Which quantum encoding method actually works on today's noisy hardware?

Quantum machine learning promises exponential data compression by encoding classical inputs into quantum states — but noisy hardware ruins the math. This survey of 66 papers builds the first framework that jointly weighs circuit cost, expressiveness, and noise tolerance, then distills it into a decision guide: once gate error rates exceed roughly 1 in 1,000 (where most current devices sit), the theoretically elegant amplitude encoding collapses, and plain angle encoding wins. A useful reality check for anyone reading breathless QML benchmarks.