Fixed original plan A
Hub holding, north backup and south backup. This baseline never changes.
Resilience Designer / Joint closures
Eight crates are ready at the harbour at 09:00. Each depot needs four at noon. Book transport and holding space before you know which routes will close. Then see what gets through on the same bookings.
Checking the complete small booking menu. Controls, results and saved links unlock when ready. A linked setup stays in the address if loading fails.
Walk through the booking decision
Follow five comparisons from one closed corridor to a cheaper plan that leaves a depot empty. The walkthrough uses the fictional booking example; you can restore your workshop and its prior Undo.
One case to inspect
Click a corridor to close or reopen it. Route numbers count crates in this dispatch. The direct pair gives north (N) and south (S) separately. Hub space is booked holding capacity. Grey dashed routes marked ○ are not booked. The route table gives departure times.
This is one maximum-total allocation. Partial totals do not promise a share for each depot.
Across the whole chosen failure family
Four crates total could mean four at one depot and none at the other. This target does not protect both orders.
Fixed original plan A
Hub holding, north backup and south backup. This baseline never changes.
Selected preview B
Capacity first, dispatch later
The search chooses the cheapest plan meeting the selected minimum in every case. Budget determines whether you can afford it. You can also inspect a manual plan; it is labelled separately.
Changing the failure allowance, minimum or available contract reruns the search. Budget only changes affordability. Undo restores the last input or booking change.
Open the original dispatch board ↗The same bookings are retained across rows. Routing changes once each case is known. Counts of passing cases are not probabilities.
| Closed corridors | A north | A south | B north | B south | Inspect |
|---|
| Action | Place or route | Departure / arrival | Crates |
|---|
A valid shipment schedule supplies a lower bound on delivery. A separating capacity cut of the same size supplies an upper bound. Together they certify this declared case’s maximum total; they do not certify a real operation or a fair partial allocation.
| Plan | Cost | Worst total | Meets minimum? | Within budget? | Inspect |
|---|
The original example has four places, six corridor groups, five reservation packages and two four-crate orders receivable only at noon. A route takes one hour. Storage consumes declared capacity for each hour of waiting. The original 8-unit plan reserves hub holding and both depot backups.
The optional authored contract adds two three-hour direct routes, each with four crates of capacity per departure, for 9 units. They share one closure group. It is a declared what-if, not a real supplier, measured price or update to the original example. The original example keeps its own assumptions and file format.
Every no-closure and single-closure case is included. With a two-closure allowance, every pair of distinct declared groups is added. That gives 7 or 22 cases in the original menu, and 8 or 29 with the optional contract. No probability or independence assumption is assigned. A single group can already contain several routes.
All 32 original or 64 extended booking combinations are checked. The search minimizes reservation cost subject to meeting the chosen total in every case. Tied minima use fixed mask order. An above-budget optimum stays above budget; a complete infeasibility result applies only to this declared menu. The full comparison retains equally priced booking variants.
Bookings are chosen before the failure is known. A separate dispatch is then calculated for each known full-run closure case. This is capacity, storage and receiving-window screening. It does not model a surprise mid-run outage, failure rates, individual vehicles, crews, loading operations or total logistics cost.
The unchanged native maximum-flow engine supplies the dispatch. World-row flow and cut checks certify each case. A separately authored crate-journey oracle checks all 2,560 original/extended plan-and-case pairs, removing closed capacity differently. That reference shares the stated world assumptions; it cannot validate fictional rates or actual operations.
Saved JSON and links retain inputs, manual/search mode, inspected case and map choice. Reopening recalculates results. The file contains no prior Undo; the import can be undone during this visit. Saved results and network objects are not trusted. Workshop JSON reopens here; the original board retains its separate file format. The CSV contains every current plan/case row.
APTIFOLD / AGENTIC AI
Aptifold develops its own agentic AI to investigate questions, test ideas and build working software. This experience is one example of that work.
Company links open in a new tab, so you can keep this experiment open.