快速筛选
Quick screening
判断一篇论文是否值得继续投入时间。
Decide whether a paper deserves deeper time.
结构化论文阅读
把论文读成可以验证、比较和迁移的研究材料。
目录
Contents
为什么做Why
针对什么问题Problem
可以用来做什么Use cases
怎么使用How to use
五步工作流Five-step workflow
质量门槛Quality gates
为什么做
Why this workflow exists
普通摘要很容易丢掉页码、实验条件和失败边界。读完之后，我们往往记得一句结论，却说不清它由什么证据支持、能否迁移到自己的系统。
A conventional summary easily loses page references, experimental conditions and failure boundaries. After reading, we may remember a claim but cannot explain what supports it or whether it can transfer into our own system.
这套流程的目标，是让论文阅读从“看过”变成“可以验证、比较和继续积累”。
The goal is to move from having read a paper to being able to verify, compare and build on it.
针对什么问题
Problems it addresses
- 01
结论缺少证据定位
Claims without evidence
核心结论没有回到页码、章节、公式、图或表。
Key claims are not tied back to pages, sections, equations, figures or tables.
- 02
条件和边界被省略
Missing conditions and limits
看不到基线是否公平、消融是否充分、成本是否可以接受。
Baseline fairness, ablations, cost and deployment limits remain unclear.
- 03
多篇论文无法累积
No accumulation across papers
每篇笔记格式不同，后续无法比较方法、数据和结论。
Inconsistent notes make it hard to compare methods, data and conclusions later.
- 04
输出只有一种形态
One output for every purpose
组会、复现、方法迁移和方向综述需要不同结构。
Group meetings, reproduction, method transfer and surveys require different structures.
可以用来做什么
What it can produce
标准深读
Deep reading
重建问题、方法、实验、消融、失败与复现风险。
Reconstruct the problem, method, experiments, ablations, failures and reproducibility risks.
方法迁移
Method transfer
提炼可以进入自己系统的机制和最小验证实验。
Extract mechanisms that can enter another system and define a minimum validation experiment.
多篇综述
Multi-paper synthesis
用统一矩阵比较共识、分歧、方法演化和研究空白。
Compare consensus, disagreement, method evolution and research gaps in one matrix.
怎么使用
How to use it
输入可以是 arXiv 链接、DOI、本地 PDF、论文题目，也可以是一组论文。把来源交给 Codex 后，再选择阅读模式即可。
Inputs can be an arXiv link, DOI, local PDF, title or a collection of papers. Give the source to Codex, then choose the reading mode.
读：https://arxiv.org/abs/2210.03629
Read: https://arxiv.org/abs/2210.03629
深读这三篇并做横向综述
Deep-read these three papers and create a synthesis
把这篇论文做成 12 页 PPT，重点讲方法和真机实验
Turn this paper into a 12-page deck focused on the method and real-robot experiments
五步工作流
Five-step workflow
01来源审计Source audit版本、作者、图表完整性Version, authors, completeness
02证据抽取Evidence绑定页码、章节、图与表Bind claims to pages and figures
03结构化深读Deep read问题、方法、实验、边界Problem, method, evidence, limits
04批判复核Critique基线、消融与复现风险Baselines, ablations, reproducibility
05复用输出Reuse笔记、矩阵、汇报与方案Notes, matrix, deck and plan
质量门槛
Quality gates
- 核心结论有页码、章节或图表定位。
- Key claims have page, section, figure or table references.
- 明确区分作者主张与独立判断。
- Paper claims and independent assessments stay separate.
- 复述实验设置、基线、指标和数据规模。
- Experiment setup, baselines, metrics and data scale are reconstructed.
- 指出失败案例、适用边界、成本和复现风险。
- Failure cases, limits, cost and reproducibility risks are explicit.