Repository navigation
Guard local rewrites and document F-Droid limits - #2
Merged
Merged
Conversation
Gr33nOps
marked this pull request as ready for review
October 27, 2026 10:20
|
Strix is installed on this repository, but we couldn't run this PR security review because this workspace's trial has ended. Add a card to resume code reviews here. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The F-Droid device review found that a German local rewrite became English and changed
13 uhrto13:00. This branch gives local models source-language cues, rejects clear translation and number/script drift before saving, and retries some failed rewrites in the source language. Mixed-language drafts can retry by passage. It also rejects a newly observed English rewrite that silently drops an explicit missed-meeting constraint, ignores silence-only dictation, and asks users to compare each result with the original before sending.Local models are best effort. The app and release criteria do not promise perfect meaning preservation, equal quality in every language, strong tone changes, or a speed guarantee. The phone tests show why: Gemma and Qwen kept the reviewer's German sample in German, but a Japanese Gemma output changed who could attend a meeting; English Casual and Professional were sometimes indistinguishable; Qwen and Llama dropped facts in other runs. The app rejected several of those outputs, but its checks are not a general semantic verifier. Llama 3B took roughly two minutes and also lost meaning in one run.
On Redmi 12 / Android 15, the isolated QA app completed encrypted backup export and settings-only restore, local WebDAV sync and conflict review, and a mock OpenAI-compatible cloud stream. Spoken dictation captured and inserted the user's words but heard “Thursday” as “Lutersday.” Android registered the selected-text handler, but this phone's selection menu did not show Rescripto, so return-to-caller remains unverified. Real cloud and Nextcloud accounts were not used. See
docs/FDROID_DEVICE_RETEST.mdandtesting/mobile/reports/2026-10-25_fdroid_language_retest.mdfor exact outputs and limits.Validation: Flutter 3.44.9 reports 476 passing tests and clean analysis; an Android arm64 release-style APK builds. This PR remains draft. No release tag, signed APK, reproducibility comparison, or F-Droid recipe update has been published.
Manual Release workflow preflight passed on commit 54b08b2: run 40485716462. Its unsigned APK artifact is available; no signing or publication occurred.