- Units follow the math单位自己跟着算
-
Write
DAU × ARPU × 30 day and it knows the answer is
money, not headcount. Units carry through the whole expression on their own.
写 日活 × 客单价 × 30 day，它知道结果是钱，不是人数。
单位跟着算式一路推导，不用自己盯。
- What an AI feature will cost算 AI 功能要烧多少钱
-
2k tokens/user × $3/Mtok gives cost per user; multiply by
DAU for the monthly bill. You supply the price — a built-in table would go stale,
and a stale price table is worse than none.
2k tokens/user × $3/Mtok 得到每用户成本，再乘日活就是月账单。
模型价格自己填——内置的价格表过期比没有更危险。
- How long the test has to run这个实验要跑几天
-
Sample size for the lift you care about, divided by daily traffic:
DAU = 12k users/day，然后 samplesize(2.4%, 10%) users / DAU → 11.2 day
检出目标提升需要多少样本，除以日活就是天数：
样本量(2.4%, 10%) users / 日活 → 11.2 天
- Points versus percent涨了 0.61 个点还是涨了 12%
-
Conversion from 5% to 5.61% is both "+0.61 points" and "+12%".
5% → 5.61% gives you both, so you don't quote the wrong one.
转化率从 5% 到 5.61%，两种说法都对但差别很大。
5% → 5.61% 两个数一起给，汇报时不会说错。
- Name your assumptions给数字起名字
-
Write
revenue = DAU × ARPU × 30 instead of
12000×39×30. Change one assumption and the rest recomputes —
and it still makes sense three days later.
写 月收入 = 日活 × 客单价 × 30 而不是 12000×39×30。
改一个假设整段自动重算，三天后回来还看得懂。
- Paste it somewhere else算完贴给别人看
-
⌘⇧A copies the whole draft with results
aligned; ⌘⇧M copies it as a Markdown table.
⌘⇧A 复制整份草稿，算式和结果对齐排好；
⌘⇧M 复制成 Markdown 表格。
- Money that adds up钱算到分不出错
-
0.1 + 0.2 is 0.3. Most calculators give you
0.30000000000000004 — floating point and money don't mix.
0.1 + 0.2 就是 0.3。多数计算器在这里会给你
0.30000000000000004——用浮点数算钱迟早出事。
