Forty-one captured traces across nine areas. The average frame rate is respectable and almost irrelevant: this game's problem is a repeating 34 millisecond spike that arrives every time the streaming system crosses a cell boundary.
The average is fine. The experience is not, and the reason is a single repeating stall that appears in thirty-eight of our forty-one captures at almost exactly the same interval.
End-to-end button-to-photon latency for six titles at 2,400 samples each. The spread between best and worst is 41 milliseconds, which is roughly two and a half frames of advantage.
A stealth game whose lighting model is not decoration but the entire ruleset. Our capture data shows what it costs, and it is the rare case where the cost is obviously worth paying.
Fourteen late-game saves timed across identical turn sequences. Simulation cost scales with entity count far more steeply than the interface suggests, and the threshold is uncomfortably early.
Test
The same traversal repeated forty-one times to isolate where the engine loses a frame and where it loses eleven.
Latency
High-speed camera measurement rather than in-engine reporting, because a game cannot honestly measure its own delay.
Review
What it costs to make every shadow authoritative, measured across twelve levels and reported honestly.
Simulation
Fourteen saves, identical sequences, and a scaling curve that gets ugly around hour ninety.
Audio
Recorded audio latency and occlusion accuracy in a horror title that leans on both more heavily than its marketing admits.
Display
Comparability beats representativeness. The reasoning behind a fixed display and what it costs us.
FrametimeDesk