hummbl-hf/agent-wicked-problems-40k

HUMMBL 40k Multi-Agent Wicked Problems & Coordination Corpus

A foundational 40,171-event empirical dataset capturing real-world multi-agent coordination, epistemic problem decomposition, failure mode taxonomies, and strategic intelligence surges generated across the HUMMBL autonomous agent fleet.

Dataset Overview

The dataset provides structured visibility into how autonomous agents navigate complex, ill-defined ("wicked") problems, coordinate across distributed execution surfaces, and surface edge cases in agent governance.

Structure & Fields

Each record represents a discrete coordination or intelligence event:

Field	Type	Description
source	string	Originating coordination ledger or telemetry bus stream
line	integer	Sequential monotonic entry sequence
type	string	Interaction type (intel_surge, coordination_status, inquiry, etc.)
timestamp	string (ISO 8601)	Exact temporal timestamp of event emission
text_head	string	Core semantic payload, query, or problem statement
problem_ids	list[string]	Mapped wicked problem and failure taxonomy codes
primary	string	Primary governing problem category (e.g. P2, P7, P12)
int_code	string	Intelligence collection classification (e.g. OSINT, MASINT, CYBINT)
goal_id	string	Scoped autonomous agent session or mission identifier
sources_count	integer	Number of verified evidentiary citations supporting the event
validation_status	string / null	Formal verification status

Key Applications

1. Multi-Agent Coordination Research: Benchmark how agents decompose ambiguous requests into structured inquiry loops.
2. AI Safety & Alignment: Train classifiers to detect circular coordination locks, prompt injection surfaces, and failure mode emergence.
3. Epistemic Routing: Fine-tune router models to classify incoming technical inquiries into domain-specific reasoning operator lattices.

Usage

from datasets import load_dataset

dataset = load_dataset("hummbl/agent-wicked-problems-40k")
print(dataset["train"][0])

Privacy & Redaction

This dataset has been processed through the HUMMBL automated pre-push sanitization engine:

• All sensitive credentials, API keys, tokens, and authorization headers have been completely redacted.
• Private network IPs and local host paths have been sanitized.
• Purely public research, coordination mechanics, and domain ontologies are preserved.

Citation & License

Published by HUMMBL, LLC under the Apache 2.0 License.

• GitHub: github.com/hummbl-io/oss