← Entropy Atlas
Reproducible randomness
Randomized algorithms
Random choices can improve expected runtime, balance structures, or avoid pathological inputs.
Evaluation card
What “plenty” means here.
- Required property
- A generator appropriate to whether inputs are benign or adversarial, plus unbiased bounded sampling.
- If it fails
- Predictable choices let crafted input force worst-case behavior, or sampling bias breaks a proof assumption.
- Relevant attacker
- Sometimes none; sometimes a client can choose inputs after learning the random strategy.
- The Plenty Line
- Use reproducible PRNGs for ordinary analysis and a keyed or cryptographic strategy when adversarial prediction matters.
- Why more is not automatically better
- Generator strength beyond the threat model does not improve the algorithm’s mathematical bound.
Where it appears
Recognize the pattern.
- 01quicksort pivots
- 02skip lists
- 03randomized load balancing
The Plentropy rule
Protect the requirement, then stop.
Use reproducible PRNGs for ordinary analysis and a keyed or cryptographic strategy when adversarial prediction matters.
Once that claim has comfortable evidence and margin, improve the next limiting factor instead of worshipping a larger entropy number.