2022 与 2023–2024 年样本相比，「数形结合」由 25% 变为 60%。这是收录题目的结构差异，不外推下一年。
按前后两个有样本时期比较分布；这是历史观察，不是未来预测。
GAOKAO INTELLIGENCE / EXAM TRENDS
2022–2024 北京卷数学 · 函数与导数命题趋势
用历史真题回答三件事：发生了什么、证据在哪里、接下来练什么。分析基于当前已收录公开样本，不等同于全国全量试卷。
9public past questions
3papers
3years with samples
3structural changes
Methodology:北京卷数学 · 2022–2024explains historical samples only and does not predict future questions · View evidence criteria
01 / WHAT CHANGED
What changes do these past papers show?
Conclusions are first generated by deterministic aggregation, then backed by question and paper evidence. Dimensions with too few samples get no conclusion.
2022 与 2023–2024 年样本相比，「中等」由 75% 变为 100%。这是收录题目的结构差异，不外推下一年。
按前后两个有样本时期比较分布；这是历史观察，不是未来预测。
2022 与 2023–2024 年样本相比，「逻辑推理」由 50% 变为 80%。这是收录题目的结构差异，不外推下一年。
按前后两个有样本时期比较分布；这是历史观察，不是未来预测。
02 / EVIDENCE TRACK
Representative Questions by Year
Every node links back to real questions. A node reflects the structure observable in that year, not every paper from that year.
03 / CORPUS READINESS
First Check Whether Each Year's Data Is Comparable
Years with too few public questions or insufficient knowledge map coverage are never reported as "not tested" and are excluded from early-versus-late comparisons.
| year | 公开题 | 主知识标签覆盖 | 函数与导数证据题 | 比较状态 |
|---|---|---|---|---|
| 2022 | 20 | 80% | 4 | 参与比较 |
| 2023 | 21 | 81% | 2 | 参与比较 |
| 2024 | 21 | 81% | 3 | 参与比较 |
| 2025 | 21 | 62% | — | 语料未就绪 |
| 2026 | 12 | 42% | — | 语料未就绪 |
04 / REPRESENTATIVES
Which Questions Best Illustrate This Change?
One question is picked per year, favoring fully tagged samples that show the structural difference, rather than simply ranking by difficulty.
Where Do the Conclusions Come From?
题目公开状态与结构化标签来源是两套证据：公开题通过发布门槛；知识点和属性标签仍可能来自 AI，不能描述为人工审核趋势标签。
发现 1 处题型字段与结构化标签不一致（Q139355579372665）。统计统一采用 qb_questions.type，冲突不静默覆盖。
- 公开题状态
- approved / machine_verified
- 知识点标签来源
- ai: 17
- 属性标签来源
- ai: 85 / rule: 6
- 排除重复试 Paper
- 4
沿着这条变化练 3 道题
先识别基础结构，再做代表题，最后换题面验证方法迁移。题目由现有 Question Graph 与趋势证据共同选择。
1. 基础识别→2. 代表结构→3. 换题迁移