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. One frozen run of both keyboards at the same pixel spreads, every cell complete. 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 |
|---|---|---|
| 14 px tap spread · native | 134 (15 wrong) | 103 (10 wrong) |
| 23.5 px tap spread · native | 127 (15 wrong) | 106 (21 wrong) |
| 33 px tap spread · native | 110 (26 wrong) | 100 (38 wrong) |
| 23.5 px tap spread · web text box | 125 (15 wrong) | 106 (21 wrong) |
Clean words changed, of 200: LocalType 0 and Gboard 0 to 1 through 23.5 px, and 4 against 5 at 33 px, where the paired difference in correction is +0.05 [−0.01, +0.11], a tie. Taps at exact key centers are not shown: Gboard changes behaviour under them, so they do not measure it. Next-word prediction is at parity: the strip offered the target before typing for 16 against 15 of 150 human-written targets, and potential savings were 48.7% against 46.3%. Latency is within a millisecond at the median on one phone and 3 ms in Gboard’s favour on the other. Scripted taps are not human typing; no speed advantage is claimed.
