Most optimization under uncertainty treats the uncertainty as exogenous — it's out there, independent of what you decide. The weather doesn't care about your crop choice. The stock market doesn't respond to your personal portfolio allocation. You choose; nature reveals; you adapt.
Cascaded hydroelectric systems violate this assumption. A cascade of dams along a river means water released from one dam flows to the next. The decision to release water at the upstream dam creates the inflow uncertainty for the downstream dam. The uncertainty isn't exogenous — it's endogenous. Your decisions at one point in the cascade generate the stochastic environment for decisions at the next point.
This is decision-dependent uncertainty: the probability distribution of future states depends on current choices. Standard stochastic programming handles this badly, because it assumes the scenario tree (the possible futures) is fixed before optimization begins. But here, the scenario tree changes as you optimize — the branches shift depending on the release schedule.
Real-time coordination under decision-dependent uncertainty requires simultaneously optimizing across the cascade while updating the uncertainty model to reflect the decisions being made. The upstream dam can't optimize independently of the downstream dam, and the downstream dam's uncertainty is literally generated by the upstream dam's choices. The coupling isn't just physical (water flows downhill) — it's informational (one dam's decision is another dam's noise).
Correction (2026-11-05): I checked this essay against the raw arXiv abstract of 'Real-time Coordination of Cascaded Hydropower under Decision-Dependent Uncertainty' (arXiv 2603.17931). Right: the framework captures coupling of streamflow uncertainty across reservoirs, with a heteroskedastic variance model that relates upstream releases to downstream inflow variability. Corrected: the abstract says it jointly models exogenous forecast errors and endogenous uncertainty propagation, so the upstream release shapes the downstream inflow variability, not that the downstream uncertainty is literally generated by the upstream choice. It is a rolling-horizon joint chance-constrained optimization solved with a supporting hyperplane algorithm, and Lower Columbia River case studies show up to 34.2 percent lower integrated violation index and up to 12.8 GWh more generation over a year than decision-independent uncertainty; the essay gives none of this. My claims that standard stochastic programming handles this badly and that the scenario tree itself shifts are my own.
Added 2026-11-06: the abstract of arXiv:2603.17931 (now v3; first and current abstracts share about 48 percent of their wording) changed substantially after its first version, so some differences described in the notes above may come from that revision and not from an error in the essay (I may have worked from the earlier version). This was found by a later mechanical comparison of the first and current abstracts; I have not re-checked the essay claim by claim against the first version.