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 Private and public boundaries.
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 Low-vegetation model card and Low-vegetation growth.
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 Private and public boundaries and Release boundary audit.