StrokerSync Update v4 — Signal & Connection Improvements
Sync & Responsiveness
Position extrapolation: Instead of sending where the penetration IS, the plugin now sends where it WILL BE at the end of the interpolation window. This lets the device glide smoothly across the full LinearCmd duration instead of staircase-jumping between positions. At 20 Hz, each command gets ~55ms of uninterrupted movement — dramatically smoother than the old approach of interrupting every 33ms with a new target.
Adaptive send rate: Command frequency now scales dynamically with motion speed. Fast strokes use the full configured rate; slow/idle movement backs off to 5 Hz, reducing unnecessary Bluetooth traffic. The "Send Rate" slider now sets the maximum rate. Default lowered to 20 Hz to complement extrapolation (fewer commands = longer glides = smoother).
Accumulator-based timing: Replaced the simple timestamp check with a time accumulator that carries leftover frame time forward. The average command rate now stays mathematically on-target regardless of frame rate fluctuations. Capped at 2x interval to prevent burst-sending after alt-tab or loading stalls.
Predictive LinearCmd duration: The device interpolation duration is now calculated from the actual time until the next expected command, plus a small user-configurable padding. Previously a fixed value, this caused the device to either sit idle between commands (at low rates) or never complete a movement (at high rates). The "Device Smoothness" slider is now "Duration Padding" (default 10ms).
Source-space deadband: Position deadband is now applied before stroke zone mapping. Previously, a compressed stroke zone (e.g. 0.3–0.7) inflated the effective deadband by up to 2.5x.
Penetration hysteresis: The "no penetration" threshold now uses hysteresis (stop at 0.005, resume at 0.02) to prevent the device from rapidly toggling on/off when the tip barely touches the target.
Tracking & Filtering
Rolling calibration: The penetration range now adapts continuously during play without manual intervention. New extremes expand the range instantly; if the observed range narrows (position change, different angle), the stored range slowly contracts over ~45–60 seconds. Pauses and cutscenes freeze the range. This eliminates the need to open the menu and recalibrate after changing positions. Enabled by default, can be toggled off for full manual control.
Conditional spike rejection: The median filter no longer runs unconditionally. During smooth motion, raw samples pass through with zero added latency. The median only engages when a sample deviates beyond a spike threshold, limiting the 1-frame latency penalty to actual physics glitches.
Asymmetric EMA: The noise filter now smooths direction-dependently. Increasing penetration depth gets 4x less smoothing for snappy attack feel; withdrawal uses full smoothing for softer release.
Pelvis-blended vaginal direction: Vaginal canal direction now blends 70% trigger-based vector with 30% pelvis bone axis. This stabilizes depth tracking during extreme poses where the labia/vagina triggers alone give a poor canal estimate. Falls back to trigger-only if no pelvis bone is found.
Connection & Reliability
WebSocket fragmented frame support: TinyWebSocket now handles continuation frames (opcode 0x0) per RFC 6455 §5.4. Control frames (ping/pong) are processed immediately even mid-fragment.
Callback TTL cleanup: Pending Buttplug protocol callbacks now expire after 10 seconds. A sweep runs every 5 seconds to prevent unbounded memory growth during long sessions if the server silently drops responses.
Sampled LinearCmd error logging: Every 100th position command now registers an error callback. This catches persistent device errors without flooding the callback system (~one check every 3 seconds at 30 Hz).
Clean client disposal on reconnect: The old ButtplugClient is now explicitly disconnected before creating a new one during auto-reconnect, preventing ghost receive threads from accumulating.
Component cache validation: Cached rigidbody/transform references are verified every 2 seconds. If an atom reload (clothing change, appearance preset) silently destroys a component, the cache is rebuilt immediately rather than waiting for a full null-out failure.
Added Hand tracking for jerking off or getting jerked off. Have not had time to test it 100% so it might be buggy.
