# When NOT to use


## Non-finite or special float values

`NaN`, `Infinity`, `-Infinity` are **deliberately unsupported** — construction
throws `DecimalError` (`NON_FINITE`). If your domain needs them (statistics,
sensor streams, scientific computing), use `number` or a domain-specific type.

## Transcendental / irrational math

There is no `sqrt`, `log`, `sin`, `exp` — a `Decimal` is always a finite
rational base-10 number. Irrational results would require an infinite
expansion. Use `number` (or `Math.*`) for geometry, statistics, physics.

## Hot-loop performance on integers

`bigint` arithmetic is fast but still allocates per operation. For pure
integer counters, indexes, pagination, bitmasks — `number` (or `bigint`
directly) is cheaper. `Decimal` pays for exactness you may not need.

## Ratios that must stay rational forever

`div` truncates at a finite scale (PG-faithful default or explicit). If you
need to *keep* `1/3` as a symbolic rational for later exact arithmetic, you
need a rational-number library (numerator/denominator pairs) — `isPeriodic`
only tells you a truncation happened.

## Rule of thumb

- Values that must match `numeric` columns or money semantics → `Decimal`
- Counts, indexes, flags → `number`/`bigint`
- Scientific/statistical math → `number` + domain library