The planning stack

Brushline keeps three planning scales and several policy comparisons separate. They read the same decision units and write versioned artifacts.

Strategic

The strategic layer selects segment-year-method actions under budget and capacity constraints. The stateful MILP carries risk forward from year to year and is exact for the declared linear risk state on small instances. The heuristic path uses Fresh Solutions and cluster repair for larger instances.

Useful pages: Stateful strategic MILP, Stateful heuristic, Stateful decomposition.

Tactical

The tactical layer groups segment actions into treatment units, assigns contractors and methods, checks access windows and capacity, and builds a multi-year program. It is a prototype allocation layer, not a committed contractor schedule.

Useful pages: Tactical program, Contractor capacity and cost, Access and road status.

Operational

The operational layer creates work orders and a deterministic crew/machine shift schedule, then exposes the result as a timeline. The scheduler respects method skills, shift hours, travel, and blocked days. It does not model every real-world labour rule.

Useful pages: Operational work plan, Shift scheduling, Operations timeline.

Policy comparisons

The prototype also contains policy experiments above the base plan: static frontier plans, scenario-robust plans, sampled chance constraints, two-stage deferral, rolling re-optimization, weather-aware execution, forecast-skill sensitivity, and capacity remedies. These pages are experiment records; they are not additional certifying standards. Start with Uncertainty and recourse and use the reference page for the specific comparison.

Artifact flow

The normal flow is:

dataset → plan → run → tactical → resources → operations → shifts

Each artifact has a schema version, a validator, and named errors. The map is in Contract index.