CS2 movement constants: gravity 800, friction 5.2, stopspeed 80, accelerate 5.5

How it was measured

The instructive part of this page is not the numbers, it is a wrong number that looked right. Take the per-tick change in vertical velocity of an airborne player and read its median: on current demos that gives −12.0, hence 768 u/s², and it is stable, repeatable and completely false. The cause is that 86.8% of airborne z rows land on a 1/64 u grid, so the median snaps to a grid value rather than to the true one. The derivation tool now prints that figure explicitly labelled unquotable, so nobody rediscovers it as a finding.

Fitting whole arcs instead gives 800.003 with a median absolute deviation of 0.121 over 454 jump arcs, and two independent current-build recordings give 799.93 and 800.26. Three sources converging on 800.00 is what makes it VERIFIED rather than merely measured.

Friction and stopspeed come out of the same recordings without needing to label which frames were decelerating. Below the knee of the decay curve the per-tick speed drop is a constant equal to stopspeed × friction × dt; its measured median is 6.5010 (MAD 0.0043), and 80 × 5.2/64 is 6.5 exactly.

Crouching is a timeline, not a state. The duck amount ramps linearly at 6.4 per second, reaching 1.0 in exactly ten ticks, but the collision hull is binary — you stand 72 units tall until the ramp completes and then you are 54. On the ground your feet stay planted through the change; in the air the origin re-centres by ±9 units, half the difference. That rate was byte-verified in the binary, superseding an older demo-fitted "14 ticks in, 12 out" that had absorbed network latency into the mechanic.

One negative result is worth as much as the constants: there is no instantaneous horizontal slowdown on landing from a single jump. The ratio of speed after landing to speed before is 1.0000 across every airtime bin, over 220 landings.