Web port had the task semantically wrong (framed as lexical decision /
real-word judgment). Unity VisualOrtho is a press-and-hold same/different
visual orthographic comparison on 5-character consonant strings
(B D F G K L N S T) plus a symbol phase. Reauthored from scratch.

Trial mechanic: PointDown -> fixation -> stimuli -> PointUp -> answer
panel (= / !=). All five Unity outcome states recorded
(not_recorded_too_early, not_recorded_no_start, triggered_softtimer,
correct, incorrect).

Stimuli: shipped Forms A/B/C verbatim from VisualOrthoManager.cs
(_configs_A/B/C_letters and _symbols, 40 trials each: 20 SAME / 10 TL /
10 SUB) plus the 8-trial _configs_unaided practice. Sequences are
frozen as Unity plays them, not regenerated from
VisualOrthoGenerator.

Symbol phase: ships the bespoke target_<L>_symbol.png sprite atlas from
Assets/Sprites/VisualOrtho/ for B/D/F/G/K/L/N/S/T (CSS filter:invert
to render white on dark). Replaces the scaleX(-1) rotate(180deg) hack.

Audio: 120 OGGs (DE/FR/IT, no EN VO in Unity) copied verbatim from
Assets/Addressables/VoiceOver/VisualOrtho/. Per-screen Replay button.
Practice feedback plays aided_right/wrong/tooLate/tooEarly/noStart.

Localisation: 4 locales (FR/IT/DE/EN). Strings are verbatim Unity I2
under _VisualOrtho/visualortho_*; cells empty in DE / EN I2 are omitted
from the locale dict and t() falls back to the verbatim FR string at
runtime, mirroring Unity's I2 culture-chain.

Tutorial condensed to single instructions screen + 8-trial unaided
practice with feedback (no Auto/Aided/criterion-retry).