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

# Acceptance testing

> The repeatable checklist to run after any refactor, design change, or site reorg - what's automated (link/redirect/image/accessibility/encoding checks, a scripted page sweep), what needs a manual look, and the known blind spots of each.

This is the checklist to run after any change big enough to risk breaking something across the site - a folder move, a nav overhaul, a design change, or a pre-launch pass. It complements [Automated tests][2], which covers what runs automatically on every PR - everything here is run on demand, against a live deployment (the site or a PR preview).

## Automated checks

These run from the repository root with no browser needed. Several use the Vale CLI or `mint` CLI directly - both already used elsewhere in this repo.

| Check                            | Command                                                                                 | Covers                                                                                                                                                                                  |
| -------------------------------- | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Internal/external/redirect links | `mint broken-links --check-anchors --check-external --check-redirects --check-snippets` | Every link type the CLI supports, run in one pass instead of the per-PR add-on's internal-only default                                                                                  |
| Accessibility                    | `mint a11y`                                                                             | Color-contrast ratios against `docs.json`'s theme colors, and missing alt text on images/videos                                                                                         |
| Image references                 | `python tools/check-image-references.py`                                                | Every markdown/reference-style/HTML image reference resolves to a real file - catches the leftover-relative-path class of bug the root-absolute media convention should have eliminated |
| Encoding corruption              | `python tools/check-encoding.py`                                                        | Invalid UTF-8 and mojibake substrings (double-encoded accented characters, corrupted smart quotes) across every language                                                                |
| Language leakage (proxy)         | `python tools/check-vale-spelling-density.py --vale <path-to-vale.exe>`                 | Flags files whose per-language Hunspell spelling-alert density is a statistical outlier versus that language's own median - a real signal without adding a new dependency               |
| Redirects against a live site    | `python tools/check-redirects-live.py --base-url <site-url>`                            | Every non-wildcard `config/redirects.json` entry actually resolves, and lands on the expected page - not just that the JSON is well-formed                                              |
| Nav paths resolve                | `python tools/verify-nav-paths.py <config/nav-*.json>`                                  | Every page referenced in nav exists on disk (already used elsewhere, listed here for completeness)                                                                                      |

### Known limitations

The language-leakage check has a real blind spot: Danish, Norwegian, and Swedish share enough vocabulary that a paragraph of one leaking into another may not add many extra misspellings, since the words can still be valid-ish in the wrong dictionary. It works far better for a leak with little shared vocabulary, such as German into Dutch. Treat an outlier as worth a 30-second look, and pair it with the manual per-language skim below for the Nordic-language direction specifically.

`tools/check-redirects-live.py` can only test wildcard entries (about 8% of `config/redirects.json`) with a partial sanity check on the destination's own prefix, and only when that prefix is itself a real path (`/foo/*`, not a filename-prefix pattern like `/foo/bar-*`). A wildcard rule reported clean has not been proven correct for every possible match, only that its target namespace is alive.

## Scripted page sweep

`tools/benchmarks/smoke-test-pages.mjs` (reuses this folder's existing Playwright dependency - run `npm install` in `tools/benchmarks/` first if `node_modules` is missing) navigates a fixed set of pages against a live URL and writes one report instead of requiring an interactive session per page:

```sh theme={null}
SMOKE_BASE_URL=<site-url> RUN_LABEL=<label> node smoke-test-pages.mjs
```

For each page it captures console errors, failed network requests (any response 400 or above), and a screenshot at both a desktop and a narrower width. It also runs one keyboard tab-order trace, checking how many focusable elements are reachable and whether each has a visible focus outline - a check `mint a11y` does not perform.

Edit the page list at the top of the script to change what's covered - by default it's 1 landing and 1 deep page per language, a handful of content-type-diverse English pages, and any page identified as a candidate for a layout complaint (for example, a deeply nested reference section suspected of causing a cramped sidebar).

<Note>
  An embedded third-party video (for example, a YouTube `nocookie` embed) can show up as a failed request under headless Chromium even though it loads fine in a real browser - confirmed by curling the same URL directly. Don't take a single repeated failed request at face value. Check whether it is the same third-party embed on every affected page before treating it as a real break.
</Note>

## Manual review

Reading every screenshot the script produces defeats the point of scripting it. Only open the ones a judgment call actually needs:

* **Layout/sidebar candidates:** the screenshots for pages flagged as deep-nesting/long-label candidates, at both widths captured.

* **Per-language landing pages:** a quick skim for anything that reads as an obvious untranslated stub or broken layout. This is a backstop for the language-leakage blind spot above, not a substitute for it.

* **Search relevance:** run a fixed list of representative queries and read the returned titles; the automated check only confirms a query returns *some* results, not that they're the *right* results.

* **Keyboard tab order:** the script reports how many elements were reached and whether each had a visible outline; whether the *order* makes sense is a human judgment the trace alone cannot make.

## Reporting what you find

File anything confirmed real as its own GitHub issue rather than fixing it in the same pass that found it, unless the fix is genuinely trivial. See [Reviewing a PR][3] for the general review conventions this checklist feeds into.

## Related content

* [Mintlify deployment and CI/CD][4] - how to reach a live PR preview to test against.
* [Performance benchmarking][5] - the sibling `tools/benchmarks/` scripts this page's Playwright sweep reuses.

[2]: ./automated-tests

[3]: ./review/index

[4]: ./deployment

[5]: ./performance-benchmarking