Cherry-pick isolated HeliBoard settings (content-audit findings)

[AsafMah]

Summary

Content-level audit (tree diff, not version numbers) of Helium314/HeliBoard main vs this fork found a set of isolated, low-coupling settings present upstream but absent here. These are the only clean cherry-pick candidates — the rest (theming/typeface refactor, foldable split, floating rework, emoji-search rework) are entangled with HeliBoard infra this fork's base predates, and the gesture-data-gathering subsystem is deliberately excluded (see the other issue).

Method: diff of PREF_* keys in Settings.java (HeliBoard 137 vs fork 174) + tree diff of latin/utils, settings/screens, assets/layouts.

Cheap pulls (each ~self-contained)

• Autocorrect confidence threshold — auto_correct_confidence
• Punctuation suggestions toggle — punctuation_suggestions
• Spell-checker suggestions — spellcheck_suggest
• Swipe down on toolbar to hide keyboard — toolbar_swipe_down_to_hide
• Toolbar-only with hardware keyboard — only_toolbar_with_hw_keyboard
• Touchpad / space-bar edge scroll — touchpad_edge_scroll (+ keyboard/TouchpadHandler.java)
• Key-gap / bottom-row scaling — key_gap_scale, bottom_row_scale (+ settings/preferences/KeyboardScalePreference.kt)
• HCESAR layout — assets/layouts/main/hcesar.json (single file, trivial)

Entangled — evaluate, likely skip

• Foldable split keyboard (split_keyboard_folded[_landscape] + FoldableUtils.kt)
• Floating-keyboard rework (floating_* prefs + FloatingKeyboardUtils.kt) — conflicts with the fork's existing floating keyboard; this is a replacement, not an add. Needs a decision before touching.
• Theming/typeface system (Theme.kt, Icons.kt, KeyboardTypeface.kt) — infra refactor, not a single feature. Pulling it is a project, not a cherry-pick.
• Emoji search rework (emoji/EmojiSearchActivity.kt + common/Emoji.kt) — fork already has a variant.

Notes / caveats

• Each pull MUST follow the 5-file settings pattern here (Settings.java / Defaults.kt / SettingsValues.java / strings.xml / a *Screen.kt entry) + add a SettingsContainerTest.kt case.
• Strategic ordering caveat: the priority direction is sending fixes up to LeanBitLab, not pulling features down piecemeal. Treat this as opportunistic, low-priority QoL.
• Each is small enough for a single-responsibility PR; do not batch.

[AsafMah]

Deep-dive results (changelogs + docs + strings + layouts + manifest + source, both directions)

Method: content-level diff of heliboard/main vs origin/main across 35 changelogs, full_description.txt, all Settings.java PREF keys, full source trees, assets/layouts/**, strings.xml name sets, AndroidManifest.xml, and build.gradle.kts. (Histories don't share recent commits, so no commit diff; versions ignored.)

Headline: the fork is substantially ahead of HeliBoard, not behind. Fork-only: 28 features, ~60 pref keys, 246 strings, 28 source files. HeliBoard-only worth-having is a short QoL list.

A — Worth pulling (cheap, isolated)

1. HCESAR layout — copy 1 file (assets/layouts/main/hcesar.json). Zero risk.
2. Toolbar swipe-down-to-hide (toolbar_swipe_down_to_hide) — ~3-4 files, no conflict.
3. Touchpad edge-scroll acceleration — port the EDGE_THRESHOLD_PERCENTAGE/TOUCHPAD_ACCELERATION_FACTOR logic from HeliBoard's TouchpadHandler.java into the fork's existing TouchpadView.java (don't adopt the class rename). 1-2 files.
4. Only-toolbar-with-HW-keyboard (only_toolbar_with_hw_keyboard) — ~3 files.
5. Bottom row scale (bottom_row_scale) — ~4-5 files; needs a slider widget decision (HeliBoard's KeyboardScalePreference.kt vs fork's existing SliderPreference).
6. Punctuation-suggestions customization (punctuation_suggestions) and 7. spellcheck suggestions toggle (spellcheck_suggest) — ~3 files each.
7. Key gap scale (key_gap_scale) — but the fork already has finer PREF_NARROW_KEY_GAPS (5 levels); likely redundant.
8. known_dict_hashes.txt asset (dict integrity) — 1 file; ChecksumCalculator.kt already shared.
9. Default emoji skin tone (emoji_skin_tone) — verify fork doesn't already have it.

B — Entangled / expensive (skip unless a stated goal)

• Full-screen Emoji Search Activity — needs KeyboardTypeface.kt + latin/utils/Icons.kt + latin/utils/Theme.kt + common/Emoji.kt + manifest + KeyboardSwitcher state (≥6 HeliBoard-only files). Fork already has the simpler inline EmojiSearchAdapter.kt. Not worth it.
• Foldable split keyboard (FoldableUtils.kt + split_keyboard_folded[_landscape]) — ~5-7 files; only valuable for foldable hardware.
• KeyboardTypeface / Icons / Theme (latin.utils) — the fork's settings/Icons.kt + settings/Theme.kt + CustomFontPreference.kt already cover these in a different package; pulling HeliBoard's would collide. No action.

C — Do NOT pull

• GestureData* subsystem — explicitly temporary NLnet research tool (// will be removed once the project is finished), phones home, conflicts with PREF_DISABLE_NETWORK + the offlinelite no-INTERNET contract. (Tracked separately in Roadmap: adopt NLnet open gesture recognizer as glide backend; build two-thumb on top #75.)
• HeliBoard's floating keyboard (FloatingKeyboardUtils.kt) — the fork's FloatingKeyboardManager.kt is architecturally superior (true TYPE_APPLICATION_OVERLAY window vs shrinking the IME view; HeliBoard's has a TODO about visual glitches). Do not regress.

Key gotchas

• auto_correct_confidence (HeliBoard) vs auto_correct_threshold (fork) — same slider, different key/semantics. Can't be silently swapped; would need an AppUpgrade.kt migration. Recommend keeping the fork's.
• Several "missing" items from the first shallow pass are actually fork-has-equivalent (emoji search, floating, icons/theme, key-gap) — net new HeliBoard value is small.

Full report retained internally (heliboard-gap.md). Recommend doing only A1-A4 if/when we touch this; everything else is redundant or not worth the coupling.

[AsafMah]

Closing — the recommended cheap pulls (HCESAR, touchpad edge-scroll, toolbar swipe-down-to-hide, only-toolbar-with-HW-keyboard) shipped in #79 (verified on-device). The remaining HeliBoard-only items are intentionally not pulled per the deep-dive: emoji-search / floating / icons-theme / key-gap are already better-or-equivalent in the fork, GestureData* is do-not-pull, and foldable/font are entangled and low-value. Nothing left worth taking.