Updated 2026-09-14. Only the user’s 240 Hz OLED with software black insertion is proven. U6 Pro at 144/240 Hz and Dell at 120 Hz remain optically unverified with these changes. The old subtraction-based “LightBoost” button was removed. Its description incorrectly equated subtracting a predicted ghost with switching off the backlight, and called assumed Hisense response times measurements. Those claims are withdrawn.
2026-09-22: the user explicitly rejects the earlier Hisense video-derived timings. Do not use them as calibration inputs or manufacturer specifications. The user now identifies the TV as the US 65U6SF-PRO (65-inch Fire TV, SKU 6673625). See the model-specific manufacturer evidence and failed optical result.
The BFI control correction built under build/u6-bfi-control-fix makes all everyday
BFI toggles clear neutral gray reset. A gray reset sequence no longer appears checked as
black insertion. Panel drive buttons explicitly select manual phase/shutter timing;
previously, an enabled guarded LCD aperture could override their requested shutter or
reject a sequence change. Manual phase marks are disabled in guarded timing mode.
These are software corrections, not evidence of clean U6 Pro stereo.
Close the currently running app, then use Launch-Panel-Experiments.cmd. This build lives
under build/panel-experiments/bin/Release; its profiles and reports are beside that EXE.
The build script seeds copies of existing profiles only when a destination does not exist.
The normal executable and original calibration files remain available through Launch.cmd.
Profile version 10 adds guard_level; old executables cannot load version 10 files.
Select the correct display and Match output rate, then open Panel experiments. For optical calibration use fullscreen output, with the controls on another display. The embedded preview covers only part of the panel and cannot establish full-panel behavior.
| Experiment | What is sent | 240 Hz: frames/eye/s | 144 Hz | 120 Hz |
|---|---|---|---|---|
| Direct | L R | 120 | 72 | 60 |
| Black reset | L B R B | 60 | 36 | 30 |
| Neutral reset | L G R G, adjustable uniform gray G | 60 | 36 | 30 |
| Preload | L L R R, emitter trigger on second copy | 60 | 36 | 30 |
| Extended preload, Advanced > Sequence | L L L R R R | 40 | 24 | 20 |
| Hold plus reset, Advanced > Sequence | L L B R R B (or G) | 40 | 24 | 20 |
These rates are arithmetic, not promises about the TV’s internal panel cadence or the glasses’ behavior at every emitter rate. The new buttons preserve the current phase modulo the new cycle and use equal 1500 us starting shutters (or the emitter limit), full image area, ordinary image gain, zero black lift, and no ghost subtraction. They do not guess response times. Longer runs are deliberate low-rate experiments and can visibly flicker.
Repeat direct, preload and neutral-reset comparisons. At these rates, extra settling refreshes cost substantial per-eye cadence: 36 or 30 frames/eye/s. They do not secretly deliver 72 or 60. If preload improves the whole screen, it is evidence that time between eye changes matters. It does not prove the panel’s exact transition time.
The saved hardware report identifies the Dell as S3220DGF. Its manual identifies a VA panel and Fast / Super Fast / Extreme response controls. Compare those settings separately, including inverse ghosts, and clear the optical marks each time. The manual does not document a LightBoost or strobe-backlight control. Dell manual, pp. 14 and 35.
For the TV, compare fixed-rate Game/PC operation, interpolation off, and local dimming off versus on if its actual menu permits. This is an isolation experiment, not a claim about which settings this particular U6 Pro firmware exposes. Record resolution, input, HDR, overdrive, dimming and motion settings; changing them invalidates comparisons.
The nine-row target has three pairs of columns in every row: white/black, middle-gray/black, and light-gray/dark-gray. Left-eye content is brighter in columns 1/3/5; right-eye content reverses each pair. This tests different transitions and reaches close to both screen edges.
At one fixed shutter width, inspect one lens at a time. Find one contiguous acceptable phase range for each screen third. Stop the sweep and let the output settle before marking its start and end. Desired targets must remain visible throughout the range; darkness is not successful isolation. Inspect all three rows in each third and all gray pairs.
Use common measured phase intersects all six ranges (two lenses times three thirds), including ranges that wrap through phase zero. It chooses the center of the widest overlap, keeps the shutter unchanged, and refuses a setting where the emitter would clamp its width. No overlap is reported as no overlap, not replaced by the model’s least-bad phase. This calibration does not treat phase-range width as light-pulse width.
Marks reset when output stops or relevant app settings change. Clear them manually after changes in the display’s own controls, glasses position, or hardware. These are visual observations, not photodiode measurements. Export optical observations records the configuration, six ranges, intersection, notes and host timing counters. Recheck the chosen phase with the captured game: passing a static pattern does not establish moving-content quality.
Black reset asks the previous image to transition toward black. Neutral reset instead asks all pixels to converge toward one intermediate level before the next eye. The hypothesis is that a different transition may erase image-dependent residuals sooner on some LCD modes. It may also increase haze, activate local dimming differently, or fail completely. There is no preselected response-time claim or guarantee of improvement.
The gray slider is an sRGB code value, not a voltage sent to liquid crystals. SDR sends that code; HDR converts it to linear scRGB against the app’s 200-nit reference white. It does not modify the eye images’ blacks or emitter packets. Startup, pause and reacquisition frames stay black. The old leakage model is disabled for this experiment because it has no neutral-reset dynamics. The game hook does not implement neutral reset; use the app’s presenter and the proven window-capture source.
Backlight strobing and shutter-glasses timing must cooperate with pixel settling. Sending a black image, increasing vertical blanking, or changing a model’s illumination flag does not itself switch the physical backlight off. Multipass refresh before viewing is a documented approach to LCD settling; our preload experiment uses duplicate scans and glasses timing, without claiming the monitor’s internal overdrive or backlight is controlled. Blur Busters’ engineering discussion.
For direct L/R, an approximate full-screen settled interval is T - S - R: refresh period
minus actual panel scan duration minus settling time. The requested shutter and timing
margin must also fit. Preloading N identical refreshes changes this to N*T - S - R.
This is an explanatory bound for a simple sample-and-hold model, not a measured property
of either display. Gray transitions, overdrive and scanning backlights need richer measurements.
If the goal is to keep 60 frames/eye/s at a 120 Hz input, investigate accelerated panel scanout/longer blanking or a real synchronized backlight mode. Repeated frames alone cannot retain that cadence. A custom mode must be derived from the actual pixel clock and totals, then checked optically: a TV can buffer and re-time the input. The previous fixed VT 2777 / 3332 instructions are withdrawn; their acceptance and optical effect were never established.
If a TV menu offers a backlight strobe, measure whether it is synchronized and whether it flashes once per input refresh. PWM, local dimming and interpolation are not interchangeable with a clean stereo strobe. The app’s Illumination setting only changes its model.
Its native-display guide restores NVIDIA’s legacy stereo path and links a modified Strobelight utility. Strobelight’s author lists specific ASUS/BenQ LightBoost monitors and DVI/DisplayPort requirements, not generic LCDs, the Dell S3220DGF or Hisense TVs. Strobelight requirements.
The inspected D3D11 compositor converts stereo into spatial display formats. Its overlay presenter uses a separate D3D11 output path. These files do not supply a generic LCD backlight-control implementation. Their useful distinction is between generating stereo images and driving a display already capable of separating them. No code was copied from that project.
Footage240fps/U6_PRO contains LEFT, RIGHT and TV MP4 recordings and extracted images.
The LEFT and TV video streams report 240000/1001 fps (about 239.76), 1920x1080. Their frame
spacing is about 4.171 ms. The inspected TV sample and left-lens contact image show
different relative target brightness at different screen heights. They support investigating
spatial/temporal mixing. They do not independently separate camera exposure/readout,
display scanout, liquid-crystal response and shutter-glasses timing. Periodic sampling can
sometimes reveal sub-frame structure, but requires a justified capture model. No such
validated fit or exposure calibration accompanies the previously asserted 1.5 ms rise /
4 ms fall values in the inspected material; those values are not used as measured inputs
to the new experiments.
A measured, gray-to-gray pre-emphasis pass followed by a target pass could extend preload: write a deliberately chosen first image, then the actual eye image, and open the lens after the second pass. This acts on transition history rather than subtracting the opposite eye’s visible ghost. It needs measured transitions for this panel/mode, including overdrive and clipping limits; otherwise another invented LUT simply repeats the previous mistake. This is a research direction, not implemented or validated in this build.
The official emitter remains the active backend. Replacing it with an RP2040 does not by itself establish faster LCD settling, panel scan behavior, or physical backlight control.