GAME DESIGN / REWARD REGRESSION
Test the unlucky tail.
Keep the promise.
A 1% drop rate is not a guarantee in 100 attempts.
Compare reward curves and make your design targets testable.
Load the example or edit both models. Calculations describe the next first reward, conditional on the misses entered.
01 / BASELINE
The current design
02 / CANDIDATE
The proposed change
Soft pity adds the step on the first boosted attempt, then again on each later miss. A guaranteed attempt takes priority. Enter 0 to disable soft or hard pity.
Candidate acceptance checks
Use success ≥ a target, or still missing ≤ a limit. Tests use log probabilities so rounding cannot turn tiny odds into a guarantee.
| Check name | By attempt | Rule | Threshold (%) |
|---|
Free edition: two-model analysis and curve CSV export up to 1,000 attempts. The full edition adds saved projects, acceptance checks, shareable reports and the Node.js regression runner.
Who is still waiting?
| Measure | Baseline | Candidate |
|---|
Inspect a particular attempt
P50 / P95 / P99 are the first future attempt at which that fraction has acquired the reward. “Beyond horizon” is not an estimate. A capped mean counts an unsuccessful player as having spent the whole horizon; it is not the average among winners.