Table 1.

Experimental results for Jarzynski’s equality. We include the confidence intervals for the left hand side of (5), which we obtain after bootstrapping the observed values of $\mathrm{\Delta }{E}_{\mathit{ext}}(\mathit{x})$ for the forward process 1000 times and estimating $⟨e^{-\beta \mathrm{\Delta }{E}_{\mathit{ext}}(\mathit{X})}⟩$ by its mean for each set of bootstrapped data. In our experiment we have $\mathrm{\Delta }F=0$ in the right hand side of (5), resulting in a theoretical prediction of $⟨e^{-\beta \mathrm{\Delta }{E}_{\mathit{ext}}(\mathit{X})}⟩=1.0$. Note, that for most subjects the value of 1.0 lies inside the confidence interval, which does not hold when assuming unsuitable loss functions, as discussed at the end of the Results. Participants that achieve at least $50\%$ adaptation (c.f. Fig. 3) are shaded by a green background color .

Participant	Confidence interval	Participant	Confidence interval
1	(0.03, 48.59)	6	(0.04, 3.75)
2	(0.03, 137.58)	7	(0.01, 0.50)
3	(0.01, 3.63)	8	(1.98, 518130.21)
4	(0.49, 63.48)	9	(0.76, 77.24)
5	(0.46, 1.37)	10	(0.26, 48758.33)