The Teachometry benchmark Teachometry 基准 tutor-eval-v0.2a@0.2a.6
The benchmark, made observable. 让基准变得可观察。
Teachometry turns complex tutoring behavior into a structured, transparent benchmark. Explore authored cases, rubrics, and evaluation structure to understand how AI tutors diagnose, guide, adapt, and support learners in realistic scenarios.
Teachometry 将复杂的教学行为转化为结构化、透明的基准。通过人工编写的案例、评分标准和评测结构,了解 AI Tutor 在贴近真实教学的情境中如何诊断、引导、适应并支持学生。
Browse Cases 浏览案例 Read the Methodology 阅读方法论
Open data 开放数据 Research-oriented 面向研究 Community-driven 社区驱动
Dataset snapshot 数据集快照 Public development set 公开开发集
Authored cases 人工编写案例 48 Public scenarios 公开情境
Evaluation dimensions 评测维度 5 Versioned rubric schema 版本化评分标准结构
Subjects 学科 5 Authored subject areas 人工编写的学科领域
Capabilities 能力 27 Tutoring taxonomy tags 教学分类标签
Subject × locale · authored case counts 学科 × 语言 · 人工编写案例数量
Subject 学科 English 简体中文
Mathematics 数学 9 9Programming 编程 5 5Science 科学 4 4History Or Social Studies 历史或社会研究 3 3Language 语言 3 3
Darker dots = a larger share of that subject. Counts, not model scores. 圆点越深,表示占该学科案例的比例越高。这里展示数量,不是模型分数。 View all coverage 查看全部覆盖情况
Data coverage 数据覆盖
A diverse set ofrealistic scenarios. 覆盖多样的仿真教学情境。
The benchmark spans authored subjects, learner contexts, and tutoring capabilities, with structured evaluation criteria.
该基准覆盖人工编写的学科、学生语境和教学能力,并配套结构化评测标准。
tutor-eval-v0.2a@0.2a.6 128 rubrics · Developer Preview 128 项评分标准 · 开发者预览
Explore all cases 探索全部案例
Subjects 学科 5 categories 类别
Mathematics 数学 18 (38%) Programming 编程 10 (21%) Science 科学 8 (17%) History Or Social Studies 历史或社会研究 6 (13%) Language 语言 6 (13%)
Share of all cases. n = 48 占全部案例的比例;n = 48。
Learner levels 学生水平 5 categories 类别
Middle School 初中 20 (42%) Secondary 中学 12 (25%) Upper Elementary 小学高年级 8 (17%) Elementary 小学 6 (13%) Beginner 入门 2 (4%)
Share of all cases. n = 48 占全部案例的比例;n = 48。
Student states 学生状态 8 categories 类别
Novice 初学者 10 (21%) Overconfident Incorrect 过度自信但答案错误 8 (17%) Partial Understanding 理解不完整 8 (17%) Conceptual Misconception 概念误解 6 (13%) Others 其他 16 (33%) View all 8 student states 查看全部 8 种学生状态 Novice 初学者 10 (21%) Overconfident Incorrect 过度自信但答案错误 8 (17%) Partial Understanding 理解不完整 8 (17%) Conceptual Misconception 概念误解 6 (13%) Procedural Error 步骤错误 6 (13%) Stuck Without Attempt 未尝试便卡住 4 (8%) Uncertain But Correct 答案正确但不确定 4 (8%) Correct Answer Wrong Reasoning 答案正确但推理错误 2 (4%)
Share of all cases. n = 48 占全部案例的比例;n = 48。
Capabilities 能力 27 capability tags 能力标签
Student Executable Action 学生可执行的行动 22 (46%) Conceptual Correctness 概念正确性 20 (42%) Procedural Prompting 步骤提示 12 (25%) Check For Understanding 检查理解 10 (21%) View all 27 capabilities 查看全部 27 项能力 Student Executable Action 学生可执行的行动 22 (46%) Conceptual Correctness 概念正确性 20 (42%) Procedural Prompting 步骤提示 12 (25%) Check For Understanding 检查理解 10 (21%) Counterfactual Adaptation 反事实适应 10 (21%) Error Detection 错误检测 10 (21%) Error Localization 错误定位 10 (21%) Explanation Depth Adaptation 解释深度适应 10 (21%) Procedural Correctness 步骤正确性 10 (21%) Productive Question 有效提问 10 (21%) Scaffolding 学习支架 10 (21%) Factual Correctness 事实正确性 8 (17%) Prior Knowledge Adaptation 依据已有知识调整 8 (17%) Answer Non Disclosure 不泄露答案 6 (13%) Clear Next Step 明确下一步 6 (13%) Difficulty Adaptation 难度适应 6 (13%) Reasoning Consistency 推理一致性 6 (13%) Uncertainty Detection 不确定性检测 6 (13%) Conceptual Prompting 概念提示 4 (8%) Misconception Identification 识别概念误解 4 (8%) Misconception Specific Adaptation 针对概念误解的适应 4 (8%) Misleading Simplification 误导性简化 4 (8%) Correct Answer Wrong Reasoning 答案正确但推理错误 2 (4%) Hint Calibration 提示校准 2 (4%) Knowledge Gap Identification 识别知识缺口 2 (4%) Overhelping Avoidance 避免过度帮助 2 (4%) Student Agency 学生自主性 2 (4%)
Cases with each category; categories overlap. n = 48 各类别案例数,类别可重叠;n = 48。
Learning tasks (8), locales (2) & evaluation coverage 学习任务(8)、语言(2)与评测覆盖情况
Learning tasks 学习任务 8 categories 类别
Guided Problem Solving 引导式解题 12 (25%) Error Diagnosis 错误诊断 8 (17%) Answer Checking 答案检查 6 (13%) Concept Explanation 概念解释 6 (13%) Others 其他 16 (33%) View all 8 learning tasks 查看全部 8 个学习任务 Guided Problem Solving 引导式解题 12 (25%) Error Diagnosis 错误诊断 8 (17%) Answer Checking 答案检查 6 (13%) Concept Explanation 概念解释 6 (13%) Knowledge Recall 知识回忆 6 (13%) Reasoning Checking 推理检查 4 (8%) Transfer Preparation 迁移准备 4 (8%) Hint Request 提示请求 2 (4%)
Share of all cases. n = 48 占全部案例的比例;n = 48。
Locales 语言 2 categories 类别
English 24 (50%) 简体中文 24 (50%)
Share of all cases. n = 48 占全部案例的比例;n = 48。
Evaluation dimensions 评测维度 5 categories 类别
Actionability 可执行性 42 (88%) Guidance 引导能力 24 (50%) Correctness 正确性 18 (38%) Diagnosis 诊断能力 18 (38%) Adaptation 适应能力 14 (29%)
Cases with each category; categories overlap. n = 48 各类别案例数,类别可重叠;n = 48。
Dimension coverage counts cases with at least one rubric in each category; categories overlap. Benchmark 0.1 · artifact schema 1. Dataset growth recalculates coverage; schema changes require an explicit presentation version review. 维度覆盖统计各类别中至少包含一项评分标准的案例数,类别可重叠。基准 0.1 · 产物 schema 1。数据集增长会重新计算覆盖情况;schema 变更需要显式审查展示版本。
Disclosure policies 披露政策 : Full Solution Allowed 允许完整解答 : 8 · Full Solution Required 要求完整解答 : 6 · Hint Only 仅提供提示 : 16 · No Answer 不提供答案 : 8 · Partial Solution 部分解答 : 10
Open & reproducible 开放且可复现
Transparent by design. 从设计上保持透明。
Public development cases and evaluation criteria are open for inspection. Public case artifacts exclude evaluator-only answers and rubrics. Coverage describes the authored dataset, not teaching effectiveness or learning outcomes.
公开开发案例和评测标准可供检查。公开案例产物不包含仅供评估器使用的答案和评分标准。这里的覆盖情况描述的是人工编写的数据集,而不是教学效果或学习结果。
View on GitHub 在 GitHub 查看 Read the data documentation 阅读数据文档