How We Detect Stick Drift

For steady-state drift, five seconds catches almost everything. For intermittent drift — sticks that spike once every few seconds — a longer window catches more spikes. We chose 5 seconds because it's the minimum needed to observe most intermittent behavior while keeping the test brisk enough that users will actually complete it. If you suspect intermittent drift and the test passes, run it three times in a row: you'll catch a stick that spikes 20% of the time.

Because drift is a 2D phenomenon. A stick that reads (0.04, 0.04) is drifting diagonally by a real distance of 0.057 — but an axis-by-axis check that flags anything over 0.05 would give it a clean bill of health. Radial magnitude via sqrt(x² + y²) captures the actual displacement the user experiences. It's the mathematically correct measurement, and it matches how games apply deadzones (radially, not per-axis, on any modern engine).

Two reasons. First, empirical: on a new controller in good condition, the Gamepad API's axis noise floor typically hovers between 0.01 and 0.03 depending on hardware and browser. 0.05 clears that noise comfortably. Second, practical: most games apply deadzones at 0.10 or higher, so drift under 0.05 will never register as input in-game. Anything at or below 0.05 is functionally invisible to gameplay.

Because above 0.15, no amount of in-game deadzone tuning fixes the problem cleanly. A 0.15 drift will punch through most default deadzones (often around 0.10 to 0.15), forcing the user to widen the deadzone artificially — which then eats precision on genuine subtle inputs. Recalibration or replacement becomes the only path to normal play. This threshold matches the fault band in scoreStickDrift() and corresponds to the well-documented threshold where major review sites (Gamers Nexus, iFixit's controller drift documentation) declare a stick clinically failed.

Because that's how the controller feels to the player. A DualSense with a pristine right stick and a drifting left stick is a drifting controller — you can't just play half the game with it. Averaging the two sticks would produce a misleadingly cheerful score. Taking the max of leftMax and rightMax reflects the actual gameplay experience: your best stick doesn't compensate for your worst.

Honest answer: the standalone Stick Drift Test uses radial magnitude sqrt(x² + y²), which is mathematically correct. The composite Controller Benchmark uses the simpler max(|axis0|, |axis1|) for pipeline consistency across its six stages. In practice both approaches produce very similar verdicts because drift almost always presents primarily on one axis; a purely diagonal drift is rare. The scoring library's scoreStickDrift() consumes whichever number the caller provides. If you want the most accurate drift number, use the standalone /tools/stick-drift-test — this is on the polish backlog to reconcile.

No — that's what the Hall Effect Checker is for, and even that tool infers rather than definitively identifies. This methodology measures behavior at rest. Hall effect sticks typically show axis noise floors around 0.005 to 0.015, potentiometer sticks around 0.010 to 0.030 — a subtle difference, not a clean signal. Sensor type inference from noise characteristics is unreliable; check the manufacturer spec or iFixit teardowns for a definitive answer.