01 · REPRESENTATION MAP
What does each operator transmit?
02 · CROSSOVER MICROSCOPE
Inspect one recombination
Change the operator, parents, or seed. Coloured loci show what is retained directly; the trace explains how remaining values are filled.
Parent 1
Parent 2
↓
Offspring
03 · CLASSIC PROBLEMS
One permutation, different meaning
04 · LIVE BROWSER BENCHMARK
Which crossover fits which problem?
Run all operators with the same initial population, seed, evaluation budget, mutation, and selection. This is a teaching-sized experiment—not a publication claim.
Best-so-far convergencegeneration → · objective ↓
Final rankinglower is better
| # | Operator | Best objective | Gap | Time |
|---|---|---|---|---|
| Run the benchmark to compare. | ||||
Fairness: tournament selection, elitism, inversion mutation p=0.2, crossover p=0.9, and identical seeded initial populations. Lower objective is better.
05 · MEASURED FLOW-SHOP PILOT
TA001–TA005 · seed 42 · 40,000 evaluations
These are measured Python/pymoo runs, not browser simulation. RPD is relative to the best observed crossover for each instance.
Makespan
Total flow time
Teaching question: Why does PMX minimize average regret while OX wins more instances, and why is OBX especially competitive for flow time?
06 · TAKE THE LAB WITH YOU