An hour out of Southampton the car carrier was on its side.

•  A car carrier sailed from Southampton on estimated weights.

•  Every Land Rover on board was entered at two tonnes.

•  The real weights sat unused in the bar codes.

•  The ship went over on the first turn.

Hoegh Osaka was a pure car and truck carrier, Singapore flagged, built in 2000, 179.9 metres long and 51,770 gross tons. It arrived at Southampton on 2 January 2015 with 932.6 tonnes already on board, and worked cargo at berth 40 from 0621 the next morning until 1750, with four more cars going on at 1857.

The plan sent to the master the day before listed 1,306 units weighing 4,008.9 tonnes. The tally handed over when loading finished listed 1,418 units weighing 4,625.6 tonnes. The extra 112 units came to 616.7 tonnes, and the ship was told about them only at the end.

None of that looked like trouble. The crew had a loading computer, a stability manual, a cargo securing manual and a chief officer who had run a departure sum that morning. The sailing itself was routine: a pilot aboard, two tugs, a channel the ship had used many times.

Every row below comes from one document: Accident Investigation Report 6/2016, published by the Marine Accident Investigation Branch in March 2016. After the event the Branch took a selection of the used trucks off the ship and put them across a certified weighbridge. The table is that weighing, and column one is the gap in per cent between the paperwork and the scale.

Read the last row before the total. A refuse truck was booked at 8.5 tonnes and weighed 14.16. Nothing about it was hidden: it had a shipping note, a sticker and a declared weight, and every one of those said 8.5. The scale was the only thing that disagreed.

Four things stood behind this, and the turn itself was the last of them.

•  The weights were estimates by policy. New cars were given estimated weights rather than their own. Every Land Rover was entered at two tonnes. A Range Rover is 2.513, and across the cars that gap came to about 345.4 tonnes.

•  The real figures already existed. Every new car carried a bar code holding its actual weight, and the tally read each one electronically as it came aboard. Those figures were never used in the tally the ship was given.

•  The ballast numbers were estimates too. Tank quantities were worked out on board and differed from the real levels. It was practice on board to alter the readings so the computer output matched the draughts the crew could see.

•  The last sum was never done. The morning calculation used the plan, not the load, and returned a metacentric height of 1.46 metres. No stability calculation was completed after cargo work finished and before the ship sailed.

At 2109 the ship was rounding West Bramble buoy when it went over to starboard past 40 degrees, the steering and the propulsion went, and it drifted aground six minutes later. Cargo broke its lashings and punched a hole through the shell plating, and deck 6 took water once it was under.

All 25 on board came off. The pilot, the master and the chief officer stayed until 0209 the next morning and left by helicopter. There was no pollution. Salvors refloated the ship and brought it to a berth in Southampton on 22 January.

The Branch sent recommendations to the Maritime and Coastguard Agency, the Association of European Vehicle Logistics and the International Chamber of Shipping. Its chief inspector put it in one line: the assumed distribution of ballast bore no resemblance to reality.

•  An estimate nobody reconciles becomes a fact. Two tonnes a car was an assumption never checked against a scale. By sailing day it was what the ship believed it carried.

•  The true number is often already in the building. The bar codes held the real weight of every car and the tally read them. The number existed and was left out of the sheet that mattered.

•  The last check is the one that counts. A sum run in the morning described a load that never arrived. The one that would have caught this was the one due after the ramp came up.

•  Some things carry their own proof. Gold is the other case. One bar states its weight and its assay on itself, and a scale settles the question without a second document.

I ran a desk in 1998 where one position was carried at a modelled mark for eleven weeks because the screen holding the traded figure sat on a system nobody opened.

Ask of any holding which of its numbers came off a measurement and which came off a convenient assumption.