Footstep Consistency Audit

Each footstep is decided separately in three places. Each one runs UpdateStepSound, then IsMakingFootstepSounds decides whether the step is silent. The marks show where each defect sits.

Every audible step plays the surface's quiet or loud volume, picked by comparing speed with a fixed run speed: 80 u/s crouched, 220 u/s standing. Crouched steps are then scaled to 65%. Those two numbers come from the Source SDK, which crouches at a third of run speed. NEO's speeds sit on the wrong side of both.

So a crouch walk was about 1.6 times as loud as a run, and a sprint was only loud for a recon carrying a 0.85 weapon. Support cannot sprint. Section 3.5 has the fix, and 4.3 the test log counts.

The BALC is the only weapon with no slowdown (1.0), and only the Juggernaut carries it. A Juggernaut sprint reaches 256 u/s and was loud on master. This paper leaves the Juggernaut mode out of the sprint figures; it will be reviewed separately.

The first row is engine behavior inherited from the Source SDK. Each later step was a reasonable answer to the problem in front of the team at the time. Together they explain why master's rule depends on the aim and walk keys, and why some questions were deliberately left open.

#1097 also links a Discord thread and a change document, which are not summarised here.

Formula part 1, crouched line. Crouching used to be silent only while also aiming or holding +walk, so a plain crouch walk was audible. The threshold is now the player's own crouch walk speed plus 10%, including weapon slowdown. That is the same limit as 0.825 of run speed in the formula.

// neo_player_shared.cpp, IsMakingFootstepSounds
if (GetFlags() & FL_DUCKING)
{
    if ((IsInAim() || IsWalking()) &&
        speed <= GetCrouchSpeed() * 0.7f)
        return false;
}

if ((GetFlags() & FL_DUCKING) && !isOnLadder)
{
    float crouchSilentGrace = 1.1f;
    // plus turn headroom, see Turning below
    // ground speed: other clients see vertical
    // speed when a player steps up stairs
    if (groundspeed <=
        GetCrouchSpeed_WithActiveWepEncumberment()
        * crouchSilentGrace)
        return false;
}

Formula part 2. Side-stepping at normal speed and pressing crouch often plays one more footstep. The game only counts a player as crouched once the 0.4 s crouch animation finishes, and a running step fires every 300 ms, so one step almost always lands in between. Walking settles in about 50 ms and is rarely affected, but follows the same rule so both keys behave alike.

Holding a movement key while turning the view, or switching keys such as W+D to W+A, really raises ground speed: acceleration builds toward the new direction faster than friction removes the old one. The gain reaches about 50% of crouch walk speed, more than the 10% headroom, so looking around while crouch walking made footsteps. The turning chart in section 2.1 shows the threshold against the simulated gain.

float crouchSilentGrace = 1.1f;
// crouch sprinting stays audible while turning
if (!m_fIsSprinting)
    crouchSilentGrace += 0.45f *
        Min(m_flFootstepTurnRate / 180.f, 1.f);

Walk speed and crouch walk speed are the same: class speed × weapon × 0.75. Master still measured standing walk and aim against 0.7 × class run speed, which ignores the weapon. The headroom swung from about 10% to 38% by gun, had no allowance for turning, and the BALC was always audible. Walking and aiming now use the crouched rule.

// neo_player_shared.cpp
bool CNEO_Player::IsInSilentMovementMode() const
{
    return (GetFlags() & FL_DUCKING) || IsWalking() || IsInAim();
}

if (IsInSilentMovementMode() && !isOnLadder)
    // silent up to c × (1.1 + t)

Aiming moves at 0.6 × speed, slower than walking, so it keeps about 37% headroom under the shared limit, as on master. Ladders are always audible.

The SDK picks a quiet or loud step volume by comparing speed with a fixed run speed, 80 u/s crouched and 220 u/s standing, then scales crouched steps to 65%. Those numbers fit the SDK's own crouch, a third of run speed (about 63 u/s), so its crouched steps stayed under 80 and quiet. NEO crouches at 0.75 of run speed, so every crouch walk passes 80 and gets the loud volume. Standing, an assault sprint stays under 220 with every weapon and a recon sprint only passes it with a 0.85 weapon, so most sprints get the quiet one.

// neo_player_shared.cpp
bool CNEO_Player::IsQuietFootstep() const
{
    return !m_fIsSprinting;
}

// baseplayer_shared.cpp, UpdateStepSound
fvol = pNeoPlayer->GetSurfaceStepVolume(psurface, bQuietStep); // 0.2 or 0.5 on concrete
if (GetFlags() & FL_DUCKING)
    fvol *= NEO_FS_CROUCH_VOLUME_SCALE; // 0.65

Other surfaces keep their own quiet and loud pair, such as dirt at 0.25 and 0.55. The step interval still follows speed, so step timing does not change. The sprint state and the crouch flag reach every client, so the server, the stepping player and everyone listening pick the same volume.

A flat jump falls too slowly for the engine's landing sound, which needs 350 u/s of fall. The footstep players hear on landing is the regular step whose timer ran out in the air, firing on the first tick back on the ground. Two things had broken that. Since #2151 UpdateStepSound restarted the step timer even in the air, so the first step after landing came at a random phase up to half a second later. And with crouching a quiet mode, a crouched landing at walk speed made no sound at all.

A standing jump in place still makes nothing, since speed is under 50 u/s on landing, as before. The engine's hard-landing sound is unchanged.

Behavior change: players who are out of sight but within hearing range get the server's footsteps again, as in the stock SDK.

Top ground speed for each class, posture and gait on flat ground, and whether a step at that speed is silent. Speeds come from CalculateSpeed capped by the movement code's per-gait limit. The right-hand column gives the silent movement threshold.

• Standing, the threshold applies only while aiming or holding +walk. It is the same number as the crouched one.
• Every posture is silent below 50 u/s.
• Thresholds shown without turning; turning adds up to 0.45 c, see the formula.
• Aiming multiplies speed by 0.6, which is silent in every posture with every weapon.
• VIP and Juggernaut use the Assault numbers. Support cannot sprint.
• Top speeds only. Accelerating, slowing down and the post-jump slowdown for non-recon classes pass through lower speeds.

Top ground speed in u/s with each weapon held, from CalculateSpeed: class speed × weapon speed scale, then × 0.75 for crouching or for standing with +walk. Crouch and +walk come out equal for every class and gun. What differs is the silent movement threshold each one is measured against.

• Bold: in the class's loadout menu or given at spawn. Plain: only by picking it up. –: the class cannot pick it up (CanBePickedUpByClass).
• Run is always audible. Crouch and +walk are both measured against crouch speed × weapon × 1.1, so both are silent for every gun here.
• * PBK56S is only in builds with NEO_BUILD_WEAPON_PBK56S, which is off by default.

Four play passes on a listen server. One bot mirrored the host's inputs (bot_mimic 1) so every step could be compared as the stepping player, the server, and another client heard it. neo_footstep_debug logged each footstep decision, and chat messages marked what was heard in game.

Server footstep origin (pass 1)

Crouched threshold (before, and pass 4 after)

Step volume on master

Shared threshold

Held step and landing, review passes

Test setup

sv_cheats 1; neo_bot_quota 1
neo_footstep_debug 1; neo_footstep_debug_bots 1
con_logfile footsteps.log
nb_stop 1            // walk up to the bot
nb_stop 0; bot_mimic 1

Reported as footsteps while turning quickly in place. Every movement tick was sampled while turning. Speed stayed at exactly 0 and no footstep played, locally or as other clients hear it.

• Pass 2, no lag: turning at up to 3,428° per second.
• Pass 4, net_fakelag 80 and 2% loss: standing, leaning, and all three classes.
• Every step on a tick with no movement keys held was the player slowing down after a run, at 57 to 170 u/s. Those are real movement.

footsteps/sound-consistency-docs sits on top of the PR branch with the ground surface trace fix, the logging, the experiment and these docs.