## 2026.10.2

### Highlights

- Dictating no longer wipes the clipboard. (#153)
- The window connects to the engine in about a second instead of waiting for the speech model to load, and shows "Getting ready…" until it can dictate. (#158)
- Decoding starts the moment the shortcut is released, instead of up to 100 ms later. (#160)
- Each dictation adds one `Dictation timing:` line to the log, with how long each step took and never the text. (#159)

### Changes

- Each dictation adds one `Dictation timing:` line to the log, with how long each step took and never the text. The status line now reads "Transcribing N s of audio", because the old wording looked like a processing time. (#159)

### Fixes

- Dictating no longer wipes the clipboard. Dictate pastes through the clipboard, then puts back exactly what was there before, once the app has read the dictation: text, images, file lists and rich text on Windows, and the copied item on Linux. If you copy something yourself mid-dictation, Dictate leaves it alone. On Windows, dictated text stays out of Win+V clipboard history and cloud clipboard sync. (#153)
- The window connects to the engine in about a second instead of waiting for the speech model to load, and shows "Getting ready…" until it can dictate. A model that fails to load is shown in the window instead of the engine exiting. (#158)
- Decoding starts the moment the shortcut is released, instead of up to 100 ms later. (#160)

### Complete contribution record

This audited record covers the complete 20386bdb17364f196858c3ab26e654de5f2e43e7..2d6aa476ccee05cdf7f066ec876a2880b78dadc0 history: 5 in-range PRs + 0 retained seed-only PRs = 5 unique PRs. The generation manifest also supplies direct commits as editorial input; the grouped notes above prioritize user impact.

#### Pull requests

- **PR #153** Related #149.
- **PR #159**
- **PR #160**
- **PR #158** Related #155.
- **PR #161**