Agent 04 · Monte Carlo + cross-entropy method
All amounts in today's dollars. Plans run to age 95.
Funded: savings compared with what the rest of the plan needs.
Monte Carlo
Instead of assuming a steady 7% a year, each future draws a random return for every year: mostly ordinary years, occasional crashes, bonds that wobble less. Running 2,000 lifetimes shows the whole range, from lucky to unlucky, and how often the money runs out.
Self-training
The planner's strategy is four numbers: how its stock share depends on age, how it depends on being ahead or behind, and how much it trims spending in bad times. Each round it tries 40 versions on the same 400 futures, keeps the best eight, and samples the next round around them.
Fair comparison
Every score shown comes from 2,000 futures the planner never trained on, so it cannot win by memorising lucky paths. It is a simplified model: no taxes, fees, pensions or health costs. Use it to build intuition, not as financial advice.