The overflow moment
A polysynth with four voices is four separate instruments behind one keyboard. Hold four notes and all four are busy; press a fifth and the allocator must free a voice by silencing a note you are still holding. That forced choice is voice stealing.
Last-note priority: the policy that won
Which note dies? Almost every modern instrument answers the same way: the oldest one — the note that has been sounding longest gives up its voice to the newest key press. The logic is musical: the newest note is the one your ear is following, and the oldest has had the most time to register. Keys Warrior’s own allocator implements exactly this policy, and the 4-voice chord pages exercise it on real chords.
A steal, step by step
Here is the exact scenario, run through this site’s own allocator: hold C4, E4, G4, and Bb4 on the 4-voice polysynth — a C7, all four voices busy. One beat later, add D5 on top. The allocator scans for a free voice, finds none, and applies last-note priority: C4, the oldest note, is cut at the moment D5 arrives, and D5 takes over its voice. What you hear is a C7 that quietly loses its root as the 9th blooms on top — which, not coincidentally, is the same trade the chord reducer makes on paper: dropped: 5th, root — kept: 9th, 3rd, 13th, b7 is its verdict on a full C13 in four voices.
CI replays that exact scenario and asserts the victim, the thief, and the truncation — the guide and the allocator cannot drift apart.
Hearing it coming
Voice stealing is not a malfunction — but unplanned, it sounds like one: a pad note vanishing mid-sustain, a bass note cut off by a grace note. Players on small polysynths learn to budget: leave a voice free when you plan an ornament, or thin the chord yourself before the machine thins it for you. The chord charts here run every quality through the budget in advance, so you can see which extended chords will force a steal before you play one.
Stealing versus reducing
Stealing happens in real time, note by note. This site’s chord reducer does the same arithmetic ahead of time — dropping doubled tones first, then the dispensable fifth, then the root — so the chord that reaches the voices already fits. Same constraint, two responses: the machine’s reflex and the player’s plan. The allocation guide covers the bookkeeping underneath both.