Four optimizations left the list. None of them for lack of room.
They passed verification and did nothing, or did harm. Two are still sold elsewhere as a performance gain. This page is the autopsy, and the reason Maximize now checks the effect on Windows instead of the value written to the registry.
The trace below is frametime across a session — higher means a frame that took longer. Each switch hands one removed optimization back to the product. Turn it on and watch what it does to the frame.
Frametime · session
Average
8.3ms
Worst 1%
8.7ms
Clicks dropped
0
Link drops
0
More interrupts, not fewer. It promised a predictable timer; it delivered microstutter and a more irregular frametime, the exact opposite of the advertisement.
The red dots are dropped clicks. The queue is not latency: it is overflow capacity. With a 1000 Hz mouse and a short queue, every DPC spike above ~20 ms overflows and throws events away, including the click. In CS2 that is the crosshair dead on target and the shot that never fires.
The red band is the link going down. Five to fifteen seconds, worse on Wi-Fi. And since Game Mode usually activates after the game has connected, that is the player dropping mid-match.
That cache was useful. Emptied, Windows re-reads everything from disk mid-match. The spikes that appeared are I/O, not CPU.
The trace did not move. That is the point. These are TCP-only tweaks, and competitive games use UDP — they never touch the traffic that matters. Worse: they only count after the interface restarts, and the session reverts on exit. The old verification still said “applied”, because it read the registry instead of the effect.
A demonstration trace, deterministic — identical on any machine. On your PC the measuring is done by the app's Live Metrics tab, with your own hardware's numbers.
The three cases, one by one.
Three cases, four identifiers: Nagle and the TCP stack tweak left together, for the same reason. Every one kept its revert path on purpose, so an older install still undoes whatever it applied.
sys-dynamic-tick
Turn off the kernel dynamic tick
Looked like
A more predictable timer, less jitter.
What happened
It raised the interrupt and DPC rate. Result: microstutter and a more irregular frametime — exactly the opposite of the promise.
Removed · not coming back
periph-input-queue
Shrink the mouse and keyboard input queue
Looked like
Shorter queue, less buffer, less input lag. A widespread forum myth.
What happened
The queue does not reduce latency — events are consumed as they arrive. It is overflow capacity. With a 1000 Hz mouse, any DPC spike above ~20 ms overflows and discards events, the click among them.
Removed · revert kept
net-disable-nagle · net-tcp-tuning
Disable Nagle and tune the TCP stack
Looked like
Less network latency, less bufferbloat.
What happened
Two failures at once. They are TCP-only tweaks, and competitive games use UDP. And they only count after the interface restarts, while Game Mode is a session that reverts on exit — the value never got to count.
Removed · placebo
The lesson that survived
Check the effect, not the value written.
Nagle is the teaching case: the verification screen said "applied" because it re-read the registry and found the value sitting there. It was right about the registry and wrong about the machine, the tweak only takes hold after the interface restarts, and Game Mode is a session that reverts on exit. The value never got to count.
After that, verification started re-reading Windows itself. Of the 47 tweaks, 39 are confirmed by re-reading the system. The other 8 are labelled "applied" rather than "in effect", because they leave no state to re-read — closing a browser writes nothing anywhere.
One detail separates checking from pretending: the 13 services are read from the Start value in the registry, not from Get-Service. Get-Service reports a trigger-started service as "Automatic" even when it is disabled — anyone relying on it starts reporting success where there was none.
Verification
39/47
confirmed by re-reading Windows
8
labelled “applied”, because they leave no state to re-read
Verification method for each of the 47 optimizations
Method
What is re-read from Windows
How many
Registry DWORD
the exact numeric value that was expected
12
Registry string
the exact text value that was expected
5
Service Start
4 = disabled, read straight from the registry
13
Live command
power plan, QoS policy, DNS and adapter
9
None: labelled “applied”
nothing: closing a program writes no state
8
Check it yourself: the Verify now button sits on the Game Mode screen itself, and it runs on the free plan.
Four things this product forbids itself.
These are not architectural opinions. Each one ran here, caused concrete damage, and became a written rule.
Forbidden practices, the damage they caused and what is done instead
The practice
What it caused
What is done instead
Restarting the network adapter to apply a tweak
Drops the link for 5 to 15 seconds, worse on Wi-Fi. And Game Mode usually activates after the game has opened its connection: it disconnected the player mid-match.
Apply without restarting and label it effective after reboot. Physical adapters only, never a VPN.
Purging standby memory before playing
That cache was useful. Emptying it forces Windows to re-read everything from disk mid-match: I/O stutter.
Only under real memory pressure, and outside the presets.
Freezing programs that inject an overlay
Suspending Discord, Steam, NVIDIA or OBS hangs the game when it tries to sync with the overlay.
They can be closed. Never frozen.
Detecting success from message text
The check looked for English phrases. On Portuguese Windows, every update was marked as a failure.
Exit code and re-query by identifier. Same for reading ping: time=, tempo=, Zeit=.
The button that checks is free.
You do not have to take a single line of this page on faith. Install it, turn on the Basic preset and press Verify now: it re-reads Windows and tells you, item by item, what is actually in effect.