LocalType and Gboard: typos corrected in the editor
100 invented sentences, each with two planted typos and two clean words, typed by scripted touches into the same test editor after each keyboard’s own reset, scored only from the text that landed. Pixel 9a shown; the Pixel 10 was within two words in every cell and gave LocalType identical text.
| Condition | LocalType 0.5.14 | Gboard 18.4 beta |
|---|---|---|
| Exact key centers · native field | 133 (16 wrong) | 4 (0 wrong) |
| Exact key centers · web text box | 133 (16 wrong) | 91 (9 wrong) |
| 14 px tap spread · native | 134 (15 wrong) | 103 (9 wrong) |
| 23 px tap spread · native | 127 (20 wrong) | 105 (19 wrong) |
| 33 px tap spread · native, 87 sentences | 58% (101 of 174) | 55% (103 of 200) |
Clean words damaged, of 200: LocalType 0 to 1 and Gboard 1 to 2 at these spreads. At 33 px the paired difference is +0.03 [−0.05, +0.10], a tie. Gboard reaches the editor 1 to 4 ms sooner at the median. Next-word prediction has no controlled result either way. Scripted taps are not human typing; no speed advantage is claimed.
