Limitations and boundaries ========================== Brushline is a working prototype. This page names the boundaries so a review does not build on a guess. Each entry says what the boundary is and what to do instead. Network and data ---------------- **The BC network is a public-data prototype, not an asset-owner plan.** It uses OSM lines and towers, BC VRI context, and Copernicus terrain. It does not carry BC Hydro or MKRF asset records, clearance standards, outage schedules, or field inspections. **Private artifacts stay private until approval.** The public demo is synthetic. See :doc:`private-public`. **The 590- and 10,030-segment cases are replicated stress instances.** They test scaling and algorithm behavior; they are not real networks. Vegetation and evidence ----------------------- **The growth and treatment model is not field-calibrated for a specific utility.** Its parameters, uncertainty, and applicability domain are prototype values. See :doc:`../reference/low-vegetation-model-card` and :doc:`../reference/growth-model`. **Remote sensing is context and initialization evidence, not a stem map.** LiDAR and canopy products carry applicability and coverage states. Missing or out-of-domain evidence is reported as unknown or partial, not defaulted. **The risk metric is a model quantity.** It is length-weighted state risk. It does not certify clearance, fire behaviour, reliability, or safety. Optimization ------------ **Exactness is limited to the declared model and small instances.** The stateful MILP is exact for its linearized risk state. At 59 segments it is usable; larger cases use heuristics or cluster repair. Heuristic status is not global optimality. **The scenario, robust, and chance plans are sampled experiments.** Chance constraints are enforced on training paths; out-of-sample violations can occur. Robust plans are scenario-robust, not distributionally guaranteed. **Recourse is not a stochastic dynamic program.** Multi-year and weather recourse use deferral or re-optimization with a short lookahead. They do not solve for an optimal policy over a distribution. Weather, calendar, and capacity ------------------------------- **The weather calendar is assumed, not forecast.** Five outage days and independent 20% daily weather loss are scenario assumptions. The calendar-aware case observes the calendar before deferral; the forecast-aware case uses a synthetic forecast. The forecast-skill sweep is a diagnostic, not a forecast product. **Whole-action deferral is conservative.** Calendar-aware execution drops an action when it cannot fit the observed calendar; it does not give partial treatment credit. This can make aware cases look worse on scope or risk than naive partial execution. **Capacity remedies are operational what-ifs.** Window extension and added-crew comparisons hold the crew mix fixed, use prototype day rates, and do not model every seasonal overhead, supervision rule, access constraint, or labour agreement. Costs and commercial use ------------------------ **Costs, hours, productivity, access, and capacity are demo assumptions.** The plan and bid artifacts are inspectable, but they are not a commercial offer, a contractor commitment, or an asset-owner budget. **The planning brief is lab-review material.** It is not a public release of network-derived data and it is not an asset-owner decision record. Release and field use --------------------- **No field validation has been performed.** There is no field campaign, field-calibrated coefficient, safety certification, or asset-owner acceptance. **Release remains gated.** Data-sharing, release approval, field validation, and asset-owner review are external. See :doc:`private-public` and :doc:`../reference/release-boundary`.