Amparo-MX: Mexican Federal Legal Reasoning Dataset

The first ML-ready structured dataset of Mexican federal jurisprudence (juicio de amparo), sourced from the Semanario Judicial de la Federación (SCJN) — Mexico's official federal judicial record with public thesis.

Why it matters

Currently, several assessment datasets exist for legal NLP in common law jurisdictions. Mexico's amparo trial, a distinctive constitutional instrument with no direct equivalent in other systems, is entirely omitted from open-source machine learning datasets. The subset on indigenous rights categorizes terms of Nahuatl origin incorporated into formal legal Spanish, a field previously lacking NLP resources. No comprehensive public dataset of Mexican indigenous jurisprudence exists. Despite growing judicial attention to indigenous rights over the last decade, these rulings remain dispersed across multiple eras and courts, with no structured ML-ready corpus. This dataset fills that gap.

Dataset Size & Quality

Version	Rows	Grade	Percentile	Notes
v1	50	—	—	Raw extraction, no augmentation
v2	50	C	8.4%	Structured sections + local edits
v3	50	A	57.7% (9.6/10)	Enhanced via Adaptive Data
v4	98	A	57.7% (9.5/10)	ES(97) + HT(28) + QU(24) enhanced pairs · Filtered

Language Coverage (v4)

Language	Code	Records	Field
Spanish (legal)	es	97	All fields
Haitian Creole	ht	28	enhanced_prompt / enhanced_completion
Quechua	qu	24	enhanced_prompt / enhanced_completion

Multilingual coverage is concentrated in the fine-tuning fields (enhanced_prompt, enhanced_completion). Base legal fields (hechos, criterio_juridico, justificacion) are Spanish-only, reflecting the source jurisdiction.

Schema

Field	Description
hechos	Case facts — factual background
criterio_juridico	Legal criterion — core reasoning
justificacion	Doctrinal justification
materias	Subject areas (Penal, Civil, Constitucional…)
terminos_indigenas	Legal Spanish terms referencing indigenous concepts — NOT Nahuatl words, but formal Spanish expressions mediating indigenous rights
contiene_indigena	Whether case involves indigenous rights
epoca	Judicial era (9a–12a Época)
instancia	Court instance
enhanced_prompt	Fine-tuning ready prompt (Adaptive Data) — multilingual
enhanced_completion	Fine-tuning ready completion (Adaptive Data) — multilingual

Linguistic Finding

Initial hypothesis: Nahuatl terms (tequio, calpulli, tlatoani) would appear in federal rulings involving indigenous communities. Indigenous law is entirely expressed through formal legal Spanish. The terminos_indigenas field captures how indigenous concepts are translated and incorporated into legal language, rather than the original terms themselves.

Nahuatl term	Legal Spanish equivalent used in rulings
TLATEKIWILI	"usos y costumbres"
TEQUITL	"trabajo comunitario (tequio)"
IPATLAKA AXKAYOTL	"comunero"

This is itself a research finding: indigenous communities appear as legal subjects in federal courts, but their own terminology is invisible. The judicial system mediates indigenous rights through Spanish legal categories — a form of institutional translation that this dataset makes legible for NLP research.

Representation bias note: Indigenous communities appear as passive subjects in case facts, but their own conceptual frameworks are not present in the text. Models trained on this data should account for this structural asymmetry.

Dataset Splits

• train: v4 (98 rows, Grade A, Enhanced + Expanded World + Filtered)
• raw_v3: v3 (50 rows, Enhanced via Adaptive Data)
• raw_v1: v1 (50 rows, raw extraction)
• raw_v2: v2 (50 rows, structured + local edits)

Roadmap (Coming in v5+)

• 200+ real cases via SCJN sequential scan across 9a–11a Épocas
• Nahuatl↔Spanish legal terminology mapping
• Baseline eval metrics (ROUGE, legal reasoning accuracy)

Downstream Tasks

• Legal subject classification: given hechos + criterio_juridico → predict materias
• Argument extraction: identify reasoning chain structure
• Cross-lingual transfer: indigenous legal terminology in formal Spanish
• Institutional bias analysis: how judicial language mediates indigenous rights

Source & License

• Source: Semanario Judicial de la Federación (SCJN) — public official record
• License: CC BY 4.0
• Created with Adaptive Data by Adaption
• Submitted to the Uncharted Data Challenge

Citation

@dataset{amparo_mx_2026,
  author    = {V. Alva, Jalil},
  title     = {Amparo-MX: Mexican Federal Legal Reasoning Dataset},
  year      = {2026},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/jvz37/amparo-mx},
  note      = {Sourced from Semanario Judicial de la Federación (SCJN).
               Quality augmentation via Adaptive Data by Adaption Labs.}
}

About

Built by Jalil V. Alva as part of the Uncharted Data Challenge 2026. Augmented using Adaptive Data by Adaption.