": "cr:key", "md5": "cr:md5", "parentField": "cr:parentField", "path": "cr:path", "personalSensitiveInformation": "cr:personalSensitiveInformation", "recordSet": "cr:recordSet", "references": "cr:references", "regex": "cr:regex", "repeated": "cr:repeated", "replace": "cr:replace", "sc": "https://schema.org/", "separator": "cr:separator", "source": "cr:source", "subField": "cr:subField", "transform": "cr:transform" }	sc:Dataset	AgentDisruptBench	An evaluation methodology for evaluating AI agent resilience under runtime tool-call disruptions, designed for the NeurIPS Evaluations & Datasets Track. Contains 100 base tasks and variants across 4 domains (Retail, Travel, Finance, DevOps) with 20 disruption types and 9 disruption profiles.	http://mlcommons.org/croissant/1.0	https://github.com/Kavirubc/AgentDisruptBench	https://opensource.org/licenses/MIT	0.1.0	2026-04-03T00:00:00	{ "@type": "sc:Organization", "name": "AgentDisruptBench Contributors" }	[ "AI agents", "benchmarks", "resilience", "tool-calling", "disruptions", "evaluation", "LLM", "fault-injection", "reliability" ]	false	[ { "@type": "cr:FileObject", "@id": "retail-tasks", "name": "retail.yaml", "description": "20 standard retail domain tasks (difficulty 1-5)", "contentUrl": "python/agentdisruptbench/tasks/builtin/retail.yaml", "encodingFormat": "application/x-yaml", "sha256": "5c00f518f411c868ab8ee81c35c6...	[ { "@type": "cr:RecordSet", "@id": "tasks", "name": "Benchmark Tasks", "description": "The 100 benchmark tasks across 4 domains and 3 task types.", "field": [ { "@type": "cr:Field", "@id": "tasks/task_id", "name": "task_id", "description": "Unique task identi...

AgentDisruptBench

An evaluation methodology for measuring AI agent resilience under runtime tool-call disruptions.

Overview

AgentDisruptBench provides a systematic methodology, backed by 100 base tasks and variants across 4 domains, a systematic 20-type disruption taxonomy, and 9 disruption severity profiles, to study and measure how well LLM-based agents handle real-world tool failures.

Task Statistics

Domain	Standard	Adversarial	Impossible	Handover	Total
Retail	20	2	2	1	25
Travel	20	2	2	1	25
Finance	20	2	2	1	25
DevOps	20	2	2	1	25
Total	80	8	8	4	100

Disruption Taxonomy (20 Types)

Category	Types
Timing	timeout, latency
HTTP Status	http_429, http_401, http_403, http_500, http_502, http_503
Response Content	malformed_json, truncated, null_response, missing_fields, type_mismatch, schema_drift, wrong_data
Behavioral	intermittent, flapping, quota_exhausted, auth_expiry, cascading

Key Metric: R(k, ε, λ) Reliability Surface

k-consistency: Pass rate across repeated seeds (same task, same profile)
ε-robustness: Pass rate across task-wording variants
λ-fault-tolerance: Pass rate across disruption profiles

Files

tasks/retail.yaml — 20 retail domain tasks
tasks/travel.yaml — 20 travel domain tasks
tasks/finance.yaml — 20 finance domain tasks
tasks/devops.yaml — 20 DevOps domain tasks
tasks/adversarial.yaml — 8 adversarial trap tasks
tasks/impossible.yaml — 8 impossible tasks
tasks/handover.yaml — 4 handover tasks
tasks/variants.yaml — 18 ε-robustness task variants
profiles/ — 9 disruption profile definitions

Usage

pip install agentdisruptbench

from agentdisruptbench import TaskRegistry, DisruptionEngine
registry = TaskRegistry.from_builtin()
tasks = registry.filter(domain="retail", max_difficulty=3)

Citation

@inproceedings{agentdisruptbench2026,
  title={AgentDisruptBench: An Evaluation Methodology for AI Agent Resilience Under Runtime Tool-Call Disruptions},
  author={AgentDisruptBench Contributors},
  year={2026}
}

License

MIT

Downloads last month: 22

Size of downloaded dataset files:

7.1 kB

Size of the auto-converted Parquet files:

35.6 kB

Number of rows: