Swipe Typing Technical Specification

Overview

Swipe typing routes each completed gesture to one of two decode engines — the CTC trie-beam engine (default) or the geometric (SHARK2-style) engine — selected by the swipe_engine_mode preference plus the active layout and language. Both engines feed the same downstream suggestion pipeline.

The ONNX transformer (“neural”) engine that used to be the default was removed on 2026-08-18. CTC replaced it on measured accuracy (89.31 vs 74.62 top-1 on test-2400, docs/history/audits/2026-08-17-neural-vs-ctc-parity.md) while geometric covers every cell CTC does not serve. The archived spec is docs/history/neural-engine/neural-prediction-spec.md.

Key Components

Component	File	Purpose
Engine Router	swipe/SwipeEngineRouter.kt	Mode + layout → engine selection (Mode.CTC/GEOMETRIC)
CTC Adapter	swipe/CtcEngineAdapter.kt	CTC engine boundary: layout/trie/session memos, contraction display, warm-up
CTC Core	swipe/ctc/ (CtcBeamDecoder.kt, CtcFeaturizer.kt, CtcLexiconTrie.kt, …)	Pure-JVM CTC Viterbi trie beam over ONNX emissions
ONNX Session Loader	onnx/ModelLoader.kt	Builds the CTC encoder’s OrtSession (XNNPACK-first)
Geometric Adapter	swipe/GeometricEngineAdapter.kt	Geometric engine boundary
Geometric Core	swipe/geometric/	SHARK2-style shape decoder
Keyboard Grid	KeyboardGrid.kt	Map coordinates to keys

Architecture

Touch Events (Pointers.kt)
    ↓
InputCoordinator.handleSwipeTyping
    ↓ SwipeEngineRouter.route(layout, mode)
    ├─ CTC       → CtcEngineAdapter (en/fr/de/es/it/pt/sv; other languages fall through to geometric)
    └─ GEOMETRIC → GeometricEngineAdapter
    ↓ (top-k candidates, engine-relative scores)
SuggestionHandler.handleSwipePredictionResults → UI

The router is TOTAL: every layout resolves to an engine, so no configuration can leave a layout without swipe typing.

Engine Routing (swipe_engine_mode)

The router (swipe/SwipeEngineRouter.kt) is layout-only; the ctc mode’s language gate lives in InputCoordinator.performCtcSwipeTyping, which reads the active dictionary language before dispatch and falls through to the geometric engine for anything CTC does not serve:

Mode	Latin layout + served language	Latin layout + other language	Non-Latin layout
ctc (default)	CTC	GEOMETRIC	GEOMETRIC
geometric	GEOMETRIC	GEOMETRIC	GEOMETRIC

Served languages are swipe/ctc/CtcLanguageSupport.SUPPORTED — seven: en, fr, de, es plus it, pt, sv. The last three are in CtcLanguageSupport.PROVISIONAL: they were enabled on 2026-08-18 on scale-transferred evidence (they read the same CKDT .bin frequency scale the λ sweep fitted, and the encoder never sees a language) and have no per-language accuracy bar, because no swipe corpus exists for them. Their numbers are val-tier at best and must never be quoted beside the test-validated four. A Latin layout missing an a–z letter cannot build a CtcLayout and also falls through to geometric, checked at dispatch time by CtcEngineAdapter.supportsLayout.

Mode.fromPref maps any unrecognised stored value — including the removed "neural" and "hybrid" — onto CTC, so a pre-v1.6.0 backup imports without error.

One engine owns each swipe end-to-end; scores are engine-relative and never compared across engines. Suggestion provenance tags the engine that actually decoded (SuggestionProvenance.forRoutedEngine). The CTC engine maps contraction aliases to display forms (“dont” → “don’t”) inside its adapter before the shared pipeline. Full CTC engine internals: docs/specs/ctc-swipe-engine.md (engineering spec).

Gesture Sampling Robustness

There is no minimum-speed gate on swipe typing — a slow swipe is not rejected for being slow. Word activation depends on registering ≥2 keys plus swipe_min_distance of path, which a slow-but-complete swipe satisfies just like a fast one (path length is bounded by key geometry, not by speed).

ImprovedSwipeGestureRecognizer.addPoint does, however, drop samples whose inter-sample gap exceeds MAX_POINT_INTERVAL_MS (500 ms). To keep a mid-gesture pause (a deliberate swiper holding still to aim) from permanently stalling the gesture, the recognizer re-anchors _lastPointTime to the resume timestamp when a long gap is seen, then resumes on the next sample. Without this re-anchor a single >500 ms gap left _lastPointTime stale, so every later sample’s delta grew larger and the remainder of the swipe was dropped — the second key never registered and no word was produced.

Engine Configuration

From Config.kt:

Setting	Key	Default	Range	Source
Prediction Engine	swipe_engine_mode	"ctc"	ctc / geometric (case-canonicalized at read; anything else resolves to ctc)	Config.kt (SWIPE_ENGINE_MODE), swipe/SwipeEngineRouter.kt
CTC Beam Width	ctc_beam_width	100	10-300 (clamped at load and per decode)	Config.kt (CTC_BEAM_WIDTH), CtcSettingsActivity.kt
ONNX Threads	onnx_xnnpack_threads	2	1-8	Config.kt (ONNX_XNNPACK_THREADS), CtcSettingsActivity.kt
ONNX Model	bundled asset	—	—	src/main/assets/models/ctc_swipe_encoder.onnx (2.91 MB)

The CTC scoring constants (gamma/lambda/beta/prune) are CtcScoringParams.tunedV2, fitted offline against the shipped lexicon, and are deliberately not user-tunable.

Geometric engine knobs live in GeometricSettingsActivity (geo_max_results, geo_frequency_weight, geo_endpoint_inset_kw).

Related Specifications

• CTC Swipe Engine - Deeper architectural reference for the shipping decoder (trie beam, lexicon merge, per-language λ)
• Gesture System Overview - Touch event routing and hasLeftStartingKey gatekeeper
• Autocorrect Specification