Layer 4 Self-Consistency Same prompt. Same model. Five answers. The harness finds consensus.

Temp 0.7

Select a preset question above

Run 1

—

Run 2

—

Run 3

—

Run 4

—

Run 5

—

🧩

Pixie

Same model, same prompt — different answers? That's not a bug, it's temperature. The harness runs it 5 times and picks the majority. One API call is never certain.