Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

I Played Out a Broke Student's Month 500 Times on a Budget i3 Laptop With the Wi-Fi Off. It Warned Him Ahead in 4 of 5 Bad Months

说明与用途

Broke Date 在学生的 i3 电脑上离线预测 500 条支出未来,提前 27 天警告资金短缺,并提供以天为单位的成本估算。

问题

A student with a monthly allowance needs to know if he can afford a purchase today without running out of money before payday, but existing tools either lack forecasting or require uploading bank data.

结果

On a simulated year, 4 of 5 months that ran out of money were flagged ahead of time (27, 21, 5 and 4 days before). The fifth month was flagged only on the day itself, with 2 false alarms. The app uses 1.0 GB RAM and answers each what‑if in 0.05 s on an i3‑1215U.

工作流

  • 1. Import bank statement into the app on the local laptop
  • 2. TabPFN forecasts 99 quantiles of tomorrow’s spend from 14 recent numbers
  • 3. Chain forecasts day‑by‑day to generate 500 simulated futures
  • 4. Evaluate each future for reaching payday and compute risk
  • 5. Present results: safe to spend today, number of futures that succeed, plan cost in days, odd spend odds
  • 6. Allow user to ask specific questions (e.g., “Can I afford ₹400 on Saturday?”) and receive a sentence answer and chart
  • 7. Provide a time‑machine view to assess past months’ accuracy
  • 8. Offer a “Plans” screen to price planned expenses in days of runway

证据

It reads his bank statement on his own laptop, learns how he actually spends, and plays out the rest of his month 500 times

来源

On a simulated year, walk‑forward with rules I wrote down first: 4 of 5 months that ran out were flagged ahead of time (27, 21, 5 and 4 days before). The fifth was flagged only on the day itself, and there were 2 false alarms.

来源

Runs on a budget laptop: an i3‑1215U with integrated graphics, 1.0 GB of memory at peak, and every what‑if answered in 0.05 s.

来源

  • 记录日期: 2026-10-04T20:11:02.214Z