# Verification and limits

Read validation claims at the right evidence level and evaluate providers on your workload.

Canonical: https://searchsdk.dev/docs/reliability/verification
Markdown: https://searchsdk.dev/docs/reliability/verification.md

<a id="fixture-evidence"></a>

## What fixture validation covers

Deterministic fixtures check protocol mapping, invalid fields, unsafe URLs, filter preflight, errors, fallback, cancellation, and deadlines. Use the repository's latest validation record for the current suite and results.

Injected responses exercise expected contracts without API credentials. They do not establish account readiness, upstream availability, actual billing, or real retrieval usefulness.

<a id="evaluate"></a>

## Evaluate your own workload

- Record dated queries and required source types.
- Record provider settings, locale, filters, result count, and time budgets.
- Use a clear rubric for source usefulness and coverage, with human review when needed.
- Measure latency and observed failures independently from content quality.
- Keep reported usage separate from estimated total spend.

<a id="benchmarks"></a>

## Benchmark work is in development

> **No published ranking**
> The benchmark workspace is in development. No universal quality ranking or fabricated latency score is presented here.

- [Benchmark development status](https://searchsdk.dev/benchmarks)
- [Provider-specific boundaries](https://searchsdk.dev/docs/providers)