File size: 4,468 Bytes
25c416e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
---
language:
- en
license: apache-2.0
tags:
- geospatial
- agent-benchmark
- jurisdictional-routing
- geometry-validation
- delegation-chain
- gdpr
- eu-ai-act
- h3
- gns-protocol
pretty_name: GEIANT Geospatial Agent Benchmark
size_categories:
- n<1K
task_categories:
- text-classification
- question-answering
---

# GEIANT Geospatial Agent Benchmark

**The first benchmark dataset for geospatial AI agent orchestration.**

Built on the [GNS Protocol](https://gcrumbs.com) — the decentralized identity system that proves humanity through Proof-of-Trajectory.

## Overview

Every AI orchestrator (LangChain, CrewAI, AutoGPT) routes tasks based on capability and availability. None of them understand *where* the task originates, *what regulatory framework* governs that location, or *whether the geometry the agent produced is actually valid*.

GEIANT fixes this. This benchmark tests three capabilities no other orchestrator has:

| Capability | What it tests |
|---|---|
| **Jurisdictional Routing** | H3 cell → country → regulatory framework → agent selection |
| **Geometry Mutation Integrity** | Multi-step geometry workflows with injected corruption |
| **Delegation Chain Validation** | Human→agent authorization cert validity |

## Dataset Statistics

**Total records:** 40

### By Family
| Family | Count |
|---|---|
| `jurisdictional_routing` | 14 |
| `geometry_mutation` | 11 |
| `delegation_chain` | 15 |

### By Difficulty
| Difficulty | Count |
|---|---|
| `easy` | 16 |
| `medium` | 13 |
| `hard` | 4 |
| `adversarial` | 7 |

### By Expected Outcome
| Outcome | Count |
|---|---|
| `route_success` | 15 |
| `reject_delegation` | 10 |
| `reject_geometry` | 7 |
| `reject_no_ant` | 4 |
| `reject_tier` | 2 |
| `reject_no_jurisdiction` | 1 |
| `flag_boundary_crossing` | 1 |

## Schema

Each record is a `DatasetRecord` with the following fields:

```typescript
{
  id: string;                    // UUID
  family: DatasetFamily;         // which benchmark
  description: string;           // human-readable scenario description
  input: object;                 // the task/cert/geometry submitted
  expected_outcome: string;      // what GEIANT should do
  ground_truth: {
    expected_ant_handle?: string;
    expected_country?: string;
    expected_frameworks?: string[];
    geometry_valid?: boolean;
    delegation_valid?: boolean;
    explanation: string;         // WHY this is the correct answer
  };
  difficulty: string;            // easy | medium | hard | adversarial
  tags: string[];
}
```

## Regulatory Frameworks Covered

| Framework | Jurisdiction | Max Autonomy Tier |
|---|---|---|
| GDPR | EU | trusted |
| EU AI Act | EU | trusted |
| eIDAS2 | EU | certified |
| FINMA | Switzerland | certified |
| Swiss DPA | Switzerland | certified |
| UK GDPR | United Kingdom | trusted |
| US EO 14110 | United States | sovereign |
| CCPA | California, USA | sovereign |
| LGPD | Brazil | trusted |
| PDPA-SG | Singapore | trusted |
| Italian Civil Code | Italy | trusted |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("GNS-Foundation/geiant-geospatial-agent-benchmark")

# Filter by family
routing = ds.filter(lambda x: x["family"] == "jurisdictional_routing")

# Filter by difficulty
adversarial = ds.filter(lambda x: x["difficulty"] == "adversarial")

# Get all rejection scenarios
rejections = ds.filter(lambda x: x["expected_outcome"].startswith("reject_"))
```

## Geospatial Moat

This dataset uses **H3 hexagonal hierarchical spatial indexing** (Uber H3) at resolution 5–9. Each agent is assigned a territory as a set of H3 cells. Routing validates that the task origin cell is contained within the agent's territory — not just lat/lng bounding boxes.

The H3 cells in this dataset are generated from real coordinates:

```python
import h3
rome_cell = h3.latlng_to_cell(41.902, 12.496, 7)
# → '871e805003fffff'
```

## Citation

```bibtex
@dataset{geiant_benchmark_2026,
  author    = {Ayerbe, Camilo},
  title     = {GEIANT Geospatial Agent Benchmark},
  year      = {2026},
  version   = {0.1.0},
  publisher = {GNS Foundation / ULISSY s.r.l.},
  url       = {https://huggingface.co/datasets/GNS-Foundation/geiant-geospatial-agent-benchmark}
}
```

## License

Apache 2.0 — free for research and commercial use.

---

*Built with [GEIANT](https://github.com/GNS-Foundation/geiant) — Geo-Identity Agent Navigation & Tasking.*
*Part of the [GNS Protocol](https://gcrumbs.com) ecosystem.*