======================== The objective ======================== There is no single thing WS4 maximizes. There is a **penalty**, and the search minimizes it: for every target, how far the realized accounts fall outside the bounds you set, weighted, summed over targets and periods. That has a consequence worth internalizing: **anything you do not constrain is free**. The search will spend an unconstrained year, an unconstrained account, or an unconstrained patch class however it likes, because nothing charges it. An empty schedule is a common first result for exactly this reason. Targets are aspirations, not feasibility constraints ---------------------------------------------------- The words ``minimum`` and ``maximum`` name the **hinge of a one-sided penalty**. They do not mean that the model must remain inside a feasible interval. A minimum charges shortfall below the target and leaves over-performance free unless a maximum is also supplied. A maximum does the reverse. This distinction is central to forest planning: * A target can be deliberately set **above the level the estate can actually reach**. That is not an infeasible target; it is a pull-up device. The penalty always has a downward gradient, so the search keeps allocating harvest toward the aspiration. * A set of level targets, one per period, supplies a pull across the whole horizon. The realized level is an outcome of the trade-off, not a value that had to be declared first. * A flow target supplies the **shape**. A dominant flow target, combined with level aspirations that are deliberately too high to satisfy, produces the highest even-flow harvest volume the estate can sustain. This is the **maximum even-flow harvest** pattern. * The target set can be jointly unreachable without being invalid. “Infeasible” describes a hard constraint problem; WS4 is solving weighted goal programming. The result is a compromise, and the report exposes every target's remaining penalty so that compromise can be inspected. Dominance is a matter of degree. Making a flow target dominant with a large weight keeps the level aspirations active as a pull. Making it dominant with a very large priority-tier multiplier can make the level tier almost irrelevant: the flow becomes a lexicographic first objective and the harvest level has little influence beyond a tie-break. Both are legitimate choices, but they mean different things. See :doc:`../howto/even-flow` for the practical pattern and :doc:`../reference/optimizer` for weights and priority tiers. A target has three parts: an account, a period, and an aspiration level or band. It is weighted — `min_weight` on the shortfall, `max_weight` on the excess. `linear` charges the weighted deviation; `non_linear` charges the weighted *squared* deviation, which is the right choice when being approximately right matters more than being on the correct side. Beyond a level, there are two other shapes of constraint: * a **flow** target constrains the shape of a profile rather than its level. `even` compares every period with a reference period, `ndy` compares every period with the one before it, and both are priced per period. Paired with level aspirations, a band around zero finds the highest flat profile an estate can sustain without the level ever being stated. * a **patch** target constrains the map: the size distribution of patches of a criterion feature. It is read at the end offset of its period, priced through a patch model, and it is what makes the search arrange harvesting into openings rather than simply count it. * a **carbon** target constrains an ecosystem, product, or whole-system account. Carbon accounts use the same period and shape vocabulary; with an HWP configuration the system accounts add product storage, end-of-life emissions, and displacement. See :doc:`../reference/carbon`. Weights are the whole tuning story. A weight that is much larger than the others makes its term dominant, which is what you want for something that must not be violated and what you do not want by accident. `ws4 calibrate` runs a short pilot search, records the largest penalty each target reaches, and writes a weights file whose terms are comparable, which is the most reliable first step when a result looks lopsided. Where to read more: :doc:`../reference/optimizer` for the objective and the search stack, :doc:`../howto/write-targets` for the vocabulary, :doc:`../howto/even-flow` and :doc:`../howto/patch-targets` for the two shape constraints.