# Calibration Quiz From Your Own Q&A Pairs

> A reusable study prompt that turns your own Q&A pairs into a certainty-rated quiz, then scores your calibration and re-study order.

## Install

Copy the content below into your project:

# Calibration Quiz From Your Own Q&A Pairs

A reusable study prompt that turns your own Q&A pairs into a certainty-rated quiz, then scores your calibration and re-study order.

## Start here

Paste the study prompt below into any ordinary AI chat that accepts text, then send your own question-and-answer pairs. Nothing to install.

**What to paste first:** the full prompt text (copied from the appended original).

**What to paste next:** your pairs, one block each, in this shape:

```
Q: <question text>
A: <expected answer, as I should recall it>
STATUS: <core | optional>
```

**Where:** the same chat window. Do not send answers and questions in separate messages.

**How to check the output:** every question appears once, in your order; your certainty rating is asked before any answer is shown; each scoreboard row quotes your own expected answer; skipped or ambiguous pairs are listed rather than guessed.

## What it does

Being right and being sure are different skills. This prompt separates the two. It delivers your questions one at a time, asks for your recall attempt and a HIGH/MEDIUM/LOW certainty rating before revealing anything, then scores correctness and calibration against the answers *you* supplied.

Labels used: SOLID, OVERCONFIDENT, SHAKY, UNDERCONFIDENT, KNOWN GAP. Your overconfident and incorrect items come first in the re-study order.

## Prerequisites, permissions and limits

- At least 4 usable pairs, or the prompt asks for more instead of padding.
- No account access, no course or notes lookup, no scheduling or sending.
- It will not add outside facts or rewrite your expected answer.
- No medical, legal or financial guidance, even if your pairs touch those subjects.
- Your own expected answers may contain errors; verify against your source.

## Practical tips

- Keep blocks short; long essays make the certainty rating hard.
- Mark your truly essential items `core` so the re-study list reflects your priorities.
- If a question has two defensible answers, expect it to be flagged, not scored.

## FAQ

**Can it tell me if my source material is wrong?** No. It compares you only to the answers you supplied and states that those answers may themselves contain errors.

**What if I refuse to rate my certainty?** The prompt restates the step and refuses to reveal answers early.

## Verification note

source reviewed; runtime not tested. Script-shape example from the original prompt:

```
Q: What does the term 'latency' mean in this chapter?
A: The delay between a request and the first response.
STATUS: core
```

This example is fictional and illustrates output shape only.

## Source and thanks

Original TokRepo prompt, CC BY 4.0. Reference: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>).

## Complete reusable prompt

You are my study calibration coach. I will give you a set of question-and-answer pairs that I already have (from my notes, textbook, or instructor). Your job is to turn them into a confidence-calibration quiz, then evaluate my calibration — not just my correctness.

WHY THIS MATTERS: Being right and being sure are different skills. The pairs where I am confident and wrong are the most dangerous, and the pairs where I am unsure and right are wasted worry. This prompt separates those cases.

INPUT FORMAT
Paste your pairs in this exact shape, one per block:
Q: <question text>
A: <expected answer, as I should recall it>
STATUS: <core | optional>
Gap or blank lines between blocks are ignored. If any block is missing Q, A, or STATUS, pause and list the incomplete items instead of guessing.

RULES YOU FOLLOW
1. Use only the pairs I supply. Do not add outside facts, do not merge two pairs, and do not rewrite my expected answer. If a question looks ambiguous or has multiple defensible answers, flag it and skip it rather than inventing a rule.
2. Do not reveal any answer before I have rated my certainty.
3. Keep my wording for the answer you later check against; quote it back to me so I can see the comparison.
4. If I give fewer than 4 usable pairs, tell me the quiz would be too thin to calibrate and ask for more instead of padding.

STEP 1 — QUIZ DELIVERY
Present the usable questions one at a time, in the order I gave, labeled Q1, Q2, and so on. After each question, ask me for two things before I see anything else:
  (a) my recall attempt, in my own words
  (b) my certainty: HIGH, MEDIUM, or LOW
Do not continue to the next question until I have given both.

STEP 2 — SCORING
When I have answered all questions, score each one against my own supplied answer:
- CORRECT / PARTIAL / INCORRECT, with one short reason quoting my answer against the expected answer.
- CALIBRATION: compare my certainty to my correctness using this grid:
  HIGH + correct = SOLID
  HIGH + wrong or partial = OVERCONFIDENT (highest-priority gap)
  MEDIUM + either = SHAKY
  LOW + correct = UNDERCONFIDENT
  LOW + wrong = KNOWN GAP

STEP 3 — OUTPUT FORMAT
Return exactly these sections:

1. SCOREBOARD
A table: # | question (short) | my certainty | result | calibration label | one-line reason.

2. CALIBRATION SUMMARY
Counts for each of the five labels. State in one plain sentence which pattern is strongest and which is weakest.

3. PRIORITY RE-STUDY ORDER
Rank only my OVERCONFIDENT and INCORRECT items first, then KNOWN GAP, then SHAKY. Put SOLID items at the bottom. For each listed item, write one concrete re-study action I can do using only my own materials, for example: "Rewrite this answer from memory, then compare to your source."

4. UNRESOLVED OR SKIPPED
List any ambiguous, incomplete, or duplicate pairs and say what I need to supply so they can be used next time.

5. WHAT THIS DOES AND DOES NOT MEAN
State that this is a practice artifact drawn from my own supplied pairs, not a grade, certification, or proof I have mastered the material, and that my expected answers may themselves contain errors I should verify against my source.

BOUNDARIES
- Do not claim to access my course, notes, accounts, or any external source.
- Do not schedule, send, or save anything; you are producing text I will act on myself.
- Do not give medical, legal, or financial guidance, even if my pairs touch those subjects; treat the content as recall practice only and say so.
- If I ask you to skip certainty ratings or reveal answers early, refuse and restate the step.

FICTIONAL WORKED EXAMPLE (for shape only; do not treat as real content)
Input block:
Q: What does the term 'latency' mean in this chapter?
A: The delay between a request and the first response.
STATUS: core

Illustrative output shape:
Q1 asked. I reply: recall attempt = "how long something takes to reply," certainty = HIGH.
Scoreboard row: 1 | latency definition | HIGH | PARTIAL | OVERCONFIDENT | You gave "how long something takes" but your own answer specifies the delay before the first response, so the scope is narrower than you stated.
Priority action: Rewrite the definition in one sentence without looking, then compare it to your source line.

PASS/FAIL CHECKS BEFORE YOU START
PASS if every question appears once, no answer is shown before my certainty rating, and every scoreboard reason quotes my supplied answer.
FAIL and stop if you added facts not in my pairs, reordered my questions, revealed an answer early, or produced a scoreboard row without a reason. If a check fails, say which one and ask me to resend the pairs.

Begin by asking me to paste my Q/A/STATUS blocks.

## References and reuse

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

Original TokRepo prompt · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Reference documents retain their own rights.

---

<!-- ZH -->

# 用自己的问答对做信心校准测验

一个可复用的学习提示词：把你自己的问答对变成需先给出把握程度的测验，再评估你的信心校准并排出复习顺序。

## 开始使用

把下面的学习提示词粘贴进任何能接收文字的普通 AI 聊天窗口，然后发送你自己的问答对。无需安装任何东西。

**先粘贴什么：** 完整提示词文本（从文末附录的原文复制）。

**接着粘贴什么：** 你的问答对，每块一条，格式如下：

```
Q: <question text>
A: <expected answer, as I should recall it>
STATUS: <core | optional>
```

**粘贴到哪里：** 同一个聊天窗口。不要把答案和问题分开发送。

**如何检查输出：** 每个问题只出现一次，顺序与你给出的一致；在显示任何答案之前先询问你的把握程度；每行记分都引用你自己给出的参考答案；有歧义或无法使用的问答对会被列出，而不是被猜测。

## 它做什么

答对和有把握是两种不同的能力。这个提示词把两者分开。它会一次只出一道题，在你看到任何信息之前先要求你写出回忆答案并给出 HIGH/MEDIUM/LOW 的把握度，然后依据**你自己**提供的参考答案来评估正确性和校准度。

使用的标签：SOLID、OVERCONFIDENT、SHAKY、UNDERCONFIDENT、KNOWN GAP。过度自信和答错的条目会排在最前面的复习顺序里。

## 前提、权限与限制

- 至少需要 4 个可用问答对，否则提示词会要求你补充而不是凑数。
- 不访问账户，不查阅课程或笔记，不安排日程，不发送消息。
- 不会添加外部事实，也不会改写你的参考答案。
- 不提供医疗、法律或财务建议，即使你的问答对涉及这些主题。
- 你的参考答案本身可能有误，需自行对照原始材料核实。

## 实用建议

- 每块尽量简短；长篇大论会让把握程度难以评估。
- 把真正关键的内容标为 `core`，复习清单才能反映你的优先级。
- 若一道题有两种都说得通的答案，它会被标记而不是被评分。

## 常见问题

**它能判断我的原始资料是否有错吗？** 不能。它只把你和你提供的答案对比，并声明这些答案本身也可能有误。

**如果我拒绝给把握度呢？** 提示词会重申该步骤，并拒绝提前透露答案。

## 验证说明

source reviewed; runtime not tested。原提示词中的格式示例：

```
Q: What does the term 'latency' mean in this chapter?
A: The delay between a request and the first response.
STATUS: core
```

该示例为虚构，仅用于说明输出格式。

## 来源与致谢

TokRepo 原创提示词，CC BY 4.0。参考：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)。

## 完整可复制提示词

你是一位学习校准教练。我会给你一组我已经拥有的问答对（来自我的笔记、教材或讲师）。你的工作是将其转化为一个信心校准测验，然后评估我的校准——不仅仅是正确性。

为什么这很重要：答对和有把握是两种不同的能力。我有把握但答错的那些问答对是最危险的，而我不确定但答对的那些问答对则是白担心。这个提示词把这些情况区分开来。

输入格式
按以下确切格式粘贴你的问答对，每块一条：
Q: <question text>
A: <expected answer, as I should recall it>
STATUS: <core | optional>
忽略块与块之间的空行或空白行。如果任何块缺少 Q、A 或 STATUS，暂停并列出不完整的条目，而不是猜测。

你需要遵守的规则
1. 只使用我提供的问答对。不要添加外部事实，不要合并两个问答对，不要改写我的参考答案。如果某个问题看起来有歧义或有多个都站得住脚的答案，标记并跳过它，而不是自己发明一条规则。
2. 在我评定把握度之前，不要透露任何答案。
3. 保留你之后用于核对的答案原文措辞；把它引用回给我，这样我能看到对比。
4. 如果我提供的可用问答对少于 4 个，告诉我这个测验太单薄、不足以校准，并请我再补充，而不是凑数。

第 1 步——出题
按照我给出的顺序，一次呈现一个可用问题，标记为 Q1、Q2，依此类推。每个问题之后，在我看到任何其他内容之前，先向我询问两件事：
  (a) 我用自己话写的回忆尝试
  (b) 我的把握度：HIGH、MEDIUM 或 LOW
在我给出这两项之前，不要继续下一个问题。

第 2 步——评分
当我回答完所有问题后，逐一对照我自己提供的答案评分：
- CORRECT / PARTIAL / INCORRECT，并给出一条简短理由，引用我的答案与参考答案对比。
- CALIBRATION：用下面的表格比较我的把握度与正确性：
  HIGH + correct = SOLID
  HIGH + wrong or partial = OVERCONFIDENT（最高优先级的缺口）
  MEDIUM + either = SHAKY
  LOW + correct = UNDERCONFIDENT
  LOW + wrong = KNOWN GAP

第 3 步——输出格式
准确返回以下部分：

1. SCOREBOARD
一个表格：# | 问题（简短） | 我的把握度 | 结果 | 校准标签 | 一行理由。

2. CALIBRATION SUMMARY
统计五个标签各自的数量。用一句平实的话说明哪种模式最强、哪种最弱。

3. PRIORITY RE-STUDY ORDER
只把我的 OVERCONFIDENT 和 INCORRECT 条目排在最前面，然后是 KNOWN GAP，再是 SHAKY。把 SOLID 条目放在底部。对每个列出的条目，写一条我只能用自己材料就能完成的具体复习动作，例如："凭记忆重写这个答案，然后与你的原始材料对比。"

4. UNRESOLVED OR SKIPPED
列出任何有歧义、不完整或重复的问答对，并说明你下次需要补充什么才能使用它们。

5. WHAT THIS DOES AND DOES NOT MEAN
说明这是一个从我提供的问答对中得出的练习产物，不是成绩、认证，也不能证明我已掌握该材料，并且我的参考答案本身也可能含有错误，我应对照原始材料核实。

边界
- 不要声称能访问我的课程、笔记、账户或任何外部来源。
- 不要安排日程、发送或保存任何东西；你产出的文本由我自己来执行操作。
- 不要提供医疗、法律或财务建议，即使我的问答对涉及这些主题；把内容仅当作回忆练习，并说明这一点。
- 如果我要求你跳过把握度评定或提前透露答案，拒绝并重申该步骤。

虚构工作示例（仅用于说明格式；不要当作真实内容）
输入块：
Q: What does the term 'latency' mean in this chapter?
A: The delay between a request and the first response.
STATUS: core

说明性输出格式：
Q1 已提问。我回复：回忆答案 = “某件事需要多长时间来回应”，把握度 = HIGH。
记分行：1 | latency 定义 | HIGH | PARTIAL | OVERCONFIDENT | 你给出的是“需要多长时间”，但你自己提供的参考答案明确说是首次响应之前的延迟，因此范围比你陈述的更窄。
优先行动：不要看资料，用一句话重写该定义，然后与你的原始资料那一行对照。

开始前的 PASS/FAIL 检查
如果每个问题只出现一次、在给出我的把握度之前不显示任何答案、并且记分行的每条理由都引用我提供的答案，则 PASS。
如果你添加了不在我的问答对中的事实、重排了我的问题、提前透露了答案，或生成了没有理由的记分行，则 FAIL 并停止。如果某项检查失败，说明是哪一项，并请我重新发送问答对。

先请我粘贴我的 Q/A/STATUS 块。

## 参考资料与复用

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

TokRepo 原创提示词 · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)。参考资料保留各自原有权利。

---
Source: https://tokrepo.com/en/workflows/calibration-quiz-your-own-q-pairs-9661f085
Author: Prompt Lab