# Rounding


`RoundingMode` accepts the classic `BigDecimal` mode names:

| Mode | Behavior |
|------|----------|
| `UP` | Away from zero |
| `DOWN` | Toward zero (truncation) |
| `CEILING` | Toward +∞ |
| `FLOOR` | Toward −∞ |
| `HALF_UP` | Nearest, ties away from zero — **default**, matches PG `round()` |
| `HALF_DOWN` | Nearest, ties toward zero |
| `HALF_EVEN` | Nearest, ties to even digit (banker's rounding) |

## Where modes apply

```ts
round("12.345", 2);                       // 12.35 (HALF_UP default)
round("12.345", 2, "HALF_EVEN");          // 12.34 (banker's)
rescale("9.876", 1);                      // 9.9
setScale("9.876", 3, "DOWN");             // 9.876 — pads or truncates to exactly 3
div(a, b, { scale: 6, rounding: "FLOOR" });// floor division at scale 6
```

## Related transforms

- `trunc(d)` — integer part (toward zero)
- `floor(d)` / `ceil(d)` — integer toward −∞ / +∞
- `normalize(d)` — strip trailing zeros (`12.3400` → `12.34`; `v`/`scale`
  minimized). Note: `normalize` changes `scale`, so `strictEq` on a normalized
  and un-normalized equal value is `false` — use `eq` for value comparison.