| Without PROBE | With PROBE |
|---|
| Skim a hundred titles, remember none | Scouting — a few scored papers, tied to your open decisions |
| "I'll read it later" → never | Analysis — a Korean page you finish in a sitting |
| Three papers, the same claimed win | Comparison — only where they part |
| "Surely someone has tried this" | Ideation — the untried idea, and how to refute it |
Every rewrite starts from the arXiv original, not the abstract. Every judgement
is made against your own pillars and Decision Log. All of it publishes
together, in Korean, on the reading site.
context/MASTER.md holds what cuts across pillars, and each context/P#.md
holds one pillar's decision log, tracked literature and anti-topics.
| Track | You run in Claude Code | It writes |
|---|
| Scouting | scheduled, every pillar in one run — setup · /scout [pillars] | scouting/P#/<date>.md and scouting/runs/<date>.md — format |
| Analysis | /analyze <arXiv id>, then /proofread <id> for looped passes against the original and its code | analysis/<id>.md — format |
| Comparison | /compare <id> <id> [<id>] | comparison/<slug>.md — format |
| Ideation | /ideate [<id | alias | topic>] | nothing in the repo — a chat answer, and a private page you answer on |
- Start by hand. One report, reviewed ruthlessly, then the schedule.
- You name the papers. A report never feeds
analysis/ on its own.
- No arXiv HTML, no rewrite. Never written from the abstract.
- The rewrite comes first.
/compare reads what /analyze wrote.
