How we check a plan

The planner places the boxes. Then a separate checker — written apart from the planner, sharing none of its packing code — measures the plan again. It runs on every plan and again after every change you make. A plan is shown as Checked only when it passes.

What the checker measures, box by box:

• Every box is inside the walls, with the clearance you set.
• No two boxes overlap.
• Support under every box: its whole base rests on the floor or on box tops at most 5 mm lower.
• Door size: each box passes the door the way it is placed.
• Payload: the weight in each container is within its payload.
• Your stacking rules: “Keep upright”, “Don't stack on top”, “Max weight on top”, “Max layers” and “Floor only”.
• Heavy on light: a box only rests on boxes at least a third of its weight (the 3× rule below).
• Counts: loaded plus left out equals the Qty you entered.

On the standard benchmark below, the checker found 0 problems in 1,500 plans.

Bischoff & Ratcliff is a public set of 1,500 packing problems — 15 classes of 100, each a list of 3 to 100 box sizes for one container of 587 × 233 × 220 cm — that researchers use to compare packing methods.

Fill is the share of the container's space the plan uses. Both rows follow the same rule as TrimLoad: the whole base of every box must be supported.

With few box sizes we are 1.14 percentage points behind the best research result. With many box sizes — closer to a long, mixed packing list — we are 1.91 points behind. Our own target there was 89 %, and we reach it on average (89.32 %), but not in every group: the hardest ones, classes 13 (88.38 %), 14 (87.76 %), 15 (86.98 %), are still below it. The research results take minutes per problem; a plan in TrimLoad takes about a second, because you are waiting for it.

Our run: all 1,500 problems, 100,000 search steps each, one desktop PC, 18 Oct 2026. Every plan was re-checked; none was below the old engine it replaced.

A heavy crate standing on cartons can crush them in transit. By default a box may only rest on boxes that weigh at least a third of it — at most 3 times heavier per box. You can switch the rule off for a plan in its settings.

We test the rule on 10 dangerous stackings we wrote for this purpose — heavy pieces 6 to 51.4 times heavier than the boxes under them. The rule must block every one:

The lightest of these is 6 times heavier, so the 3× limit leaves a margin of 2.

The rule has a cost: some cargo can no longer be stacked, so a shipment may need more containers. On 12 test lists we generated (not customer lists):

In plain words: the 3× rule blocks all 10 dangerous cases and needs 8 fewer containers than the strict rule. Switching it off saves 4 more — and with it off, the planner built the dangerous stack itself in 4 of the 10 cases.

We never plan a number we are not sure of — unsure values are flagged or blocked for you to confirm.

Before anything is planned, the confirm card shows what we read: which column is Qty, the size unit, whether a weight is per box or for the whole line. A guess we are unsure of has a dashed frame. A line with no Qty is red and is never planned as one box.

We count four kinds of mistake that must not happen, because nothing on the screen would warn you:

Today's accuracy on documents we had never seen: a final test set of 20 real packing lists — 6 of them original PDFs — with 74 lines, from public records — Colombia public procurement (SECOP) 7, Canadian Food Inspection Agency 5, Brazil environment agency (IBAMA) 4, US freedom-of-information releases 2, Published by journalists 2. The people who write the reader never see these documents; they only get the counts.

• Score: 0.474 out of 1 (first grade on 18 Oct 2026: 0.478). 1 means every Qty, size and weight read right, or asked about on the confirm card.
• The card raised 15 of the 26 warnings these documents call for.
• Before we call import ready, all four counts above must be 0 and the score at least 0.9.

On 18 Oct 2026 we made the safety rule stricter: a warning now counts only when it points at the exact value that may be wrong. Under the stricter rule the confidently-wrong count first rose to 9; it is 0 after the fix.

This is not good enough yet, and we are working on it: the score has not risen since the first grade, 2 lines can still go missing without a mark, and on these unfamiliar documents a customer makes a median of 7 corrections on the confirm card. We will publish each new grade here. Until then, read the confirm card before you plan — every value we used is on it.