What superposition actually is
Superposition means a qubit's state is two amplitudes — signed (in general complex) numbers, one per outcome — not that the qubit is "both 0 and 1 at once". Measurement turns amplitudes into probabilities by squaring their magnitudes, so an equal superposition reads 0 on about half of shots — with sampling noise that shrinks as you take more.
A qubit's state is two numbers: an amplitude for the outcome 0 and an amplitude for the outcome 1, written α|0⟩ + β|1⟩. Amplitudes are, in general, complex numbers (every circuit on this page keeps them real), and they obey one rule: |α|² + |β|² = 1, because the squared magnitudes are the measurement probabilities.
That is the whole definition. "Superposition" simply means both amplitudes are non-zero. The state after H is (|0⟩ + |1⟩)/√2 — amplitude about 0.7071 on each outcome, so probability 0.5 on each.
So is the qubit 0 and 1 at the same time?
No — and the slogan fails in two concrete ways. First, it suggests you could read both values out. You cannot: measuring the qubit yields exactly one bit, and afterwards the state matches the bit you observed (that is the subject of the next lesson). Second, "both at once" says nothing about the signs of the amplitudes — and the previous lesson showed the signs doing all the interesting work, cancelling an outcome exactly.
The accurate mental model: the state is a definite mathematical object — a direction, not a dice roll — which deterministically fixes the statistics of any measurement you might make. Gates transform amplitudes deterministically; randomness enters only at readout.
What does equal superposition look like at 128 shots?
Does more data make it exactly 50/50?
How close should the counts be?
Shot counts follow binomial statistics. Estimating a probability p from n shots carries a standard error of √(p(1−p)/n) — exactly the ± the panel prints. At p = 0.5 that is ±4.4 percentage points for 128 shots, ±1.6 for 1,024, and ±0.4 for 16,384. Halving the error costs four times the shots.
Two honest consequences. You will never get exactly 50/50, and you should not expect to — a result of exactly 8,192/8,192 would itself be a mild statistical surprise. And no finite number of shots reveals an amplitude directly: you only ever estimate probabilities, to finite precision, from samples. The histogram is a sampled measurement record, not a picture of the state.
What does this look like on real hardware?
On a physical device, superposition is a fragile physical arrangement, not a bookkeeping entry. Two clocks limit how long it survives: T1, the timescale on which an excited qubit relaxes towards |0⟩, and T2, the timescale on which phase information degrades. Today's superconducting processors typically sit at tens to hundreds of microseconds; trapped-ion qubits reach seconds or longer and hold superpositions the longest of any commercial platform.
A circuit has to finish well inside those windows, which is why gate speed and coherence are always quoted together. Both figures appear on each device's spec page, and the comparison view shows how differently the platforms trade speed against coherence.