01 / policy
Finish first
Never interrupt an ordinary job.
— completed valueLanguage · 语言 · 言語
English by default.
Same page, automatic translation. Text is sent to GTranslate / Google when translating; wording may vary from the original.
Weekly small experiment · 2026-W47
The urgent job is real. So is the half-finished work.
Give four policies the same action clock and the same urgent arrivals. Change interruption pressure, deadlines and the cost of saving a place. Completed value, missed urgency and discarded progress stay separate.
Loading the seeded interruption trace…
01 / policy
Never interrupt an ordinary job.
— completed value02 / policy
Discard progress and start urgency now.
— completed value03 / policy
Pay to save progress, then resume it.
— completed value04 / policy
A seeded random pre-emption baseline.
— completed valueTry this first
At the defaults, checkpoint-and-return averages 431.586 completed-value points across 128 seeds, versus 402.031 for drop-and-restart and 300.844 for finish-first. A free checkpoint rises to 449.773. At three save actions, it falls to 382.820 while drop-and-restart stays at 402.031.
Every tick buys one action. Ordinary jobs take nine actions. Urgent jobs take three and expire after seven ticks. Finish-first protects current progress but misses more urgent work. Restart reacts immediately but throws away ordinary progress. Checkpoint preserves that progress after paying the selected save cost.
The coin-flip policy is a simple random baseline. No policy sees future arrivals. All four receive the same trace, values and deadlines for a seed.
| Setting | Finish | Restart | Checkpoint | Coin flip |
|---|---|---|---|---|
| Default | 300.844 | 402.031 | 431.586 | 386.164 |
| 25% storm | 434.906 | 642.422 | 639.984 | 594.773 |
| Free save | 300.844 | 402.031 | 449.773 | 386.164 |
| Three-action save | 300.844 | 402.031 | 382.820 | 386.164 |
| Long deadline | 431.805 | 406.195 | 439.883 | 404.719 |
This is a deterministic single-desk queue, not an employee, language-model, biological, reliability or financial benchmark. Urgency labels are perfect; jobs have fixed work and fixed value; there is no learning, collaboration, rework, safety severity, money or human judgment.
All 1,152 matched worlds and 4,608 policy outcomes are retained across nine predeclared settings. No seed was removed. The empirical study is upcoming; this release reports scripted simulation results only.
Method & parameter guide ↓Source package ↓All batch results ↓Read the model
Previously: Work or Proof? · Explore the eight small worlds.