Deterministic schedule mutation grammar
Purpose
ws4.opt.schedule_grammar supplies a neighbour function for the P193 process neighbourhood
search. It performs add and remove moves over a fixed block universe, treatment set, and horizon.
Treatment windows
windows optionally maps a block to (start, end) offsets, with an inclusive start and
exclusive end. Omitted blocks use the full horizon. Add and offset-move neighbours iterate only
inside the window, and a schedule outside its window is invalid.
The default remains no windows, so existing add/remove and offset-move neighbour counts are unchanged.
Universe
A grammar has unique positive block ids, unique non-empty treatment names, a positive horizon, and a positive per-block event cap. A candidate schedule is valid only when every entry is in that universe and no block has duplicate events.
Moves
For each block below its cap, the grammar adds every treatment at every unused offset. It also removes each existing event. Add moves come first, then removes. Neighbours are deduplicated by canonical schedule id.
Candidate ids are SHA-256 digests of the canonical schedule payload, so the same schedule always has the same id.
Reference case
With blocks 1 and 2, treatment CC, a two-year horizon, and a cap of one, the empty schedule
has four add neighbours: {1: (CC, 0)}, {1: (CC, 1)}, {2: (CC, 0)},
{2: (CC, 1)}. It has no remove neighbours.
With a cap of two and schedule {1: (CC, 0)}, the neighbours are:
{1: (CC, 0), (CC, 1)};{1: (CC, 0), 2: (CC, 0)};{1: (CC, 0), 2: (CC, 1)};the empty schedule.
Limits
Add/remove by default; optional offset moves do not cover treatment changes.
No treatment-window grammar.
No spatial, road, adjacency, or patch constraints.
The optimizer’s existing schedule semantics are unchanged.