Special-R1

Product or Programme · All products and programmes
Evidence · 3 claims (2 for, 1 against) · 1 study (1 design), q2 · 0 of 1 report an effect size · 3 claims rest on one study

Description

Special-R1 is a released artifact suite comprising a ZPD-filtered multi-domain tutoring corpus, the SPED-Tutor-Mix disability-aware evaluation set, seven model checkpoints, and an evaluation pipeline created by Unggi Lee et al.

Components

• Released Special-R1 artifacts: multi-domain ZPD-filtered corpus, SPED-Tutor-Mix evaluation set, seven checkpoints, and disability-aware evaluation pipeline: The article releases the training and evaluation artifacts behind Special-R1. The training corpus contains approximately 54k problems across four domains (mathematics, reading comprehension, social reasoning, life skills), filtered to a ZPD window of 1-60% solve rate by the Llama-3.1-8B student simulator and annotated with baseline accuracy. The evaluation set is the persona-augmented SPED-Tutor-Mix test split in which "every item carries a disability profile so that persona-aware metrics can be computed." The disability rubric is released as disability_types.yaml alongside the code, together with all seven model checkpoints and the evaluation pipeline. (Unggi Lee et al. (2026))

Claims

• The full Special-R1 model raises persona-aware Fit from 6.75 to 8.40 and SPED-rubric Helpfulness from 0.720 to 0.768 over a non-disability-aware baseline [+W]
• Disability-aware training does not degrade out-of-domain educational knowledge on OpenLearnLM [+W]
• Specific learning disability in mathematics is the weakest persona profile across all GRPO-trained models [-W]

Related Products and Programmes

• Open audit dataset of 1,800 LLM history-tutor responses across five personas

Key Sources

• Unggi Lee, Jihoi Na, Yeil Jeong, Haeun Park, Yeonju Jang. (2026). Reinforcement Learning for Special Education: Aligning LLM Tutors to Diverse Learners through Disability-Adaptive Training. https://arxiv.org/abs/2605.30670

Page metadata

Field	Value
type	product
id	special-r1
title	Special-R1
description	Special-R1 is a released artifact suite comprising a ZPD-filtered multi-domain tutoring corpus, the SPED-Tutor-Mix disability-aware evaluation set, seven model checkpoints, and an evaluation pipeline created by Unggi Lee et al.
status	review
product_kind	dataset
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.