Probability is not a schedule
A displayed 20% success rate does not promise one success every five attempts. Each attempt can still fail, and short sessions can land far above or below the long-run average. Treat the rate as a model input, not a countdown.
Expected attempts
When every attempt has the same independent success probability p, the long-run expected number of attempts is:
Expected attempts = 1 ÷ pExpected failures = (1 − p) ÷ p50% success rate
Expected attempts: 2
Expected failures: 1
20% success rate
Expected attempts: 5
Expected failures: 4
For several enhancement stages, the calculator evaluates every stage and adds the expected attempts and costs. It does not average all rates into one shortcut.
How failure correction changes the model
If the client increases the rate after each failure, attempts are no longer identically distributed. The calculator steps through the rising probabilities, caps them at 100%, and sums the chance of reaching each next attempt. This is sometimes called a pity or failstack-style mechanic, but the displayed in-game rule—not the nickname—must control the calculation.
Rate after n failures = min(100%, base rate + correction × n)When protection is worth comparing
Protection has value only relative to the loss it prevents. The calculator compares the expected unprotected recovery loss with the protection cost paid on attempts. A cheaper expected value is not automatically the best personal choice: limited materials, item replacement difficulty, and risk tolerance still matter.
Unprotected = attempt cost + expected failure recoveryProtected = attempt cost + protection costExpected cost versus a safer budget
The expected value is a long-run average. A median and a 90th-percentile reserve answer different questions; their order depends on the distribution, so do not assume the median is always higher than the average.
An illustrative 50% example
Assume a single stage, independent attempts, a constant 50% success chance, equal cost per attempt and no failure correction. This is a mathematical example, not an AION 2 rate.
| Measure | Attempts | Meaning under these assumptions |
|---|---|---|
| Median | 1 | Half of runs succeed on the first attempt. |
| Expected attempts | 2 | The long-run average across many runs. |
| 90th-percentile reserve | 4 | 93.75% of runs succeed within four attempts; 6.25% need more. |
Real costs and probabilities can change after failure or across stages. Use the calculator with your own inputs and verify its model before making a spending decision.
What the model cannot promise
The estimate is only as reliable as its inputs. It cannot detect an unpublished server-side rule, hidden cap, event bonus, rounding rule, or a Global value copied incorrectly from another region. That is why every preset shows its region, source status, and update date.
Start with an example, load a verified regional reference, or enter live-client values.