How the recommendation works¶

Also in the app

Press ? in Fab Dispatch to read this in the help center, or / to ask the assistant.

The recommendation card picks one plan using an explicit rule for the planning goal you choose. It never declares a winner on a difference that could be noise.

Planning goals¶

Best value: cost × latency (default). Never gives up bottleneck coverage. Among plans whose cost is within the noise margin of the cheapest, it picks the fastest to compute.
Protect bottleneck tools. Most bottleneck tool-downs covered, then the fastest response.
Maximise coverage. The most priority-weighted work served, then the lowest cost.
Lowest total cost. The lowest total cost (the objective every strategy minimises, workload balance included), whatever the solve time.

The noise margin¶

On a single shift, re-sequencing one job can move the cost by a few points. So a cost difference smaller than 2% of the cheapest plan, or 5 points, whichever is larger, counts as a tie, not a win. Calling a 1% difference a win would be a false positive. The shaded band on the chart shows this margin.

Best value, step by step¶

Coverage guard: drop any plan that covers fewer bottleneck tool-downs, or noticeably less priority-weighted work, than the best plan.
Cheapest: find the lowest-cost plan that's left.
Ties: every remaining plan within the noise margin of it is equivalent.
Fastest: of those, the one with the shortest solve time wins.

The card states which case applied, and the trade-off line names what you'd give up by picking something else.

The cost × latency chart¶

Each point is a strategy: total cost (down is better) against solve time (left is better, on a log scale). The line joins the efficient frontier: strategies no other strategy beats on both cost and speed. A strategy off the frontier is dominated.