If you only watch the rep counter, you will mis-train. 67 Speed Challenge scores each rep with a multiplier tied to recent speed. Here are the engine numbers from the shipping client so you can practice with intent.
Rep registration
A rep is registered when the active input path accepts a completed oscillation (webcam peak–valley per hand, or keyboard alternation). Each rep pushes a timestamp into a rolling buffer and awards points based on the current combo level at that moment.
Rolling RPS window
Reps-per-second is computed from timestamps inside the last 2000 ms:
RPS ≈ (number of reps in last 2 seconds) / 2
That window is short on purpose: stop pumping and the combo level falls quickly. Burst, coast, burst creates a sawtooth multiplier; smooth density pays more.
Combo thresholds and points
| Level | RPS threshold | Label | Points per rep |
|---|---|---|---|
| 0 | < 3.5 | PUMPING… | 10 |
| 1 | ≥ 3.5 | HYPER! | 20 |
| 2 | ≥ 6.0 | GODLIKE! | 35 |
| 3 | ≥ 8.5 | SUPER 67!!! | 60 |
Illustrative math: 40 reps entirely at level 0 → 400 points. The same 40 reps held in level 2 → 1400 points. Density dominates raw count.
Strategy implications
- A controlled mid-fast pace that holds level 1–2 often beats a chaotic sprint that spends half the run at level 0 after a collapse
- Webcam players who lose tracking mid-combo fall hard—protect signal quality (setup)
- Keyboard players should focus on even alternation; missed alternation breaks rhythm the same way
- Use the first second to lock form, then push into higher bands
What the HUD is telling you
Combo text, colors, screen shake, and pitch-shifted sounds are not decoration—they encode level. If the UI is calm, you are leaving points on the table or protecting form after a stumble. Learn which is which by watching RPS feel across a session logged in the training plan.
Fair comparison
Compare scores within the same mode. Read mode differences and fair play before treating leaderboards as absolute truth.
Worked examples (same reps, different density)
Imagine two 20-second webcam runs that both finish with 48 counted reps. Player A clusters almost all reps into the first 8 seconds, then fatigues and dribbles single pumps. Player B keeps a slightly lower peak but stays inside the HYPER band for 16 seconds. Because points are awarded per rep at the moment of the rep, Player B usually posts the higher score even with equal totals. That is the core training insight: protect the rolling window.
Another example: 30 clean reps entirely at level 0 yield 300 points. Only 24 reps held at level 2 yield 840 points. Chasing rep count alone is a beginner trap. Watch the combo label like a fuel gauge.
Webcam vs keyboard scoring parity
Once a rep is accepted, the point table does not change by mode. What changes is how hard it is to feed the detector: vision noise can silently delete peaks that your muscles “felt.” Keyboard deletes fewer invisible reps but introduces finger-order errors if you double-tap one key. Log modes separately when you study progress charts.
HUD audio as a coaching channel
Pitch and timbre shift with combo level. If you practice without staring at numbers, sound alone can tell you when you fell out of HYPER. Toggle sound from the footer if you are in a quiet office; otherwise use it as a free metronome of success.
Practice drills tied to the math
- Band hold: Aim only to stay in HYPER for a full run—ignore SUPER chasing.
- Recovery reps: Intentionally slow for one second mid-run, then re-enter level 1 without panic flailing.
- Blind audio: One run watching only the preview skeleton and listening for pitch; glance at score after.
- Mode contrast: Same day, one cam PB attempt and one kbd PB attempt; compare density notes, not raw ego.
Myths
- “Max shake always max points.” False if tracking drops or the window empties.
- “Particles mean I’m optimal.” Particles fire on reps; they do not guarantee high multiplier bands.
- “Leaderboard score is SI-unit speed.” No—it is an arcade function of detection + density.