/tmp/bucket

experiments, poems, art, and stray thoughts

0%

#1

A calendar view where my to-do list and diary live.
Easy to insert to and view,
and eventually, I can see them from a broader perspective.
I have this desire to look back on myself from a broader perspective,
though I’m not really sure what the benefit would be.
Maybe I simply want to know, “What would that look like?”

#2

想看一些更宏觀的東西
因為害怕眼光太狹隘而做了錯誤的決策(吧?)

examples:
learn Rust
p=np 問題
主權個人 書
首先把牛做成球 小說

1. Start from a conversation moment. Write down a real situation where people in Taiwan chat, like replying to a boss, a couple saying goodnight, a family group, or complaining about work. Then list 8 lines people actually type in it. If you can’t get to 8 easily, the idea is too narrow.

2. Search the store with those words. Open the LINE Sticker Shop and search the theme and 2 or 3 of the lines, for example 社畜, 婉拒 or 已讀. For each search, note three things:

  • roughly how many results there are;
  • how good the top 10 look;
  • whether the top results are official brands or individual creators.

3. Read what you find.

  • Gap: few results, or the ones there are look old, ugly or off-topic. This is the best case.
  • Proven but beatable: lots of results, and the top ones are by creators and look average. Demand is real and you can compete on style.
  • Skip: the top results are all brands, celebrities or famous characters. You’d be paying for their exposure.

4. Check the rankings for format, not topics. Look through the store’s top-selling and rising lists for Taiwan. For each of 10 entries, note:

  • whether it has captions;
  • how big the face is;
  • whether it’s daily replies or a theme;
  • whether it’s one character or a pair.

Copy the pattern that keeps showing up, never the character.

5. Test the idea on yourself. Go through your own LINE chats from the past week and count the moments you would have sent this sticker. Zero means probably not.

6. Score it from 1 to 3 on each of these:

  • how often people would use it;
  • how open the search results are;
  • how much you like it.

Make anything that scores 7 or more. The “how much you like it” score is what keeps your own taste in the process.

7. Keep a record. After 60 days, compare Taiwan sales against your scores, leaving out your own gift purchases. Within a few months you’ll see which kind of score predicts sales for you, and that becomes your own research method.

Steps 2 to 4 are the slow part, and I can do them for each idea and hand you the scores. You’d only do steps 1, 5 and the “how much you like it” score.

#1

思緒:
threads
mvp unreal learning proj?
貼圖的策略
著色書送審
生日
股票大跌

#2

  • 條件不一樣,操作就不一樣。之前收入豐厚的操作不適用於無收入,完全不一樣。
  • 槓桿一定要規劃起終點。這次沒有規劃終點,錯過很大的獲利還小虧。
  • 開槓桿真的要想清楚如果下跌(有收入情況下壓力真的小很多)

比較小的點:

  • 適時規雜訊,不要看太多相關影片
  • 留一些餘韻,好可以做一些小操作可以給足「情緒價值」,例如,下跌時還可以小買,但還是分批小買就好,少量多餐

#3

flower
sculpture

#1

reset my macbook air. it got hacked, tho the hack happened 2 years ago.
it was an interview. i pulled a scam repo and ran it. it installed things and cloned my chrome profiles or smth.
details:

  1. local app sends reqs to the server which always returns code as plain text
  2. the local app executes the returned code, which creates a hidden python script with multi-layer encoding
  3. that py script then pulls some other files
  4. i had claude-code scan my mac. it found my chrome profile or smth had been duplicated, and the creation time matched when i ran the repo

yah, that’s why i reset the storage on my mac.
also happened to clear out the local projects from my previous two companies… new start!

#2

adopted margo the cat. paid $250 cad

#3

thinking of how to manage my knowledgebase

  • markdown + blog + hackmd
  • both public and private
  • easily chain with ai
  • support chaining with social media

Take these N feature requests and implement them in parallel using independent subagents, one per feature branch. For each: (1) write a short spec, (2) write failing tests first, (3) implement until all tests pass without modifying unrelated files, (4) run a self-review that flags any change not directly tied to the stated goal and reverts scope creep, (5) commit only the intended files and open a PR. Keep the full existing test suite green for every branch. Summarize each PR and any tradeoffs at the end.