What this actually tests
Press a key and its matching key on the virtual layout above turns orange while held, then keeps a green outline once you have tried it. That is a straightforward test of whether a key sends a signal at all: a key that sends no signal never lights up here, no matter how hard or how many times you press it.
What it cannot test is more subtle. It cannot detect key chatter, where a worn switch fires two rapid keydown events from one press, since a bounce that fast usually looks identical to a normal press in the event log. It also cannot see media keys, Fn-combo shortcuts, or anything your operating system intercepts before it reaches the browser as a standard key event, which is why a laptop's brightness or volume keys may do nothing here even though they work fine everywhere else.
Rollover and ghosting: why some key combos silently drop a key
Most keyboards are wired as a grid, or matrix, of rows and columns rather than giving each key its own dedicated wire, which keeps manufacturing cheap. The tradeoff shows up when three particular keys are pressed together: the electrical signal can complete a path through all three switches and trigger a fourth, unpressed key on the same rows and columns, a false detection called ghosting. Most modern keyboards catch this and instead just refuse the third key, a milder failure called jamming, but either way a combo you actually pressed does not register correctly.
A keyboard with true two-key rollover reliably handles any two keys held together, which is the baseline every working keyboard should meet. Beyond that, keyboards without per-key diode isolation typically manage four to five simultaneous keys correctly for the most common typing sequences, since manufacturers arrange the matrix so frequent combinations avoid triggering ghosting even if the keyboard is not fully diode-isolated. Keyboards built for gaming, stenography, or fast typing often add per-key diodes specifically to remove this limit entirely, a property called n-key rollover (NKRO): every key is scanned independently, so any combination of any number of keys registers correctly regardless of what else is held.
The 6-key number you may have heard, and why it barely applies anymore
USB keyboards can report input in two different formats. The optional "boot protocol" packs currently-held keys into a fixed 1-byte modifier field (covering both Ctrl, Shift, Alt, and Win keys) plus exactly 6 bytes for everything else, capping it at 6 simultaneously reported non-modifier keys. It exists only so a BIOS or other pre-OS environment, which cannot parse a full HID report descriptor, can still read basic keyboard input using a fixed, predictable format. The mandatory "report protocol", which every modern operating system uses once it has actually booted, has no such limit and supports full n-key rollover if the keyboard hardware itself does.
That is why this tool will often show a rollover count well past 6 without anything unusual happening: your keyboard, OS, and browser are using the unrestricted report protocol, not the BIOS-era boot protocol. The 6-key figure only becomes a real ceiling in the narrow case of typing directly into BIOS setup or a similar pre-boot screen.
Worked examples
Hold W, A, S, and D at once, a common gaming combination: that is 4 non-modifier keys, landing in the typical 3-to-5 matrix rollover range most keyboards handle for common sequences, well short of the boot-protocol ceiling.
Hold Ctrl, Shift, and Escape together (the Windows Task Manager shortcut): Ctrl and Shift are both exempt modifier keys under the boot-protocol accounting above, so only Escape counts as a non-modifier key, for a rollover of 1, the simplest baseline case regardless of how many modifiers rode along with it.
Braille input is an extreme real-world case: Braille2000 and similar braille-keyboard software require up to six keys pressed at once to form a single character, landing exactly at the boot-protocol's 6-key ceiling, which is part of why braille and stenography keyboards are built with full per-key diode isolation rather than a cost-cut matrix.