Validation
How Load Planner is validated
A load plan is only useful if a loader can build it and a weighbridge will accept it. Rather than compare against another product, we trace every decision the engine makes to a public document and an automated test, check the geometry against textbook answers, and measure fill on the standard academic instances next to published results.
1. Calculation rules
Each rule is applied without exception, and each is covered by an automated test that fails if the engine stops honouring it.
| Rule | Source |
|---|---|
| Payload is a hard limit. Cargo over the container rating goes to the next container or is reported, never packed. 40 one-tonne slabs fit by geometry; 28 are loaded in a 28,200 kg box and 12 are reported. | CSC safety approval plate: maximum gross mass minus tare |
| Nothing is placed on a unit marked non-stackable or top-only, and such a unit is never placed on lighter cargo. | CTU Code 2014, Annex 7 s1.6 and s3.2.3 |
| A heavier unit never rests on a lighter one (5% tolerance for like-for-like cartons). | CTU Code 2014, Annex 7 s3.2.3: "Heavier cargoes should not be stowed on top of lighter cargoes" |
| Each unit's "max weight on top" is respected, with loads propagated down multi-layer stacks. | CTU Code 2014, Annex 7 s1.5 and s1.6 (stacking strength of packages) |
| Every elevated unit rests on at least 75% of its footprint; nothing floats or cantilevers. | Engineering choice; the Bischoff and Ratcliff literature imposes a comparable stability constraint |
| Longitudinal centre of gravity reported against the +/-5% of mid-length guidance, with the 60%-in-half-the-length rule of thumb. | CTU Code 2014, Annex 7 s3.1.4 and Appendix 4 |
| VGM is Method 2 arithmetic: sum of unit masses plus the container tare, presented as arithmetic, not a declaration. | IMO MSC.1/Circ.1475 s5.1.2 and s12.1 |
| Cargo must pass through the door opening in its as-placed orientation, not just fit the interior. | Container specification sheets: door apertures are smaller than the interior |
| Units marked "this way up" only rotate about the vertical axis; others may be laid on any face. | ISO 780 handling mark "This way up" |
| Stop 1 is the first drop: it sits at the doors and loads last, and each stop occupies its own slice of the length. | Multi-drop practice: last in, first out |
| Road axle estimate uses the NHVR General Mass Limits as default limits; geometry and tares are the user's. Statics only: a lever-arm split between kingpin and axle group. The weighbridge decides. | NHVR General Mass Limits information sheet, 11 August 2026 |
2. Geometry against known answers
| Fact | Result |
|---|---|
| EUR pallets 1,200 x 800 in a 20' GP (5,898 x 2,350 mm interior) | 11 |
| EUR pallets in a 40' GP (12,032 x 2,350 mm) | 25 |
| EUR pallets in a 45' HC (13,556 x 2,352 mm) | 27 |
| 5,800 mm beams in a 5,898 mm interior; 6,000 mm beams | 8 fit, 2 reported |
| A 1,200 mm crate beside a 1,165 mm pallet in a 2,350 mm interior | Single rows only |
| Container presets: interior, door, tare, payload | Conservative values from three shipping lines' specification sheets |
3. Fill on the Bischoff and Ratcliff instances
The 700 single-container instances BR1 to BR7 from the OR-Library (Bischoff and Ratcliff, Omega 1995) are the standard benchmark for container loading. They carry no weights, so this exercises geometry only; our 75% support rule stays on. Published figures are mean volume utilisation for a 1995 constructive heuristic and a 2003 tabu search, from the citation below. A heuristic built to produce loads a crew can follow will sit below a metaheuristic that searches for minutes; the useful question is how far, and whether the gap moves.
| Class | Box types | Load Planner mean | min / max | CA (Bischoff & Ratcliff 1995) | HBal (Bischoff et al. 1995) | TS (Bortfeldt et al. 2003) |
|---|---|---|---|---|---|---|
| BR1 | 3 | 80.1% | 62.2 / 93.2 | 83.4% | 81.8% | 93.2% |
| BR2 | 5 | 79.4% | 69.9 / 90.6 | 83.6% | 81.7% | 93.3% |
| BR3 | 8 | 78.1% | 70.3 / 85.4 | 83.6% | 83.0% | 92.9% |
| BR4 | 10 | 77.7% | 69.9 / 86.3 | 84.2% | 82.6% | 92.4% |
| BR5 | 12 | 77.5% | 70.0 / 84.7 | 83.9% | 82.8% | 91.6% |
| BR6 | 15 | 76.9% | 68.2 / 83.0 | not in the cited table | ||
| BR7 | 20 | 76.7% | 70.7 / 82.9 | not in the cited table | ||
Engine load-planner-core 0.1.0, run 2026-09-11, 700 instances. Published columns: Araújo, L. J. P. and Pinheiro, P. R., "A Hybrid Methodology Approach for Container Loading Problem Using Genetic Algorithm to Maximize the Weight Distribution of Cargo", IntechOpen, Table 1. Instances: OR-Library, container loading.
4. What this does not prove
- That your cargo weights, dimensions and container tare are right. The plan is arithmetic on what you enter.
- That a plan meets the road rules on your route. The axle estimate is statics with editable defaults; the weighbridge decides.
- That the packing is optimal. It is buildable first, dense second, and the fill table above says by how much.