JM
Library · Eval datasets

Eval datasets

Curated benchmarks distilled from SME-validated eval runs. Versioned, hashed, licensable. The Atlas IP layer.

Datasets
3
1 available · 1 beta · 1 roadmap
Curated items
9,430
SME-validated, hashed
Source runs
29,000
Production traffic origins
Avg SME validation
96 %
Inter-rater agreement weighted
Registry · 3 datasetsSorted by recency
DatasetLangItemsSource runsSME val.LicenseStatus
Tagalog Healthcare Adversarial Prompts — 2026 Q2
atlas-ds-tl-healthcare-2026q2·v0.4.0-beta
tl
Tagalog
4,28012,90097%Eval + finetune
$28k
beta
Korean Clinical Register & Safety Set — 2026 Q1
atlas-ds-ko-clinical-2026q1·v1.2.0
ko
Korean
3,1409,70096%Eval + finetune
$32k
available
Japanese Keigo-Stability Adversarial Set
atlas-ds-ja-keigo-2026q2·v0.1.0-roadmap
ja
Japanese
2,0106,40095%Eval-only
tbd
roadmap
MaranoAtlas
Atlas IP · 15 runs in lineage graph

Atlas command palette

Jump to runs, SMEs, rubrics, projects, or screens.