▶ Try it in your browser · no install, works on a phone

---

What this is

A shot timer for the Pixel Watch 4 and a companion phone app.

A shot timer is the basic instrument of practical shooting. It beeps to start you, listens for your gunshots, and reports how long each one took — your draw (time to the first shot) and your splits (time between consecutive shots). Competitive shooters live by these numbers.

Dedicated timers cost $110–190, are one more thing to carry, and sit clipped to your belt where you cannot see them. SPLIT puts the timer on your wrist, and adds one thing no dedicated timer can do: it cross-checks each bang against a recoil impulse from the watch's accelerometer, so the shooter in the next bay does not end up in your string.

The watch is the instrument and the system of record. The phone is a durable replica for review and analysis. Neither depends on the other to function.

Why a watch microphone can do this at all

This is the question that decides whether the project is possible, so it is worth stating plainly.

A 9mm at the shooter's ear runs 160–165 dB SPL.¹ The MEMS capsule in a watch clips at roughly 120–130 dB.² Every shot overdrives the microphone by 30–45 dB, so the amplitude coming off the converter is meaningless — it is simply railed.

That does not matter, because a shot timer does not need a level, it needs an arrival time.

Better still, the saturation is itself the most useful signal available. Your own muzzle rails the converter for milliseconds; a shooter two bays over arrives around 30 dB down and never rails at all. Gating on how long the signal stayed railed rejects neighbours in a way that a plain loudness threshold cannot.

Features

Sample-accurate timing	Onset resolution of one sample — 20.8 µs at 48 kHz — rather than one audio buffer (~10 ms).
True start instant	The clock starts when the tone physically leaves the speaker, not when a thread woke up.
AGC defeated	Pins the rawest microphone source the device exposes, so gain does not duck after the first round.
Neighbour rejection	Clip-run gating, plus an optional accelerometer recoil gate.
Nine drills	Bill, Failure to Stop, F.A.S.T., 1‑Reload‑1, Blake, El Presidente, Casino, Dot Torture, Freestyle.
Enforced standards	Shot count and par are enforced; the string auto-stops and grades itself.
Split analysis	Draw, every split, fastest, slowest, and split σ — the number that says whether you shot a cadence or got lucky once.
USPSA hit factor	Minor/major scoring with A/C/D/M/NS entry.
Works out of range	Completed strings are written on the watch first and sync to the phone whenever Bluetooth returns.
Haptic start	A wrist buzz cuts through hearing protection when a watch speaker does not.
Auto-repeat	Re-arms itself so you can run reps without touching the watch.

Requirements

To use the browser prototype: any modern browser. Microphone detection needs HTTPS, which the hosted link provides.

To build and install the apps:

JDK	17
Android SDK	Platform 36, build-tools 36.0.0
Watch	Wear OS 6 (API 36); minSdk is 30
Phone	Android 10+ (minSdk 29)
Gradle	Wrapper included — do not install Gradle separately

Note on compileSdk: this project pins compileSdk 36 and holds several dependencies back to match. Platform 37 is not published in the SDK manager yet. See Version pinning.

Quick start

The fastest path — no install

Open https://gptmadeit.github.io/split-shot-timer/ on your phone. Tap the face (or press space on a desktop) to log shots by hand, or grant microphone access to detect live fire. Add it to your home screen and it runs full-screen.

Install the apps

Download the APKs from the latest release:

adb -s <watch-serial> install -r split-wear-<version>-debug.apk
adb -s <phone-serial> install -r split-mobile-<version>-debug.apk

Run adb devices to list serials. The release APKs are debug-signed — fine for sideloading, not for Play distribution.

Build from source

git clone https://github.com/GPTmadeit/split-shot-timer.git

cd split-shot-timer && ./gradlew :wear:assembleDebug :mobile:assembleDebug

APKs land in wear/build/outputs/apk/debug/ and mobile/build/outputs/apk/debug/.

Usage

Your first string

1. Open SPLIT on the watch and grant microphone access. It must be granted while the app is open — that is an Android rule for microphone foreground services, not a choice made here.
2. Tap the drill name at the top to pick a drill. Start with Freestyle (no shot cap, no par).
3. Press START. The bezel fills brass and counts down a random 1–4 second delay so you cannot anticipate the beep.
4. On the tone, shoot. Each detected shot burns an orange tick into the bezel and updates the readout.
5. The string ends when you press STOP, or automatically on the last shot of a drill with a fixed count.

Reading the face

The bezel is a time tape. The acoustic envelope draws inward from the band; each shot burns a tick outward at its angular position in the string. By the end of a run the ring is the string — you read your cadence at a glance without parsing digits, which matters when your eyes belong on the target.

Under the clock: DRAW (time to first shot), SPLIT (live time since the last shot), and SHOTS (count, or count/target on a fixed drill).

Calibrating — do this before your first live string

Swipe to Settings and watch the level meter while the range is active around you.

1. Turn sensitivity down until ambient range noise stops lighting the meter past the brass marker.
2. Leave clip gate on. It requires the converter to actually rail, which is what rejects the bay next to you.
3. Lower echo blanking if you shoot splits faster than the current dead time. Indoor ranges need more blanking; outdoors you can go lower.

On the phone

Completed strings appear automatically. Tap a string to expand it for split bars and USPSA scoring. A rosette badge marks your best draw of the session, and the trend chart tracks draw time across strings.

If the watch is in range while you shoot, the phone mirrors the running string live. If it is not, nothing is lost — the strings arrive when the link returns.

Configuration

All settings live on the watch, under Settings.

Setting	Default	What it does
Sensitivity	−22 dBFS	Onset threshold. Lower = less sensitive.
Clip gate	On	Require the converter to rail. Rejects distant gunfire.
Recoil gate	Off	Require a matching wrist impulse. See the warning below.
Echo blanking	60 ms	Dead time after each shot, to reject reverb.
Start delay	Random 1–4 s	Also: instant, random 2–5 s, fixed 3 s.
Start signal	Beep + haptic	Haptic-only is useful in muffs.
Par tone	On	Second tone at the drill's par time.
Auto-repeat	Off	Re-arm automatically after 5/8/12 s.

Warning

The recoil gate ships off, deliberately. Recoil transfer depends on grip and on which wrist wears the watch. Support hand on a two-handed grip reads cleanly; strong hand reads harder; a one-handed string on the off hand may read nothing at all. Rimfire and heavily buffered PCCs may fall under the threshold entirely. Turn it on only after you have verified it against your own gun.

Troubleshooting

Shots are being missed. Raise sensitivity (a less negative dB value). If you shoot suppressed, turn the clip gate off — a suppressed host can sit below the microphone's clipping point, so there is no rail to detect.

More shots recorded than fired. Indoor reverb is being counted as extra shots. Raise echo blanking to 90 or 130 ms. If the very first "shot" is spurious, the start tone is leaking in — file an issue, that is a bug.

The neighbouring bay is registering. Turn the clip gate on, and lower sensitivity. If it persists and you shoot two-handed with the watch on your support wrist, try the recoil gate.

Nothing is detected at all. Check Settings shows a source and sample rate (e.g. UNPROCESSED @ 48000 Hz). If it says the mic is off, the permission was denied — reopen the app and grant it.

Strings are not reaching the phone. Both apps must be installed and the watch paired. Strings are held on the watch and sync when the link returns, so give it a moment after you are back in range. Nothing is lost in the meantime.

The browser page renders tiny, or the ring is blank. Reload it. If it persists, open an issue with your browser and OS.

Gradle fails with "requires Android Gradle plugin 9.x" or "compile against version 37". A dependency has been bumped past what this project pins. See Version pinning.

FAQ

Does this work on other Wear OS watches? It should build and run on any Wear OS 5+ device with a microphone (minSdk 30), but it has only ever been reasoned about for the Pixel Watch 4 and has not been run on any hardware at all.

Do I need the phone app? No. The watch app is standalone and declares itself as such. The phone app adds review, trend analysis and hit factor scoring.

Is it as accurate as a CED7000 or PACT timer? Unknown, and anyone claiming otherwise without measurements is guessing. The timing method here resolves to one audio sample, which is far finer than the 0.01 s these timers display. But resolution is not accuracy: end-to-end accuracy depends on hardware latencies that have not been measured. Treat it as unverified until it is characterised.

Why does it need microphone permission all the time? It does not. The microphone foreground service can only be started while the app is open, and it stops when you leave. There is no background listening.

Does it record or upload audio? No. Audio is analysed sample-by-sample in memory and discarded. Nothing is written to disk and there is no network code in either app.

Why is the release a pre-release? Because nothing has been tested on hardware. See Status.

Can I use this in a match? No. Use a certified timer. This is a practice tool.

Architecture

Three modules:

core/     models and the watch↔phone wire contract  (pure Kotlin, unit tested)
wear/     the instrument: mic, tone, clock, state machine
mobile/   the replica: receiver, log, analysis
docs/     browser prototype and diagrams

Two transports, deliberately different, because at a range your phone is in a bag on the bench and Bluetooth drops constantly:

• MessageClient → /split/live — low latency, best effort, silently dropped when the link is down. Mirrors a running string. Nothing the timer needs travels here.
• DataClient → /split/string/<id> — replicated and durable. Survives the phone being out of range for an entire session.

Full detail, including the timing method and the reasoning behind each detector constant, is in ARCHITECTURE.md.

Development

./gradlew build

That compiles both apps for debug and release, runs the unit tests, and runs Android Lint.

Task	Command
Unit tests	./gradlew testDebugUnitTest
Android Lint	./gradlew lintDebug
Assemble APKs	./gradlew :wear:assembleDebug :mobile:assembleDebug
Kotlin formatting	see CONTRIBUTING.md

Reports land in */build/reports/.

Version pinning — read before upgrading

Versions in gradle/libs.versions.toml are not "latest". They are the newest stable releases whose AAR metadata still permits compileSdk 36 / AGP 8.13.2. Anything newer gates on compileSdk 37, which is not published in the SDK manager yet. These were determined by reading aar-metadata.properties out of each artifact, not by guessing.

One deliberate exception: material3 is pinned to 1.5.0-alpha18. MaterialExpressiveTheme and MotionScheme exist in 1.4.0 stable but are compiled internal there, so the Material 3 Expressive API is unreachable. alpha18 is the newest build that exposes it publicly and still allows compileSdk 36.

Dependabot will open PRs that bump past these pins. Those PRs are expected to fail CI until platform 37 ships — that failure is the intended signal, not a broken pipeline.

Status and honest limitations

./gradlew build passes clean: both modules compile for debug and release, 42 unit tests pass, and Android Lint reports no issues on :wear or :mobile. Manifests are verified against the packaged APKs. CI runs all of this on every push.

Warning

Nothing has been run on real hardware, and no live fire has ever been recorded through it. The four detector constants are reasoned starting points, not measurements:

Constant	Value	Basis
Clip-run threshold	3 samples	estimate
Echo blanking	60 ms	estimate
Recoil window	±40 ms	estimate
Recoil threshold	18 m/s²	estimate

Expect to characterise all four against your own firearm and range. If you do, please tell us — that is the single most valuable contribution this project can receive.

Other known limitations:

• The browser prototype cannot truly pin an unprocessed audio source. Browsers apply AGC and noise suppression by default and the page can only request they be disabled.
• There are no instrumented (on-device) tests. Unit tests cover :core logic only; the audio, sensor and UI layers are untested by machine.
• Release APKs are debug-signed. Play distribution requires your own signing key.

Contributing

Contributions are welcome — especially real-world detection data. Start with CONTRIBUTING.md. By participating you agree to the Code of Conduct.

Security

Please do not open public issues for vulnerabilities. See SECURITY.md for private reporting. Both apps are offline by design: no network permission, no telemetry, no audio retention.

Support

Questions and setup help: SUPPORT.md.

Licence

MIT.

References

1. Gunshot sound pressure levels at the shooter's position — Wideners, Silencer Shop.
2. MEMS microphone acoustic overload point — Cirrus Logic AN0290, Infineon.

_{Not affiliated with Google. Pixel and Wear OS are trademarks of Google LLC.}