A digital material is two halves, and one of them is usually missing

The situation

Thistlewaite's material library holds 88 digital materials, built over four seasons. Every one of them has been scanned: photographed under controlled light on the studio's own scanner, so the colour, the weave, the surface roughness and the way light catches the twill line are all captured at the right physical size. They look superb.

29 of the 88 carry a measured set of physical properties. The other 59 carry a preset copied from a fabric somebody judged to be similar.

That sentence is the whole lesson. A digital material is two halves — what it looks like and how it behaves — and only the first half is visible when you decide whether the material is any good.

The two halves, listed

The appearance half is easy to buy, quick, and immediately rewarding. You scan a fabric and a beautiful swatch appears. The behaviour half costs money, takes days, produces a sheet of numbers nobody enjoys reading, and changes nothing you can see in a still image.

So it gets skipped. And it is the half that decides whether the virtual sample is worth anything, because a garment's shape comes from the cloth's stiffness and weight fighting gravity. Course 22.2 approves fit and drape on virtual samples. This is the data those samples run on.

An empty field is not empty

Here is the mechanism that makes this hard to see, and it is worth stating carefully.

When a physical property is not measured, the field does not stay blank. Software cannot simulate cloth with a blank stiffness, so it uses a value. A default, or whatever was inherited from the preset the material was copied from. The simulation then runs perfectly, produces a smooth result, and reports no problem, because from its point of view there was no problem. It had a number.

A default is a number the software chose, shown as a number somebody measured. There is no flag on it, no shading, no asterisk. And a reviewer looking at a virtual sample cannot tell a measured drape from an assumed one, because both look like cloth.

That is why "we never got a wrong number" is a true statement about a library of presets, and a meaningless one. Nobody got a wrong number. Everybody got an unowned one.

The evidence, read the right way

Last season Thistlewaite settled 17 styles on virtual samples. Afterwards the technical team counted how many of those needed an extra physical round, because the sample did not hang the way the screen had shown.

Read that table the wrong way and you get a weaker finding: five of the six failures were on preset materials. True, and nearly useless. Most of the library is presets, so most of anything would be. The share of failures tells you about the library's composition, not about the risk.

Read it as rates and the finding holds: 9.1% against 83.3%, on comparable styles run through the same process by the same people. That is the number worth quoting, and it is worth quoting carefully. 17 styles is a small sample, one measured failure against eleven is a thin denominator, and the technologist who chose which fabrics to measure may well have measured the difficult ones. Say all of that when you present it. It is still a factor of nine.

An extra physical round on this brand costs USD 340 and 9 days.

The arithmetic, and why the money answer is the wrong one

Measuring a fabric's physical properties costs Thistlewaite USD 138 and 6 working days at the laboratory it uses.

Two things fall out of that table, and the second is the important one.

First, measure by re-use, not by importance. 20 of the 59 preset materials appeared in more than one season. Those 20 carry most of the exposure and cost a third as much to fix, and the payback more than halves. A material record is bought once and used for as long as the fabric is bought, which makes frequency of use the right sorting key. Not how expensive the garment is, and not how much the designer likes the fabric.

Second, the money case is the weak case, and you should not lead with it. USD 3,400 a year against a library that costs USD 8,142 to complete is a payback of over two years, which any finance director can decline without being wrong. The real cost of a preset is the 9 days an extra round adds to a style that had already been approved, arriving in the week the fabric was supposed to be booked. Days are what a virtual sampling programme is bought to save, and days are what a preset quietly gives back.

The specification to write

A digital material is worth asking for the way you would ask for a fabric: as a list somebody can supply against.

1. A scan at a stated physical scale, so the weave renders at its real size. A texture with no recorded size renders the grain at whatever the software assumes, and a twill line twice its true width is a garment nobody will recognise.
2. The colour on the same terms as lesson 1 — measured, with its conditions, not a value read off the scan.
3. The mechanical properties measured on the finished bulk quality, on named instruments, with the test conditions recorded.
4. The state of the record — measured or inherited — as a field, visible to everybody who opens it.
5. The date and the fabric reference the record was built from, so it can be retired when the quality changes.

Item four costs nothing and changes behaviour. If a reviewer can see that a material's drape is inherited, they stop treating the virtual sample as settled — and that judgement, made in February, is worth more than any of the measurements.