Elamur: **Mova** · [Parse](https://mova.elamur.ai/free/en/parse/) · [Dictionary](https://md.dictionary.elamur.ai/free/en/) · [Translate](https://md.translate.elamur.ai/free/en/) · [Research](https://md.research.elamur.ai/free/en/) · [MCP](https://mcp.elamur.ai/) · [Groupants](https://md.groupants.elamur.ai/free/en/)

[Mova](https://md.mova.elamur.ai/free/en/) · [this page in HTML](https://mova.elamur.ai/free/en/docs/tools/approx/)

[home](https://md.mova.elamur.ai/free/en/) / [docs](https://md.mova.elamur.ai/free/en/docs/) / [tools](https://md.mova.elamur.ai/free/en/docs/tools/) / mova-approx

# mova-approx

Standard-library-only contracts for experimental executable groupant surrogates. ErrorBudget admits only 1%, 2%, 5% and 10% maximum error, in that order. Exact outputs qualify at every tier. Unsupported caps and errors above the selected cap fail closed, including the next representable float above 10%. No production implementation is replaced by this crate.

Use relative_error(reference, approximation) for finite signed scalar values; zero reference requires exact zero output. Use observed_certificate for a nonempty collection of finite nonnegative error fractions. It records their maximum and sample count. Use analytic_certificate for a caller-established bound and nonblank scope. It checks the numeric contract, not the proof. Certificate fields are private and available through getters.

Three independent typed axes provide stable snake_case labels through as_str():

- MetricKind: scalar_relative, normalized_rank, objective_loss, probability_relative.
- MethodKind: quantized_numeric, certified_sample, relaxation_certificate, floating_normalization, interval_certificate.
- AssuranceKind: analytical, per_input_certificate, finite_observation.

Classification describes the caller's metric, mechanism and evidence scope. It does not certify a claim or convert finite observations into guarantees. The caller supplies the metric definition, domain and supporting evidence; a method name such as certified_sample alone establishes no assurance.

Run `cargo test -p mova-approx --lib`. The two measurement runners are the examples `groupants` and `groupants_next` (`cargo test --release -p mova-approx --example groupants`, then `cargo run --release -p mova-approx --example groupants`; the same for `groupants_next`); `cargo test -p mova-approx --all-targets` includes both. Their profiles are described in [executable Rust approximants](https://md.mova.elamur.ai/free/en/docs/repo-docs/mathematical-modeling/#executable-rust-approximants). Mathematical models, typed obernants and source binding live in `src/lib.rs.groupant/`. Registration alone does not establish validity.

The example `student_vidtvorant` runs the school student's executable vidtvorant (`world/src/knowledge.rs.groupant/student-vidtvorant.groupant/rust_approximant.rs`) on traces of `world`'s `student_probe` example; it meets no cap and registers no profile (see [The school](https://md.mova.elamur.ai/free/en/docs/repo-docs/school/), «Speed and the student's vidtvorant»).

- [← previous Native agent runner](https://md.mova.elamur.ai/free/en/docs/tools/agent-runner/)
- [next → bench — one row of fixed measurements](https://md.mova.elamur.ai/free/en/docs/tools/bench/)
- [↑ back to tools](https://md.mova.elamur.ai/free/en/docs/tools/)