Accessibility

This section is the reason the page is worth publishing. Each item is real, we know about it, and none of them is presented here as fixed.

The site is published in thirty languages and has not been tested with a screen reader in all of them. Testing has been done in English. Pronunciation and reading order in right-to-left languages, and in scripts where our display typefaces have no coverage and the platform font substitutes, have not been verified with assistive technology.

Our two typefaces cover Latin only. Cyrillic, Devanagari, Georgian, Thai, Arabic, Hebrew, Bengali, Ethiopic and CJK all fall back to the platform's own fonts. That is generally good for legibility and it does mean the typographic detail we control in English — weight, spacing, the size at which small text stops being small enough — is not something we control in those locales.

Some interactive elements are recent and less tested. The comparison matrix and the help search filter rows in place, which is the right pattern, but the announcement of a changed result count to a screen reader has been implemented rather than user-tested.

The app is held to the same intent and is audited less often than the site. It supports the platform's own text sizing and screen reader, both of which do most of the work, but we do not currently run a formal accessibility audit against each release.

No formal third-party audit has been carried out on either the site or the app. We would rather say that than imply a certification we do not hold.

The Telvio app relies on the accessibility features of the operating system it runs on: VoiceOver and Dynamic Type on iOS, TalkBack and font scaling on Android. Those cover most of what matters in an app whose interface is a keypad, a balance and a list of rates.

One design decision is worth calling out as an accessibility property rather than a feature. Because the app requires no account, there is no sign-up form, no password to enter, no email verification step and no code to retrieve from a message — the whole class of interaction that most frequently defeats people using assistive technology, or anybody who finds forms difficult, simply is not present. Install, dial, talk.

The corresponding limitation is on the same page as everything else: the app cannot call emergency services, which is set out at emergency calls. If somebody sets this app up for a relative — a common pattern, particularly for older people calling family abroad — that is the single most important thing to explain while doing it.

Email [email protected] with the subject Accessibility. Useful detail: the page or screen, what you were trying to do, what assistive technology you were using including its version, and what happened instead. None of it is required — a sentence saying “I cannot use the rates table with a screen reader” is a perfectly good report and we would rather have it than nothing.

We aim to reply within five working days with an assessment and, where we agree there is a defect, a rough timescale. Things that block somebody from completing a task are fixed ahead of other work; things that make a task harder are queued honestly rather than promised vaguely.

If you are in the European Union and believe an accessibility obligation is not being met, you may also raise it with the relevant national enforcement body. We would rather you came to us first, and we will answer either way.

Two things are worth describing because they are the difference between a statement and a practice.

Contrast is painted, not parsed. A colour value read out of a stylesheet tells you what an element was asked to be, not what it ended up looking like — a semi-transparent element over a dark band, a modern colour function a regex does not understand, a fixed header over a hero image. The check renders each element's real background by walking up to the first opaque ancestor and resolving the colour through the browser itself, then computes the ratio. It found body text on this site sitting at 4.49:1, which is under AA and looks entirely normal in isolation.

The page is checked as rendered, not as written. A separate probe loads each page in a real browser and asks whether any text in the viewport is at zero opacity, whether the document scrolls sideways, and whether anything is painted in its own background colour. That last one sounds absurd until it ships: a single grouped CSS rule once set a label's colour and its background to the same value, producing a 1:1 contrast ratio that is invisible in the markup, invisible in the DOM, and invisible in a screenshot of the source.