An asymmetric readout prediction scan holds one asymmetric assignment matrix fixed (correct rates 9/10 for prepared zero and 7/10 for prepared one) and varies the true state instead: a balanced state, a 3/4-1/4 state, and a 9/10-1/10 state each produce 80, 85, and 88 expected zero reads out of 100 shots exactly for the stated toy calibration numbers. The same fixed matrix skews every true state's observed budget in the same direction, toward zero, regardless of how close the true state already is to certain; real prediction and mitigation pipelines correct for exactly this kind of matrix bias.
A fixed asymmetric assignment matrix biases the observed budget the same direction for every true state it is applied to. Exact arithmetic here means exact results for the stated model inputs; measured inputs still carry uncertainty and significant-figure limits.
highlighted = computed this step
The same asymmetric matrix starts from a balanced state
Chapter six's matrix-quality scan fixed the true state and varied the matrix. Here the matrix stays fixed at the mild asymmetric row from this chapter's first lesson, and the true state changes instead.
ptrue=(21,21)
A fixed asymmetric matrix skews every true state the same way
Each row's observed zero probability sits above the true zero probability, because the matrix under-reports prepared one more than it under-reports prepared zero, regardless of which true state is checked.
A nearly-certain true state still carries the same bias
Even when the true state is already 9/10 toward zero, the asymmetric matrix still pushes the observed zero count above the true count: 88 expected zero reads out of 100 shots.