If possible, use 32-bit floating-point numbers for calculation in FGO. Here are several examples of how to to do it in several languages:

- C++: https://repl.it/@squaresmile/FGO-NP-gain-in-C
- Python: https://repl.it/@squaresmile/FGO-NP-gain-in-Python
- Javascript:

```
let f = (v) => Math.fround(v),
perhit = f(25),
cardBonus = f(f(1) + f(f(6) * f(f(1) + f(f(800) / f(1000))))),
finalEnemyMod = f(1),
gainBonus = f(f(1) + f(300) / f(1000)),
critBonus = f(2),
base_gain_no_round = f(f(f(f(perhit * cardBonus) * finalEnemyMod) * gainBonus) * critBonus);
console.log(Math.floor(base_gain_no_round));
console.log(base_gain_no_round);
console.log(cardBonus);
console.log(gainBonus);
```

The expected base NP gain output is 766. If 64-bit floating-point is used, the result will be 767.

If you start the calculation from the numbers the game stores (200, 300, ... which are what the API returns), the boundary between int and float are pretty clear. Buffs' values are summed up in ints and then divided by 1000f when applicable.

`!test melt m180 hp10000000 quick crit third af`

: 1243 per hit, 4972 total- Melt, third quick card, first arts card, 180% quick up + 8% quick up from passive, crit, no overkill hit

`!test abby m80 hp1 arts crit third ng30 af`

: 1149 per hit, 6894 total- Abby, third arts card, first arts card, 80% arts up, 30% NP Gain up, crit, all overkill hits