If the P&L distribution is continuous, expected shortfall is the expected value of the loss given that the loss is at least VaR. Samples from historical simulation and the Monte Carlo method are not continuous distributions, so you rank the P&L figures from the largest loss down and average the losses up to the rank that gives VaR. Japan's capital adequacy notice for banks uses the term "期待ショート・フォール" (expected shortfall), and the internal models approach under the FRTB calculates it at a one-tailed 97.5% confidence level.
Consider a hypothetical example in which the five largest losses among 200 scenarios are 30 million, 24 million, 21 million, 19 million, and 16 million yen. The 97.5% VaR is the fifth, 16 million yen, and expected shortfall is the average of the five, 22 million yen.