> ## Documentation Index
> Fetch the complete documentation index at: https://docs.truemath.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Guarantees

> What deterministic execution gives you: reproducible results, full precision across every step, provenance for every value, and versioned business logic.

TrueMath exists to make math execution trustworthy. Five guarantees define what that
means in practice.

## Determinism and reproducibility

The same inputs produce the same outputs, every time. Results do not vary across runs
or drift as models update. A calculation is a function of its inputs and the
[version](/concepts/versioning) of the logic in effect — nothing else — so a result
can be reproduced rather than re-estimated.

## Full precision

TrueMath carries values at full internal precision through every step of a
calculation. Intermediate results are not rounded before being passed to the next
step; rounding is applied only when a value is displayed. This eliminates the drift
that accumulates when rounded values feed subsequent calculations — a failure mode
that is invisible in a single step and compounding across many. See
[Units and precision](/concepts/units-and-precision).

## Dimensional safety

TrueMath treats units like a strongly-typed language, not cosmetic labels. Every value
carries its unit, and an operation that is not dimensionally meaningful is rejected
rather than silently computed. Operations that *are* meaningful produce the correct
resulting unit: a length times a length is an area; a distance divided by a time is
a velocity. This keeps a calculation from returning a confident but dimensionally
nonsensical result. See [Units and precision](/concepts/units-and-precision).

## Provenance

Every variable carries metadata about where its value came from — whether it was
stated by the user, calculated, carried forward from a prior step, or supplied as a
default. Because each value's origin is tracked, a result is not just a number; it is
a number you can account for. See [Provenance](/concepts/provenance).

## Versioned business logic

[Domains](/concepts/domains), and the [activities](/concepts/activities) within them,
move through an explicit draft-to-published lifecycle. Calculations run against a
published version of the logic, so the rules that produced a result are explicit and
controlled rather than implicit and shifting. See [Versioning](/concepts/versioning).

<Note>
  These guarantees hold regardless of which language model interprets a request. The
  model can change; the execution layer's behavior does not.
</Note>