Stewardship

Localization and Accessibility

Komms uses one versioned localization contract across desktop, Android, and iOS. English (en-US) is the required fallback and Icelandic (is) is the first complete non-English locale. Language choice is private presentation state: it cannot change identity, protocol bytes, cryptographic trust, delivery semantics, contacts, or history.

This document records implemented controls and their limits. It is not an external accessibility assessment and does not qualify a physical device.

1. Canonical localization contract

The shared catalog and shell-specific paired messages live in locales/. Generated Android resources, the desktop bundle, iOS JSON catalogs, Localizable.strings, and InfoPlist.strings are derived from that source.

The contract enforces:

stable lowercase identifiers and identical identifier sets;
identical text/plural shape and positional placeholder types;
English and Icelandic one/other plural selection;
NFC text with catalog bidi controls rejected;
first-strong isolation around inserted strings;
explicit rationale for intentionally untranslated technical values;
fail-closed unknown identifiers and malformed resources;
bounded translation expansion and complete generated-resource parity; and
strict source coverage for desktop HTML/JavaScript, Android source adapters, and SwiftUI/static iOS copy.

System, English, and Icelandic selection is available before unlock and in settings on all three shells. Unsupported system languages fall back to English. The override is stored only as local appearance state.

Run the bounded localization profile:

python3 scripts/contributor-check.py localization

Or run its individual controls:

python3 scripts/localization.py check
python3 scripts/check-localization-sources.py
python3 scripts/test-localization.py
python3 scripts/check-shell-accessibility.py

2. Accessibility surface contract

Surface	Implemented shell contract
Onboarding and modes	semantic headings and labels, described mode controls, scalable text, pre-unlock language choice, and visible security disclosures
Message requests and invitations	separate list semantics, labelled safety/name controls, 44-point actions, explicit Accept/Delete/Block/Join behavior, focus recovery, and polite status announcements
Conversations and groups	conversation-log semantics, labelled composer/actions, keyboard-operable mention and organization controls, authenticated-origin status, poll state and vote labels, and non-color delivery cues
Attachments and audio	labelled progress, explicit consent and terminal view-once language, protected preview descriptions, non-autoplay audio, and accessible playback controls
Calls	labelled availability/action state and status announcements without changing delivery semantics
Backup, recovery, and devices	secure-field classification, alert semantics, labelled comparison codes and QR frames, conflict announcements, and explicit irreversible-action copy
Settings and organization	natural focus order, keyboard navigation on desktop, labelled folders/labels/icons, scalable platform controls, and immediate language/theme feedback

Desktop controls have a 44 CSS-pixel minimum target, visible focus rings, dialog focus containment and restoration, arrow-key navigation for composite controls, a higher-contrast media profile, and a reduced-motion profile. Android uses scalable sp, theme-level 48/52 dp targets, live regions, headings, label relationships, and explicit announcements. iOS uses Dynamic Type styles, native focus order, 44-point custom controls, combined semantic groups, update traits, and no nonessential custom animation.

The automated contrast gate evaluates normal-text role pairs at WCAG AA 4.5:1 or greater in both light and dark palettes. Color remains supplemental: delivery, warning, selection, and security state also have text, symbols, traits, or borders.

3. Evidence boundary

Automated source and simulator checks can prove that required labels, focus logic, target sizes, localization resources, plural rules, and palette relationships are present. They cannot prove VoiceOver/TalkBack usability, switch-control behavior, motor accessibility, real-device font clipping, platform magnification, cognitive clarity, or the experience of disabled users.

The following remain open:

named physical-device runs at the largest supported text sizes;
complete VoiceOver, TalkBack, keyboard, switch-control, contrast, reduced motion, and screen-magnification journeys;
localization review by a fluent Icelandic reviewer; and
an independent accessibility assessment with retained findings and retest dispositions.

Simulator observations stay labelled as simulator evidence in the field-qualification matrix. No accessibility row becomes externally reviewed or field-qualified without the named execution.

Edit this page on GitHub ↗