Dog Brain

Technical analysis

A living reference to the phones and watches Dog Brain is built and tested on, and what we have found out about each that their documentation does not say. It grows as more devices are profiled.

How to read it

  • Each fact is marked measured (read off the device, or computed from what it logged), estimated (worked out from measurements and assumptions, which are named), or vendor-stated (a figure the maker or a spec page gives).
  • Confidence is ours: high, medium or low.
  • Measured facts carry the date and the firmware build they were measured on, so a later firmware can explain a difference. "unknown" means the build was not recorded at the time.
  • Sources are this repository: the dev journal by day, files by path, commits by hash. Pages are linked, with the date they were read.
  • Devices are named by model and role (dev phone, test watch), never by serial.

The devices

DeviceModelRoleFacts
Galaxy S22SM-S901Wdev phone13
Galaxy S24 UltraSM-S928Wtest phone15
Galaxy Watch UltraSM-L705Ftest watch34
Galaxy Watch UltraSM-L705Udev watch14
Galaxy Watch (SM-R880)SM-R880second test watch ("Doe watch")6

Across devices

Per class, over measured facts only. A mean once 2 devices share a measure; outliers by the modified z-score (|0.6745 (x − median) / MAD| > 3.5, NIST e-Handbook 1.3.5.17) once 5 do. Under that, n is too small to call an outlier.

Phones

MeasurenMeanRangeDevicesOutliers
display.density_dpi2540 dpi480 to 600SM-S901W 480, SM-S928W 600n too small (needs 5)

Measured on one phone only so far: battery.counter_step_uah, drain.awake_still_ma, drain.idle_screen_off_ma, drain.screen_mostly_on_mah_per_h, motor.draw_at_255_ma, motor.shortest_felt_gap_ms, sensors.accel_max_hz, timing.buzz_fire_sd_ms, timing.end_buzz_after_fire_ms.

Watches

MeasurenMeanRangeDevicesOutliers
battery.counter_full_mah2285 mAh284 to 286SM-L705F 284, SM-L705U 286n too small (needs 5)
battery.os_estimated_mah3465 mAh240 to 578SM-L705F 578, SM-L705U 578, SM-R880 240n too small (needs 5)

Measured on one watch only so far: battery.full_charge_wh_at_puck, battery.learned_cell_mah, drain.awake_ma, drain.doze_display_off_ma, drain.mixed_use_pct_per_h, drain.screen_on_3d_ma, drain.screen_on_heavy_ma, motor.shortest_felt_gap_ms, reporting.counter_step_uah, sensors.accel_max_hz, timing.end_buzz_after_fire_ms, timing.sound_lead_ms.

Galaxy S22 SM-S901W · dev phone

The dev phone since 2026-11-17, paired with the SM-L705U dev watch. Free for any test input. Most of what is known about it is how its battery reports to apps and to the shell, found while building Sniffer's current readers.

Regional variantSM-S901W (the model Sniffer names Dev phone)
Firmware buildS901WVLSAGZH3 (getprop ro.build.PDA, read by the laptop session 2026-11-21)
One UIOne UI 8.0 (ours: getprop ro.build.version.oneui reads 80000, which we take as 8.0.0; read 2026-11-21)
OSAndroid 16 (SDK 36) (read by the laptop session 2026-11-21)
SoCSM8450 (getprop ro.soc.model; SM8450 is Qualcomm's part number, so this S22 is the Snapdragon 8 Gen 1 one (ours, from the part number))

Battery capacity

  • Rated capacity 3700 mAh. vendor-stated · medium confidence

    How: Wikipedia's infobox for the Galaxy S22 range, read 2026-11-21. Samsung's own page was not read. The page: “S22: 3700 mAh”

    Source: en.wikipedia.org/wiki/Samsung_Galaxy_S22 (read 2026-11-21)

Drain by state

  • Screen off and dozing, the phone drew 80 mA over 15 minutes (6 counter steps), with adb over Wi-Fi polling every 15 s. measured · medium confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: Charge-counter drop over the idle phase. One step is ±13.6 mA over 15 min, and the adb polling is in the figure.

    Source: design/research/motor/README.md

  • Awake with nothing running (a vibration's wakelock, motor at 0) the phone drew about 105 mA, about 25 mA more than dozing. measured · low confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: Mean of the square wave's off halves; the doze comparison is the charge-counter idle phase, not the sampler, whose doze rows lean to busy moments.

    Source: design/research/motor/README.md

How the battery reports

  • BatteryManager's CURRENT_NOW comes in milliamps, not the microamps Android documents: −241 on battery. measured · high confidence · 2026-11-17, firmware unknown: not recorded

    How: The laptop session's reads; Sniffer's ChargeNow tells the scale from the size of the number (MA_MAX).

    Source: phone-sniffer/app/src/main/java/com/dogbrain/sniffer/phone/ChargeNow.kt

  • The battery voltage extra (EXTRA_VOLTAGE), which Android's page gives no unit for, comes in millivolts: 4095 at the read with −241 mA. measured · high confidence · 2026-11-17, firmware unknown: not recorded

    How: dumpsys battery over adb, the laptop session.

    Source: phone-sniffer/app/src/main/java/com/dogbrain/sniffer/phone/ChargeNow.kt

  • The shell cannot read the battery's sysfs files: /sys/class/power_supply/battery/current_now and charge_counter answer "Permission denied". measured · high confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: adb shell cat, the laptop session. So current is read by an app (Sniffer's debug CurrentSampler) through BatteryManager instead.

    Source: dev journal, 2026-11-21

  • dumpsys battery's "current now" is not live: it holds one value for about 10 s. measured · medium confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: Repeated dumpsys reads by the laptop session. How the 10 s was timed is not recorded.

    Source: dev journal, 2026-11-21

  • Through Sniffer's debug CurrentSampler (BatteryManager), CURRENT_NOW is live, not held: over 60 s idle it changed in 135 of 146 intervals. Dozing, rows come 250 ms to 1.6 s apart, so 147 rows, not 240; while a vibration holds a wakelock the sampler keeps 4 Hz. measured · high confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: The 60 s idle run and a 5 s buzz at 255, by the laptop session.

    Source: design/research/motor/README.md

  • The charge counter moves in steps of 3370 µAh: every change in a 47-minute log is a multiple of it. measured · high confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: dumpsys battery every 15 s through three 15-minute phases; 24 changes, all multiples (ours, counted from the CSV).

    Source: design/research/motor/README.md

The motor

  • The vibration motor draws about 104 mA at amplitude 64, 104 at 128, 107 at 192 and 132 at 255: not linear in amplitude. Most of the cost is running it at all. measured · medium confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: On minus off inside one vibration of 30 × (1 s on, 1 s off), 3 shuffled repeats per amplitude (sd 3 to 10 mA); one 128 trial left out because the screen woke during it.

    Source: design/research/motor/README.md

  • cmd vibrator_manager synced oneshot blocks until the vibration ends, and a shell vibration is logged with usage COMMUNICATION_REQUEST; it still played with the ringer muted. measured · high confidence · 2026-11-21, firmware S901WVLSAGZH3

    How: The laptop session's rig and the vibrator's own history (dumpsys vibrator_manager).

    Source: design/research/motor/README.md

The display

  • 6.1 in, 2340 × 1080, 120 Hz. vendor-stated · medium confidence

    How: Wikipedia's infobox, read 2026-11-21. The page: “S22: 6.1 in (150 mm) 2340 × 1080 (425 ppi), 19.5:9 aspect ratio, 120 Hz refresh rate”

    Source: en.wikipedia.org/wiki/Samsung_Galaxy_S22 (read 2026-11-21)

  • Apps see a 360 dp wide window: 1080 px at density 480 (3 px a dp), font scale 1.0. Dog Brain's screens were first laid out at 412 dp, so the S22 is the narrow case. measured · high confidence · 2026-11-14, firmware unknown: not recorded

    How: Read over adb during the screen-by-screen walk (tools/s22-walk.sh).

    Source: dev journal, 2026-11-14; phone/app/src/test/java/com/dogbrain/phone/ui/screens/RoboSupport.kt

Not known yet

  • The real alarm patterns' cost (the fanfare, ScoreBuzz, FinalPulse): not run.
  • Drain with the screen on, and in always-on: nothing measured on this phone.
  • The motor's frequency control: only amplitude was swept.
  • Alarm delivery under Doze: the alarm soak (docs/TEST_alarm_soak.md) has not been run.
  • Calibration against the dev watch: no pair run with this phone is in the repo, so the S24 pair's 47.5 ms sound lead is untested here.
  • What woke the screen twice during the motor runs.

Galaxy S24 Ultra SM-S928W · test phone

The first phone Dog Brain ran on, and the dev phone until 2026-11-17, when it and its watch (the SM-L705F) became the test pair. All of the calibration study's 200 runs are from this phone and that watch, so most timing and motor findings are from this pair.

Regional variantSM-S928W (the model Sniffer names Test phone)
Firmware buildnot known: no build has been read off this phone into the repo
One UInot known: not read off the phone; it shipped with One UI 6.1 (Wikipedia)
OSnot known: not read off the phone
SoCQualcomm Snapdragon 8 Gen 3 (every S24 Ultra, by Wikipedia; not read off the phone)

Battery capacity

  • Rated capacity 5000 mAh. vendor-stated · medium confidence

    How: Wikipedia's infobox for the S24 range, read 2026-11-21. Samsung's own page was not read. The page: “S24 Ultra: 5000 mAh”

    Source: en.wikipedia.org/wiki/Samsung_Galaxy_S24 (read 2026-11-21)

  • On a wireless pad, 46 to 54% of the energy in reaches the battery. estimated · low confidence

    How: Given in the battery research as the yardstick for the watch's own puck efficiency. How it was measured is not recorded there.

    Source: docs/research/battery-capacity-2026-11-17.md

Drain by state

  • Whole-device drain of 303 to 416 mAh an hour on battery in the first two days' test sessions, with the screen on for most of each window (3.6 of 4.1 h, 9.3 of 10.2 h, 19.0 of 21.0 h). Dog Brain's own share was about 1% of it. measured · medium confidence · 2026-11-11, firmware unknown: not recorded

    How: Android's batterystats, read by tools/profile.sh after each install (docs/PROFILE.md). The value given is the 21 h window. These are Android's model figures, not a meter's.

    Source: docs/PROFILE.md

The motor

  • The motor reports amplitude control and plays every buzz asked of it: 70 vibrations in a Time's up, all finished, none cancelled, at scale 1.00. Samsung did not trim the short one-shots. measured · high confidence · 2026-11-14, firmware unknown: not recorded

    How: dumpsys vibrator history over adb during a Time's up, the laptop session; matched against the buzz file.

    Source: dev journal, 2026-11-14

  • No frequency control: no frequency profile, envelope size 0. measured · high confidence · 2026-11-11, firmware unknown: not recorded

    How: Vibrator capabilities read by the calibration code.

    Source: dev journal, 2026-11-11

  • With the watch lying on it, the phone feels its own motor 6 to 10 times stronger than the watch's. measured · medium confidence · 2026-11-11, firmware unknown: not recorded

    How: Peak swing at each buzz onset, both accelerometers, calibration's strength measure.

    Source: dev journal, 2026-11-11

  • The shortest pause between buzzes that is felt as a pause is about 66 ms on the phone (38 to 96 ms on the watch). Buzz patterns now keep every gap at least 100 ms. measured · medium confidence · 2026-11-12, firmware unknown: not recorded

    How: Calibration's fall and rise of the same motor.

    Source: dev journal, 2026-11-12

  • Felt on the hand, the phone's buzz under the Highland March was not felt at all, though the file had the same 88 kicks the watch played. Why is open. measured · low confidence · 2026-11-14, firmware unknown: not recorded

    How: The owner's report from use.

    Source: dev journal, 2026-11-14

The display

  • 6.8 in, 3120 × 1440, 120 Hz. vendor-stated · medium confidence

    How: Wikipedia's infobox, read 2026-11-21. The page: “S24 Ultra: 6.8 in (170 mm) 3120 × 1440”

    Source: en.wikipedia.org/wiki/Samsung_Galaxy_S24 (read 2026-11-21)

  • Its default density is 600 (3.75 px a dp), the densest screen Dog Brain draws for; the phone was found overridden to 560. Frames are sized for 600 and drawn down elsewhere, so they never blur up. measured · high confidence · 2026-11-16, firmware unknown: not recorded

    How: adb shell wm size and wm density, Dog Brain CLI.

    Source: watch/dogbrain-face/make-clips.js

Timing and calibration

  • The phone fires its buzzes within 1 ms of their target (81 buzzes, sd 0.9 ms). So the ±15 to 40 ms scatter between calibration rounds is the motors and the coupling, not Android's timing. measured · high confidence · 2026-11-11, firmware unknown: not recorded

    How: Logged call time against target, calibration runs.

    Source: dev journal, 2026-11-11

  • The end buzz started 25 to 29 ms after the alarm fired (the watch's 80 to 165 ms). measured · medium confidence · 2026-11-11, firmware unknown: not recorded

    How: Fire and buzz log lines from three timer runs. The value given is the middle of the range (ours).

    Source: dev journal, 2026-11-11

  • On the old Ringtone path the sound started 163 ms after the fire; the tone is now made ahead, at the last-seconds alarm. measured · medium confidence · 2026-11-11, firmware unknown: not recorded

    How: The first timer run's log.

    Source: dev journal, 2026-11-11

Sensors

  • The accelerometer runs at about 416 to 481 Hz for an app, against the watch's 100 Hz. measured · medium confidence · 2026-11-11, firmware unknown: not recorded

    How: Rates seen in calibration runs. The value given is the middle of the range (ours).

    Source: dev journal, 2026-11-11

  • In the living room that looks lit in photos, the light sensor reads 4 to 7 lux (fifty readings): the room is about 5 lux, not 50 to 200. measured · medium confidence · 2026-11-16, firmware unknown: not recorded

    How: Read-only adb read of the sensor, the laptop session. Where the phone lay is not known.

    Source: dev journal, 2026-11-16

Not known yet

  • Battery reporting: CURRENT_NOW's unit, whether the shell can read sysfs, and the charge counter's step. All were found on the S22, not checked here.
  • Drain by state measured with current rather than Android's model: idle, screen on, always-on, motor.
  • The motor's draw.
  • Brightness range and always-on behaviour of the display.
  • Firmware, One UI and Android versions.

Galaxy Watch Ultra SM-L705F · test watch

The first watch Dog Brain ran on, and the dev watch until 2026-11-17, when it became the test watch (paired with the S24 Ultra). Its light sensor, its motor and the watch face were profiled on it, and it is where the charge counter's 284 mAh was first seen.

Regional variantSM-L705F (the international build of the SM-L705; the dev watch is the US SM-L705U; Titanium Blue, the Galaxy Watch Ultra (2025) colour)
Firmware buildL705FXXS3BZF4 (read 2026-11-16 and named again 2026-11-17)
One UIOne UI Watch 8.0
OSAndroid 16 (SDK 36), Wear OS 6
SoCnot known: not in the repo and not read off the watch. Wikipedia's Galaxy Watch 7 article gives the 2024 Ultra "Exynos W1000" but lists it as SM-L900, not SM-L705, so it does not settle this model

Battery capacity

  • Samsung rates the Galaxy Watch Ultra at 590 mAh typical; the pre-launch certification gave 578 mAh rated, the figure Android's power profile carries. vendor-stated · high confidence

    How: Samsung India's Ultra page and a certification report, quoted in the battery research.

    Source: docs/research/battery-capacity-2026-11-17.md

  • Samsung's own battery information screen in Settings says 590 mAh. vendor-stated · medium confidence

    How: Read on a Galaxy Watch Ultra's Settings > Battery information by the owner. Which of the two watches is not recorded.

    Source: docs/research/battery-capacity-2026-11-17.md

  • Android reports an estimated capacity of 578 mAh (dumpsys batterystats, bt field 12): the power profile's figure, never replaced by a learned one. measured · high confidence · 2026-11-10, firmware unknown: before the first firmware read (2026-11-16)

    How: Sniffer's Stats.kt reads field 12 of the batterystats bt line.

    Source: docs/research/battery-capacity-2026-11-17.md

Drain by state

  • 2.59% an hour (about 38.6 h projected) with Dog Brain, Sniffer and DarkWatch running and mixed screen use. At that rate the battery bands read fine down to 42%, yellow 41 to 21%, red from 20%. measured · medium confidence · 2026-11-17, firmware L705FXXS3BZF4

    How: Per cent drop over time, the laptop session's read of Sniffer.

    Source: docs/research/battery-capacity-2026-11-17.md; dev journal, 2026-11-21

  • Over 17.0 h on battery with 1.3 h of screen on, whole-device drain was 10.3 mAh an hour on the counter's 284 scale: about 3.6% an hour (ours, 10.3 / 284). measured · medium confidence · 2026-11-11, firmware unknown: before the first firmware read (2026-11-16)

    How: Android's batterystats, tools/profile.sh (docs/PROFILE.md). The mAh come from the same counter, so they are on its scale.

    Source: docs/PROFILE.md

How the battery reports

  • The charge counter tops out at 284 mAh at 100%, against Samsung's 590. The % drop matches the 284 scale, so mAh read from the counter are on that scale; per cent is right. Dog Brain's battery bands run on per cent for this reason. measured · high confidence · 2026-11-10, firmware unknown: before the first firmware read (2026-11-16)

    How: BATTERY_PROPERTY_CHARGE_COUNTER at a full charge, Sniffer.

    Source: dev journal, 2026-11-10; watch/dogbrain-wear/app/src/main/java/com/dogbrain/wear/BatteryBand.kt; docs/research/battery-capacity-2026-11-17.md

  • Why: in Samsung's open-source kernel for this firmware, the fuel gauge's charge counter is the state of charge times a device-tree constant, fuelgauge,capacity_full = 284, while charging and time-to-full use battery,battery_full_capacity = 590. The cell vendor named there is ATL. The same package covers both watches' builds. vendor-stated · high confidence

    How: Samsung's open-source release for the exact firmware, read file by file by another session and relayed; not re-read by this one.

    Source: docs/research/battery-capacity-2026-11-17.md

  • The shell cannot read the battery's sysfs files: current_now and charge_counter answer "Permission denied". measured · high confidence · 2026-11-21, firmware unknown: not re-read on 2026-11-21

    How: adb shell cat, the laptop session.

    Source: dev journal, 2026-11-21

The motor

  • The motor reports amplitude control, so it gets the fading pulse curve; it flattens above about 75%, so a stressed note is made longer, not only stronger. measured · medium confidence · 2026-11-12, firmware unknown: before the first firmware read (2026-11-16)

    How: The vibrator's capability, and the calibration strength sweep (64, 128, 192, 255).

    Source: dev journal, 2026-11-10; dev journal, 2026-11-12

  • No frequency control: no frequency profile, envelope size 0. measured · high confidence · 2026-11-11, firmware unknown: before the first firmware read (2026-11-16)

    How: Vibrator capabilities read by the calibration code.

    Source: dev journal, 2026-11-11

  • Strong enough to be heard: at full strength the motor is audible on the wrist. Kicks are now held to 191 of 255, and a loop went from 19.6 s of motor to 7.9 s. measured · medium confidence · 2026-11-14, firmware unknown: before the first firmware read (2026-11-16)

    How: The owner's report from wearing it, then the buzz file cut to the kick drum.

    Source: dev journal, 2026-11-14

  • The shortest pause felt as a pause is 38 to 96 ms (about 66 on the phone). measured · medium confidence · 2026-11-12, firmware unknown: before the first firmware read (2026-11-16)

    How: Calibration's fall and rise of the same motor. The value given is the middle of the range (ours).

    Source: dev journal, 2026-11-12

  • Draw not measured. Vibration is not in Android's mAh model either: batterystats counts buzzes and seconds only (605 buzzes, 35 s, in the first profile). measured · high confidence · 2026-11-10, firmware unknown: before the first firmware read (2026-11-16)

    How: tools/profile.sh (docs/PROFILE.md).

    Source: docs/PROFILE.md

The display

  • 1.50 in, 480 × 480. Brightness runs from a measured 5.04 nits to 584 cd/m² without the light sensor, and 3192 cd/m² peak with it. Always-on brightness was not measured; it sits somewhere inside 5.04 to 584 (ours). vendor-stated · high confidence

    How: Notebookcheck's Galaxy Watch Ultra review (a measured review, not Samsung), as quoted in the always-on research.

    Source: docs/research/always-on-power-2026-11-20.md

  • Always-on must light no more than 15% of pixels (Google). Dog Brain's always-on face lights 5.49% by day and 2.01% at night, with no blue pixel lit; the store's own test weights by brightness, ours counts lit pixels, the stricter of the two. estimated · medium confidence

    How: Counted on the generated face drawn in a browser (aod-lit.test.js, aod-power-page.js), not on the panel.

    Source: docs/research/always-on-power-2026-11-20.md

  • Screen captures come off the watch at 480 px raw, 1 px a face unit. measured · high confidence · 2026-11-20, firmware unknown: not re-read on 2026-11-20

    How: adb screencap, the laptop session's review shots.

    Source: commit 8a544c04

Watch face and WFF

  • A Watch Face Format version 4 face loads, and a complication slot binds to the app's ranged-value feed and gets data. measured · high confidence · 2026-11-15, firmware unknown: before the first firmware read (2026-11-16)

    How: The theme trial face installed, logs read by the laptop session.

    Source: dev journal, 2026-11-15

  • The system keeps each complication slot's feed by the slot's place in the face, and re-picking the face does not reset them. So a slot added anywhere but last shifts every later slot onto its neighbour's feed (the bone read "light", a ring the timer's words); added last, it draws over everything. measured · medium confidence · 2026-11-15, firmware unknown: before the first firmware read (2026-11-16)

    How: Three trial faces, the laptop session's slot map and the owner's photos. A first install on a clean watch was not tried.

    Source: dev journal, 2026-11-15

  • A ranged value carrying colours did not recolour the face: the renderer logged "colors should not be null or empty" as the slot came up. measured · low confidence · 2026-11-15, firmware unknown: before the first firmware read (2026-11-16)

    How: One log line from the first trial face, the laptop session's reading of it.

    Source: dev journal, 2026-11-15

  • Samsung's status icons are drawn over the face in their own colours (the face cannot tint them) and fill rows 14 to 62, so the face's top block is laid out under row 62. measured · high confidence · 2026-11-16, firmware L705FXXS3BZF4

    How: Screen captures on the wrist and photos of the night face.

    Source: dev journal, 2026-11-16; dev journal, 2026-11-15

  • The face's countdown complication ran about 2 s behind the system's own timer chip (00:08 against 00:06) though both count to the same end. Likely the renderer redraws the complication's time text on its own schedule; not confirmed. The chip sits over the face and cannot be moved. measured · low confidence · 2026-11-12, firmware unknown: before the first firmware read (2026-11-16)

    How: Seen on the wrist; a screen recording to settle it was asked for and is not in the repo.

    Source: dev journal, 2026-11-12

  • Samsung's weather app is com.samsung.android.watch.weather/.ui.WeatherActivity; the other weather package is the weather face. The face's weather symbol now opens it with a WFF Launch. measured · high confidence · 2026-11-21, firmware unknown: not re-read on 2026-11-21

    How: The launcher activity read off the watch by the laptop session. The tap itself is not yet tested.

    Source: dev journal, 2026-11-21

Timing and calibration

  • Against the phone, the watch's clock drifted from 2.97 s to 3.35 s ahead in about 1.5 h on one day; later about 1 ms a minute, with re-reading jumps of 10 to 25 ms. So the watch keeps re-measuring the phone's clock, and a kept calibration anchors it. measured · medium confidence · 2026-11-12, firmware unknown: before the first firmware read (2026-11-16)

    How: The watch's estimate of the phone's clock over Data Layer round trips, logged.

    Source: dev journal, 2026-11-11; dev journal, 2026-11-12

  • The end buzz started 80 to 165 ms after its alarm fired, against the phone's 25 to 29. measured · medium confidence · 2026-11-11, firmware unknown: before the first firmware read (2026-11-16)

    How: Fire and buzz log lines from three timer runs. The value given is the middle of the range (ours).

    Source: dev journal, 2026-11-11

  • With the S24 Ultra: the watch's click lands 47.5 ms after the phone's (median of 8 runs on the low-latency path; 188 to 217 ms on the old Ringtone path). measured · medium confidence · 2026-11-13, firmware unknown: before the first firmware read (2026-11-16)

    How: The sound step of the calibration, 200 saved runs (docs/CALIBRATION_STUDY.md). Two runs, −0.3 and 67.0, are unexplained.

    Source: docs/CALIBRATION_STUDY.md

  • How late the watch's buzz lands has no stable default: the normal step's median moved +31, +13, +2, −4, −43, −76, +20 ms across seven sittings, though inside a sitting it scatters only 3 to 15 ms. Cause unknown. measured · medium confidence · 2026-11-13, firmware unknown: before the first firmware read (2026-11-16)

    How: The calibration study of 200 runs.

    Source: docs/CALIBRATION_STUDY.md

  • Lying on the phone, each device feels the other's motor 5 to 10 times weaker than its own, and a fixed threshold crosses later on a weak buzz; so each side alone leans (watch −14 to −18 ms, phone 0 to +9), and calibration averages the two. measured · medium confidence · 2026-11-11, firmware unknown: before the first firmware read (2026-11-16)

    How: Super calibration runs, both accelerometers.

    Source: dev journal, 2026-11-11

Sensors

  • Apps get one light sensor, STK31E15 (on-change, 1 to 10 Hz, not wake-up). Samsung keeps two private ones for brightness, one awake and one for always-on, behind the signature-level SSENSOR permission, so the system senses light while dimmed and apps cannot. measured · high confidence · 2026-11-16, firmware L705FXXS3BZF4

    How: dumpsys sensorservice, the laptop session.

    Source: design/light-profile/TESTS.md

  • In the background an app's light listener gets no events (dumpsys: "has sensor access: FALSE"), as Android 9 and later documents for on-change sensors; a foreground service is needed. measured · high confidence · 2026-11-16, firmware L705FXXS3BZF4

    How: dumpsys sensorservice, matched to Android's behaviour-changes page.

    Source: dev journal, 2026-11-16

  • A dark room reads 0 to 2 lux at the wrist with the screen on; a room lit enough for photos reads about 1 to 10 (1 to 2 turned away from the lamp, 5 to 6 toward it). A single reading cannot tell dark from lit; the night face's switch is dark under 1, light again at 2. measured · medium confidence · 2026-11-16, firmware L705FXXS3BZF4

    How: Paired tables against the S24's sensor, the laptop session.

    Source: dev journal, 2026-11-16

  • The accelerometer runs at 100 Hz at most for an app (LSM6DSV). Samsung's faster sensors need a permission only Samsung-signed or system apps get. measured · high confidence · 2026-11-11, firmware unknown: before the first firmware read (2026-11-16)

    How: The sensor's maxRate, read in calibration. This was the only watch in use until the dev watch came on 2026-11-15 (ours, from the journal's dates).

    Source: dev journal, 2026-11-11

System behaviour

  • The back is curved, so the watch rocks and walks across a phone it buzzes on; a thick phone case made no difference. Calibrating on the charger can't work, and the app refuses. measured · high confidence · 2026-11-11, firmware unknown: before the first firmware read (2026-11-16)

    How: Found by doing it.

    Source: dev journal, 2026-11-11

  • Double pinch belongs to Samsung (it can dismiss Samsung's alarms); no source found opens it to other apps on One UI 8 Watch. Google's double-pinch API needs Wear OS 7. estimated · medium confidence

    How: Research against Google's and Samsung's pages; not tested.

    Source: dev journal, 2026-11-11

  • A factory reset gave this watch a new ANDROID_ID, so Sniffer now keys devices by model, not id; its record from before the reset reads "Test watch · until 14 Nov". measured · high confidence · 2026-11-17, firmware L705FXXS3BZF4

    How: Seen in Sniffer's merged watch list; the fix is WatchMerge.ROLE_NAMES.

    Source: dev journal, 2026-11-17

Not known yet

  • The cell itself: no drain test or meter run was done on this watch (both were on the dev watch). That it holds about 590 mAh rests on the same model and firmware.
  • Learned capacity (batterystats fields 13 and 14) and the kernel's charge_full: not read on this watch.
  • Always-on current, and the display's share of it: the before/after test is written (docs/research/always-on-power-2026-11-20.md) and not run.
  • Whether the watch reports low-bit ambient or burn-in protection, and how often the WFF renderer redraws in always-on.
  • The motor's draw in mA.
  • Whether the system holds a wake lock while the motor runs; Dog Brain holds its own partial wake lock for the last-seconds beats.
  • The SoC.
  • Which firmware the 2026-11-10 to 2026-11-15 measurements ran on: the first build read is 2026-11-16.

Galaxy Watch Ultra SM-L705U · dev watch

The dev watch since 2026-11-17, paired with the S22. The battery was measured on it three ways (Sniffer's drain test, a USB meter at the wireless puck twice, and Samsung's own kernel source), which settled that the cell is about 590 mAh and Android's charge counter is on a 284 scale.

Regional variantSM-L705U (the US build of the SM-L705; the test watch is the SM-L705F; Titanium Blue, the Galaxy Watch Ultra (2025) colour)
Firmware buildL705USQS3BZF4 (read 2026-11-17, in the drain test's log header)
One UInot known: not read off this watch (the test watch, on the same release, runs One UI Watch 8.0)
OSAndroid 16 (from the build fingerprint's version field)
SoCnot known: not in the repo and not read off the watch (see the SM-L705F)

Battery capacity

  • Samsung rates the Galaxy Watch Ultra at 590 mAh typical (578 rated). vendor-stated · high confidence

    How: Samsung India's Ultra page, quoted in the battery research.

    Source: docs/research/battery-capacity-2026-11-17.md

  • Measured cell capacity about 546 mAh: the drain test integrated 349.2 mAh of current over 64% of the battery in 1.39 h. Heavy load and heat, the stop at 3% and the seconds lost at each process restart all read low, so the cell is about 590 and healthy. measured · medium confidence · 2026-11-17, firmware L705USQS3BZF4

    How: Sniffer's drain test: CURRENT_NOW every 10 s, trapezoids over unplugged rows (DrainMath), 97% to 3%, screen held on at full brightness.

    Source: docs/research/battery-data/2026-11-17-dev-drain/README.md; commit 1b5ca56a

  • Independent check at the charger: from flat to full took 4.63 Wh into the wireless puck in 2 h 44 min (average 1.69 W, peak 2.6 W). A 590 mAh cell makes that about 46 to 50% puck-to-cell; a 284 mAh one would mean 24%, not credible. measured · medium confidence · 2026-11-17, firmware L705USQS3BZF4

    How: A USB-C power meter (FNB58) between the wall and the puck. The cell's nominal voltage (3.85 V) is assumed, not measured.

    Source: docs/research/battery-capacity-2026-11-17.md

  • Android's estimated capacity is 578 mAh, and the learned capacity (batterystats fields 13 and 14) is 0: the OS has never learned one. measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: dumpsys batterystats bt line, written into the drain log's header by Sniffer.

    Source: docs/research/battery-data/2026-11-17-dev-drain/drain-20261117-1328.csv

Drain by state

  • Screen held on at full brightness with a turning 3D shape, one busy CPU thread and a network ping every 30 s: about 163 mA. measured · medium confidence · 2026-11-17, firmware L705USQS3BZF4

    How: CURRENT_NOW in the drain log, per stretch of load.

    Source: docs/research/battery-data/2026-11-17-dev-drain/README.md

  • Adding four shell busy-loops took it to about 246 mA; a live 3D model of Lookout (Filament) to about 300 mA, about +125 mA for the 3D rendering alone. measured · medium confidence · 2026-11-17, firmware L705USQS3BZF4

    How: CURRENT_NOW in the drain log, per stretch of load.

    Source: docs/research/battery-data/2026-11-17-dev-drain/README.md

  • Under the heaviest load the battery reached 43.5 °C, near empty. The busy-loops were stopped at 42 °C, under where phones commonly throttle. measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: Battery temperature, read once a minute by the laptop session.

    Source: docs/research/battery-data/2026-11-17-dev-drain/README.md

How the battery reports

  • The charge counter falls at about 0.49 of the integrated current, 284/578: it is per cent × 284 (Samsung's device-tree constant), not charge in the cell. Both watches read about 285 to 287 mAh at full. measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: Counter drop against integrated current over the same span. Which watch read 285 and which 287 is not recorded; the value given is their middle (ours).

    Source: docs/research/battery-capacity-2026-11-17.md

  • The charge counter moves only in whole multiples of 284 µAh: all 267 changes between neighbouring 10-s rows of the drain log were 1 to 6 steps of 284 (852 µAh most often), and one 67-step jump across a restart gap. 284 µAh is 0.1% of the 284 mAh scale, so the counter is the level in tenths of a per cent times 284. measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: Ours: consecutive charge_counter_uah differences counted across the 17 drain CSVs (steps 284 ×2, 568 ×38, 852 ×130, 1136 ×65, 1420 ×30, 1704 ×1). Not in the research note before this page.

    Source: docs/research/battery-data/2026-11-17-dev-drain/drain-20261117-1328.csv; docs/research/battery-data/2026-11-17-dev-drain/README.md

  • CURRENT_NOW and CURRENT_AVERAGE come in milliamps, negative while discharging, as Android documents for the sign but not the unit (microamps). In the 10-s log the same CURRENT_NOW integer repeats in 550 of 804 neighbouring pairs, so the value is held or updated slowly; how often the gauge refreshes it is not known. measured · medium confidence · 2026-11-17, firmware L705USQS3BZF4

    How: DrainMath's unit and sign rule on the drain log; the repeat count is ours, from the 17 CSVs.

    Source: docs/research/battery-data/2026-11-17-dev-drain/README.md; watch/sniffer/app/src/main/java/com/dogbrain/sniffer/DrainMath.kt

  • An app cannot read the battery's kernel files: charge_full, charge_full_design, charge_counter, charge_now, current_now, current_avg and cycle_count are all denied (FileNotFoundException). From the shell, current_now and charge_counter answer "Permission denied". measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: Sniffer's drain log header (app), and adb shell cat by the laptop session on 2026-11-21 (shell).

    Source: docs/research/battery-data/2026-11-17-dev-drain/drain-20261117-1328.csv; dev journal, 2026-11-21

  • Voltage ran from 4347 mV at 97% to 3323 mV at 3% under the drain test's load. measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: Ours: the voltage_mv column's first and lowest values across the 17 drain CSVs.

    Source: docs/research/battery-data/2026-11-17-dev-drain/drain-20261117-1328.csv

The display

  • 1.50 in, 480 × 480; 5.04 nits minimum, 584 cd/m² without the light sensor, 3192 cd/m² peak with it (the same model as the test watch). vendor-stated · high confidence

    How: Notebookcheck's Galaxy Watch Ultra review, as quoted in the always-on research.

    Source: docs/research/always-on-power-2026-11-20.md

System behaviour

  • Wear OS kills and restarts a foreground drain service on a reinstall; the run resumes into a new file (a partial wake lock and START_STICKY keep it going otherwise). Integrating each file and adding them is how the 64% span was summed. measured · high confidence · 2026-11-17, firmware L705USQS3BZF4

    How: The drain test's file list.

    Source: docs/research/battery-data/2026-11-17-dev-drain/README.md

Not known yet

  • Idle and always-on current: the always-on before/after test (docs/research/always-on-power-2026-11-20.md) is planned on this watch and not run.
  • The motor: amplitude control, frequency control and draw were read on the test watch, not this one. The watch Sniffer's 4 Hz current sampler is built for the motor power map and not run.
  • How often the fuel gauge refreshes CURRENT_NOW.
  • Timing and calibration with the S22: no pair run is in the repo.
  • Face and WFF behaviour: the quirks found were on the test watch; whether this build behaves the same is not checked.
  • One UI version and the SoC.

Galaxy Watch (SM-R880) SM-R880 · second test watch ("Doe watch")

A second test watch, an older and smaller Galaxy Watch, read once over adb on 2026-11-17. Its companion is another phone, not either test phone, and it was disconnected from that too, so its Sniffer pushes reach nothing.

Regional variantSM-R880 (the model Sniffer names Doe watch. Which Galaxy Watch this is (Watch4 or Watch4 Classic, and the size) is not confirmed by a page read: search summaries disagree)
Firmware buildnot known: not read off the watch
One UInot known: not read off the watch
OSnot known: not read off the watch
SoCnot known: not in the repo

Battery capacity

  • Android reports an estimated capacity of 240 mAh. measured · high confidence · 2026-11-17, firmware unknown: not read

    How: dumpsys batterystats over adb, read only, the laptop session.

    Source: dev journal, 2026-11-17

Drain by state

  • Awake, it drew about 158 mA. measured · low confidence · 2026-11-17, firmware unknown: not read

    How: dumpsys battery's current now, a few reads over five minutes, with adb connected. Whether the value is held, as on the S22, was not checked.

    Source: dev journal, 2026-11-17

  • Dozing with the display off (always-on was off) it read 110 mA, then 45 mA. measured · low confidence · 2026-11-17, firmware unknown: not read

    How: dumpsys battery's current now, read only. The value given is the settled one.

    Source: dev journal, 2026-11-17

How the battery reports

  • Current reads in milliamps, negative while discharging (as on the Ultras and the S22). measured · medium confidence · 2026-11-17, firmware unknown: not read

    How: The sizes of the dumpsys reads (−45 to −158).

    Source: dev journal, 2026-11-17

The display

  • Always-on was off as found (secure setting ambient_enabled 0); it was switched on over adb. Not yet seen in a doze. measured · high confidence · 2026-11-17, firmware unknown: not read

    How: adb shell settings, the laptop session.

    Source: dev journal, 2026-11-17

Watch face and WFF

  • The Dog Brain face installs and runs on it, with the centred-colon clock. measured · medium confidence · 2026-11-17, firmware unknown: not read

    How: Installed over adb; looked at on the wrist.

    Source: dev journal, 2026-11-17

Not known yet

  • Which Galaxy Watch this is, its rated capacity and its screen: no vendor page was read. Search summaries give about 240 mAh, matching the OS figure.
  • Whether its charge counter is scaled like the Ultra's.
  • Drain over hours, always-on current, the motor and timing: only five minutes of reads exist.
  • Firmware, One UI, Android and SoC.