For decades, a mirage deceived Brazil's dams. The computer programs managing 160 hydro plants saw rivers brimming where water was already receding. Nearly 60% of the country's electricity depends on rain, and when AI predicts abundant downpours, operators generously release water through the turbines. But the rains never came — the forecasts consistently erred on the side of plenty. It's like in the desert: you see a lake, draw closer, and it's just sand.

At the core lies a simple arithmetic of water value. Every cubic meter saved today saves money tomorrow, because the rainy season gives way to drought. The computer model for optimization (cost planning) calculates this benefit. If it expects heavy rains, the water's value drops, and the water is released. Scientists tested this self-deception in three steps. First, with toy examples, they showed that even a tiny overestimation in the forecast erases the incentive to store water. Then, using real data over 12 years, they saw: in months with overly optimistic forecasts, the planned reservoir levels also ended up unjustifiably high. Finally, they ran a numerical simulation with thousands of Monte Carlo methods (random weather realizations). The weather in it changed like stochastic processes where each step depends only on the previous one — Markov chains, named after Andrey Markov. They compared two strategies: one trained on biased forecasts, and another with Bayesian inference to remove the systematic error. Such corrections could in future be embedded directly into Machine Learning algorithms and linked with control theory for grid management.

The result is like awakening from an optical illusion. Without correction, the average reservoir level by the end of the drought dropped to a third of the country's needs. Expensive thermal plants kicked in late, driving prices up by 34%. The probability of blackouts jumped to 10,3%—almost one in ten days with rolling blackouts — versus 0,9% under the honest strategy.

The mirage vanishes when forecasts are clarified. Water in reservoirs regains its real value, and Brazilians get light without overpaying.