Monitor Test

Before anything else

1. Darken the room and wait a minute. Your eyes need roughly 60 seconds to adapt before black level, backlight bleed and shadow detail mean anything. Colour and geometry are the only screens you can judge in a bright room.
2. Go fullscreen. Inline in the page the browser may resample the canvas, which quietly breaks the single-pixel patterns.
3. Set OS display scaling to 100% and browser zoom to 100% (Ctrl+0) before you trust the gamma, sharpness or inversion screens. Anything else resamples single pixels and you end up testing the scaler.
4. Change one OSD setting at a time and come back to the same screen. Contrast lives on the white-level screen, Brightness on the black-level screen, Gamma on the gamma screen, RGB gains on the 50% grey screen.
5. Step through fields with ← and → once the stage has focus, and stop the stage with the Stop button or Esc at any time. Escape leaves fullscreen in the same press.

On an OLED, do not leave the colour bars, line grids or geometry screens up for more than a minute or two. High-contrast static patterns are the exact worst case for short-term image retention.

Patterns: what they actually measure

Every one of these screens tests a chain, not a monitor. The image travels from the application to the GPU output (bit depth, RGB against YCbCr, full against limited range), through the cable and link mode, the display scaler, the picture mode and internal LUT, and finally the panel glass. A bad result tells you the chain is broken; which pattern fails tells you where to look. That is why order matters: fix resolution and signal range before you touch gamma, and fix gamma before you touch colour.

On the three uniformity fields, white shows dirty-screen effect and an uneven diffuser, 50% grey is where a colour cast is easiest to catch, and black in a dark room separates backlight bleed from IPS glow from clouding. The grey ramp is a continuous 0 to 255 strip above 32 discrete steps: contour bands mean the signal is being quantised by 6-bit dithering, dynamic contrast or a broken ICC profile, and coloured bands mean a range or profile mismatch, so set the GPU output range to Full for a PC monitor. Black level puts nine patches at 0, 1, 2, 3, 4, 6, 8, 12 and 16 on pure black, and patch 0 is a control that must be invisible; eight or nine visible is healthy, four or five means shadow detail is crushed. White level does the same at the top end with 255 as its control: if 254, 252 and 250 all look like plain white, highlights are clipping and contrast is too high.

Gamma is an estimate and you are the instrument: realistically about plus or minus 0.1, and only when the checkerboard lands on exactly one physical pixel per drawn pixel. It measures the whole chain including the OS colour profile, not the monitor's internal LUT in isolation. It is not a colorimeter and this page will never pretend it is. The two sharpness screens separate the axes: if verticals shimmer but horizontals do not, the signal is being scaled on one axis only, the classic signature of a non-native resolution with aspect stretching. Fix in this order: native resolution, then 100% or integer scaling, then the OSD sharpness back to its neutral midpoint, then disable super resolution, clarity, detail enhancement and any HDR effect processing.

On the colour screens, noisy or fringed red and magenta edges while white edges look fine means 4:2:0 chroma subsampling, so set the GPU output to RGB or YCbCr 4:4:4 and check the cable is rated for your resolution and refresh rate. A sweep row that reaches full saturation early and then sits flat means an over-saturated picture mode or a wide-gamut panel stretching sRGB across P3. Geometry is three checks: the red 1px border must be visible on all four edges or you are being overscanned, the circles must be round, and every grid line must be the same weight. Viewing angle tells TN from VA from IPS. Text clarity is deliberately real browser text rather than canvas, because canvas text is drawn with greyscale antialiasing on most platforms and would hide the subpixel fringing you are looking for.

Dead pixels

Clean the screen first: a speck of dust is by far the most common dead pixel. Kill the room lights and get closer than you normally sit, because one dead subpixel on a 24-inch 1080p monitor is about 0.09 mm across. The black field is the most revealing one, since a lit dot on black is far more visible than a dark dot on white. A dot that is dark only on the red field is a dead red subpixel, which on white shows up as a faint cyan speck that is easy to miss entirely. Photograph what you find: a phone photo of the black field with a lit pixel on it is the evidence retailers accept.

The exerciser repaints the area with fully saturated random red, green and blue noise about 20 times a second at your display's real device-pixel grid. Ten minutes is roughly twelve thousand repaints, and because each subpixel is re-rolled independently it only changes state on about half of them: call it six thousand transitions. Be clear about the evidence: this is a folk remedy with a large pile of anecdote and no controlled study behind it. It sometimes works on stuck subpixels, it cannot restore drive to a dead transistor, and on an OLED a long run is actively a bad trade.

Things that are not pixel faults at all: a dark speck that moves when you wipe (dust on the polariser), a soft corner glow that changes with viewing angle (bleed or IPS glow), a diffuse pale blotch (a pressure mark), and a whole dark row or column (a driver-IC fault, unfixable and an RMA, but not a dead pixel).

Burn-in

Dim the room, turn off adaptive brightness, and on Windows turn HDR off, because in HDR the desktop is tone-mapped and a 5% grey is no longer 5%. Start at 5% grey and step up through 10%, 20% and 50%: a mark that is obvious at 5% and gone by 50% is mild, and a mark still visible on the 50% field is a large differential worth acting on. Turn on the wipe, which is the most sensitive step, because faint marks pop out as the edge passes them. Check white, red, green and blue too: a mark that is clear on blue but invisible on red means the blue emitters in that area have aged, which is the usual OLED failure.

Blue goes first because blue light is the highest-energy visible light and the blue emitter is still the weak link, so it must be driven harder and its higher-energy excitons chew through the host material faster. WOLED tends to read as overall dimming with a colour cast; QD-OLED generates all its light from a blue layer, so blue ageing touches every colour; AMOLED phones are the classic case of a status bar that never moves. What actually helps: auto-hide the taskbar or dock, use a dark theme and wallpaper, turn on logo dimming and pixel shift, drop peak brightness for static work, set a short screen-off timer, and let the firmware run its own compensation cycle when you switch the panel off. Most OLED TVs and monitors already do that automatically, and it is far more capable than anything a web page can do.

Flicker

Start on the bright field in fullscreen and wave a pencil or your spread fingers a few centimetres in front of the panel as fast as you comfortably can. One continuous smear means the light is steady. A row of separate, hard-edged copies of your fingers means the light is being chopped, and you have just seen PWM directly. Do it again at roughly 20% brightness using the monitor's own controls: almost every panel that has PWM at all shows it far more clearly there.

The wave test defeats your retina by putting a moving object in front of a static light source, so every pulse prints a frozen silhouette at a new position. The sweep test needs a caveat, because a moving object drawn on the screen is already chopped by the display itself: an LCD or OLED holds each frame for the whole refresh interval, so a bar at 1920 px/s on a 120 Hz panel jumps 16 px between frames. The frame-step readout exists so you can tell them apart. Copies spaced noticeably closer together than the frame step are the extra ones PWM adds inside each frame.

Two numbers decide whether you will feel it. Frequency: sub-100 Hz PWM is visible to almost anyone who looks, a few hundred hertz is where most complaints live, and above a couple of kilohertz almost nobody reports symptoms. Modulation depth: a backlight that only dips 15% at 240 Hz is far gentler than one that goes fully dark at the same rate, which is why a bare frequency in a spec sheet tells you less than it appears to. IEEE Std 1789-2015 ties the two together, treating deep modulation below about 90 Hz as a genuine risk and anything above roughly 1.25 kHz as low risk at essentially any depth. Flicker free on a box normally means DC dimming, or PWM run fast enough not to matter.

If a screen gives you headaches, run the panel brighter and darken the room instead, because it is the duty cycle that hurts. On a phone, look for the vendor toggle, called DC dimming, anti-flicker, high-frequency dimming or reduce flicker. Check for a strobing mode you turned on yourself: ULMB, DyAc and ELMB are deliberate, deep flicker at your refresh rate. Then rule out the boring explanations, because an uncorrected astigmatism or a prescription two years out of date produces end-of-day headaches that look exactly like a display problem. None of this is medical advice.

The inversion patterns exist because the voltage across liquid crystal has to be reversed regularly or the material degrades, so every LCD flips polarity on a schedule. A pattern that lines up with that schedule makes the two polarities differ slightly in brightness, and the image shimmers or crawls. In real use it shows up on fine text, thin table borders and dithered gradients. Shift the phase to move the pattern onto each of the four pixel parities, because some panels only misbehave when it lands on one of them.

A second effect looks similar and is often mistaken for it. Most 8-bit panels are really 6-bit plus FRC, and many 10-bit ones are 8-bit plus FRC: they fake the missing bits by alternating between two adjacent shades on successive frames so your eye averages them. Frame rate control is a temporal trick, exactly like PWM, and on a slow panel or at a low refresh rate it can produce a fine sparkle in smooth gradients, with the graphics driver adding its own dithering on top. You can usually separate the two: inversion artefacts stay locked to the pixel grid and change when you shift the pattern phase, while FRC sparkle appears in gradients and mid-tones and moves around.

There is no camera or microphone access on this page: a browser gives getUserMedia no manual shutter control, and a camera-based flicker detector in a web page would produce a confident number built on nothing. If you want the camera method, use your phone in manual mode at 1/1000 s or faster and look for rolling-shutter bands that deepen as you dim the screen.

Ghosting

Set the stage to solid blocks at 960 px/s and go fullscreen. Check the frame rate readout matches your refresh rate first, because a dropped frame makes the object jump and the jump reads as a double image that looks a lot like a smear. Find the overdrive control in your monitor menu, where it hides under a different name on almost every brand: Overdrive, OD, Response Time, TraceFree, Rampage Response, Smart Response, AMA or Overshoot. Set it to off, look at the black row, then raise it one step at a time and stop at the last step before a bright halo appears. Confirm on the dark grey row as well, because dark-to-dark is slower than black-to-white and VA panels often want a different step there.

Even a hypothetical 0 ms panel blurs moving objects, because a normal display is sample and hold: each frame is painted once and held still for the whole refresh interval while your eye keeps gliding. The size of that smear is speed divided by refresh rate, which is the css px per frame figure in the readout: 960 px/s is 16 px at 60 Hz, 6.7 px at 144 Hz and 4 px at 240 Hz. No monitor menu touches that number, but two features attack it directly: black frame insertion and backlight strobing, sold as ULMB, DyAc or ELMB. Both shorten the time each image is held on screen, which cuts sample-and-hold blur sharply. The price is real. You lose a large part of your brightness, it usually cannot run at the same time as variable refresh, and the flicker bothers some people. Above about 32 px per frame the stepping dominates everything else. The advertised 1 ms response time is a best-case grey-to-grey figure measured with overdrive at its most aggressive setting over a 10 to 90 percent window; averaged across the full range of transitions a real panel is usually several times slower.

Refresh rate

Browsers fire a requestAnimationFrame callback once per display repaint, so the gaps between callbacks are your display frame times. The mode times consecutive frames and takes the median gap, which shrugs off one-off hiccups that would wreck an average. Hz is 1000 divided by that median: 6.94 ms is 144 Hz, 4.17 ms is 240 Hz. If the result lands within about 2 percent of a common rate it snaps to it, and the raw figure stays visible. Stability is the share of frames within 5 percent of the median, and the 1% low is the rate your slowest one percent of frames were delivered at.

Before you trust a low result: background tabs are throttled to 1 fps or less, battery saver can cap a laptop or phone at 30 to 60 fps, some browsers and remote desktops limit repaints below what the panel can do, and many 144 to 240 Hz monitors ship running at 60 Hz. G-Sync and FreeSync do not interfere, because browsers composite the desktop at the display's configured fixed rate and variable refresh mainly kicks in inside fullscreen games.

Blank screen

Black is the cleaning field: on a normal page dust and dried spray marks disappear into the image, and on black with the backlight up every one of them lights up. Move your head side to side while you look at a speck. If it shifts relative to the pixels around it, it is on the outer surface and will wipe off; if it stays locked to the pixels, it is behind the glass and no amount of wiping will touch it. White is the light panel and the uniformity check, useful for tracing on paper, hunting a dropped screw, or throwing fill on your face for a call. Soft light falls off toward the edges, which reads like a diffused softbox rather than a hard rectangle in the subject's glasses.

The brightness slider does not dim your screen the way the monitor's own control does. A web page cannot touch the backlight, so the slider lays a semi-transparent black layer over the colour. Less light leaves the panel, which is what you want for a bias light at night, but the backlight still runs at full power and the black level does not improve. The warm, neutral and cool presets are approximations: each is the blackbody colour for that temperature converted to sRGB and scaled so the brightest channel sits at full. Note that 6500 K is the sRGB white point, so the cool preset comes out almost identical to plain white, and none of them is a calibrated light source.

Before you trust any of this

• Browser colour management. Browsers treat canvas as sRGB and convert to your display profile. On a wide-gamut panel without a correct profile, full red here is not necessarily the panel's full red, so the colour screens are useful for spotting faults, not for grading.
• Zoom and scaling ruin the pixel-level tests. Gamma, sharpness and inversion are only meaningful at 100% browser zoom and 100% or integer OS display scaling, in fullscreen, at the panel's native resolution.
• Room light and eye adaptation dominate the black-level, bleed, uniformity and burn-in results. The same monitor scores wildly differently at noon and at midnight.
• Judge one screen, one setting, one change at a time. Contrast changes the white level test and the gamma test; chase them together and you will loop forever.
• On a TV, turn the processing off first. Dynamic contrast, noise reduction, motion smoothing and AI picture modes mean you are testing the processor, not the panel.
• Reported colour depth is worth nothing. Current browsers return 24 from screen.colorDepth in practice regardless of the hardware, to deny fingerprinters a signal, so it tells you nothing about whether the panel or the link is 6-bit, 8-bit or 10-bit. The grey ramp is the honest test for that.