Process-conditioned schedule evaluation ======================================= Purpose ------- ``ws4.ecosystem.evaluation`` evaluates one block-level harvest schedule through the process engine. It maps blocks to fragments, runs fire, succession, regeneration, salvage, and management, and returns a replayable final state and account table. It is an evaluation path, not a replacement for the optimizer search loop. Inputs ------ - initial cohort state and landscape edges; - block schedule and block-to-fragment mapping; - horizon, fire configuration, optional succession and regeneration sources; - optional process override, harvest fraction, scenario id, and seed. Invalid schedules, horizons, seeds, and mappings fail loudly. Outputs ------- ``ProcessScheduleEvaluation`` returns the final state, event log, per-year values, final accounts, and a JSON summary. ``write`` emits: - ``process_evaluation_values.csv``; - ``process_evaluation_accounts.csv``; - ``process_evaluation.json``. Final accounts include state totals, provisional volume, fire totals, harvest counts and distinct treated area, salvage counts, and regeneration counts. Objective adapter ----------------- ``evaluate_process_schedule`` accepts ``Target`` values and prices the final accounts with the existing weighted-goal objective. A target is accepted only when it is a level target, uses ``single`` aggregation, is not a snapshot, is not an accumulating ``product.`` account, names an existing final account, and ends at the final offset. Unsupported forms fail loudly. The evaluation summary includes ``objective``, ``target_terms``, and ``target_accounts``. Reference case -------------- A 3 × 3 grid of 10 ha fragments with one clearcut on fragments 1 and 2 yields two harvest events and 20 ha of treated area. Final biomass is below the no-harvest baseline, area is conserved, and the same schedule and seed reproduce the same summary. Limits ------ - The final account table is a reporting view; the compiled management dataset remains authoritative. - The volume bridge is provisional and uncalibrated. - The evaluator does not search schedules or replace the optimizer's account overlay. - Objective support is limited to final-account level targets; flow and multi-period targets are not supported. - Harvest is a simplified clearcut-like bridge.