A stripe is paid for by the threads that cross it
Worth reading first: A float limit gives a blind set its arrangement back · The colour order that hides least · Designing to a float limit.
A float limit gives a blind set its arrangement back found that two colour orders with the same counts, which separate exactly as many drafts as each other without a limit, come apart by a factor of eight inside one. A four-and-four order used both ways keeps 12 per cent of its bound under floats of at most three, and a two-and-two keeps all of it. The four-and-four loses because each of its stripes leaves a stretch of four cells on every crossing thread that no blind cell can reach, and one pattern in eight of such a stretch lies on one face throughout. Sixteen threads each surviving with seven chances in eight gives (7/8)¹⁶, 11.8 per cent, which the sampling confirmed.
That account applied one limit to the ends and the picks alike, and said so. Few cloths are specified that way. A weaver who limits floats does it because a long float snags, rubs or lets a thread shift, and the two systems are often different yarns doing different jobs. With the limits separated, the stretch argument says exactly which one a stripe answers to, and the answer is not the obvious one.
Which threads carry a stripe’s stretch
A blind cell is where a pick crosses an end of its own colour. There the cloth shows that colour whichever thread lies on top, so the surface cannot tell the draft’s two values apart and a designer has nothing to see. Everywhere else the cell is free: the two threads differ in colour and the surface records which is uppermost.
Follow one pick across the cloth. It is one colour throughout, so its blind cells are at the ends of its own colour and its free cells are at the ends of the other. The free cells on a pick fall exactly where the warp order’s stripes of the opposite colour lie, and a warp stripe four ends wide is a stretch of four free cells along every pick of the other colour. Follow an end down the cloth and the same holds the other way round: its free cells are the weft order’s stripes.
A float limit is a limit on a thread’s runs, so it is the pick’s limit that applies along the pick. A stripe in the warp is charged to the limit on the weft, and a stripe in the weft to the limit on the warp. It is the crossing threads, not the striped ones, that carry a stripe’s stretch, and they are held to their own system’s limit.
That inverts what the words suggest. A warp stripe is made of ends, coloured by the warper, and a designer thinking about a stripe of ends naturally thinks about the ends’ floats. Those floats are irrelevant to it. A stripe of four dark ends in a warp whose picks may float freely costs the catalogue nothing through its own stretch, however tightly the ends themselves are held.
One stretch, counted exactly
The damage a stretch does is a small count. A stretch of s free cells is repaired by nothing, since the blind cells that could break a run lie outside it, at its two edges, where they can stop a run from continuing but not from starting. So a stretch passes only if its own pattern holds no run longer than the limit.
A stretch no wider than the limit loses nothing, since no run inside it can be long enough. One cell wider, it loses exactly two of its patterns, the one with every cell on the face and the one with every cell on the back: 14 of 16 survive a stretch of four under a limit of three. Two cells wider, more patterns fail, and the share falls away quickly; under a limit of two, a stretch of nine keeps 110 of its 512 patterns.
Every thread a stripe crosses carries one such stretch, and the stretches on different threads share no cells. So under the two-colour orders, with each thread surviving on its own, the share of the bound a check keeps is a product: one factor for each pick, from the warp order’s stripes and the picks’ limit, and one for each end, from the weft order’s stripes and the ends’ limit. For the four-and-four order under a limit of three on one system, eight crossing threads at fourteen sixteenths each give (7/8)⁸, 34.4 per cent.
Two limits, two factors
The four-and-four check in both directions puts that to a test, because the product form predicts a table with a particular shape.
At three on both systems the check keeps 12 per cent, the figure the earlier essay found. Free either system and it keeps about a third: 32 and 33 per cent with the other at four, 34 and 37 with it unlimited, all within sampling error of (7/8)⁸. At four on both, every stripe fits and it keeps 94 per cent; with either system unlimited and the other at four, all of it. The table is the product of a row factor and a column factor, nearly, and the two factors belong to the two systems.
Now the asymmetric check. Four-and-four in the warp and two-and-two in the weft makes the same thirty-two blind cells as the square check does, but its stretches differ by direction: four along every pick, two along every end. Under a limit of three on the ends and four on the picks, the picks’ limit admits their stretch of four and the ends’ limit admits their stretch of two, so nothing in the closed form fails, and the samples keep 97 per cent. Turn the check a quarter, putting the wide stripe in the weft, and the stretch of four now lies along the ends, under their limit of three: (7/8)⁸ again, 34 per cent.
Same colours, same counts, same blind cells, same limits; a factor of nearly three. The only decision that differs is which way the cloth goes through the loom.
Why the closed form is a ceiling
A broken stretch is a broken cloth, whatever happens elsewhere, because no blind cell lies inside the stretch to break its run. So the closed form, which counts only those failures, can only overstate what survives. It cannot understate it.
That is what the sampling shows. Of thirty-two combinations, none lies above the diagonal by more than its sampling error, and most lie on it. A few lie a little under, by a few per cent, and three lie far under, at around half the bound where the closed form allows all of it.
The ones far under are the other mechanism the earlier essay met and could not price. A blind cell is shared by a pick and an end, and it can take only one value. When a thread’s own limit is tight enough to need a particular value in a blind cell, to break a run that its free cells have started, and the crossing thread needs the opposite value there, nothing satisfies both, and a surface that passes every stretch test still has no draft.
A limit the stripes all fit, which still costs half
The cleanest example holds every stripe inside its limit and squeezes only the system with the finer stripe.
A warp of four-and-four with picks held to four fits its stripe exactly. With a weft of two-and-two, every end’s stretches are two cells long, so an ends’ limit of two fits them too, and the closed form keeps the whole bound. The samples keep 48 per cent. With a weft of one-and-one the ends’ stretches are single cells, and a limit of two still costs 44 per cent: 56 kept. From a limit of three upward both checks keep 96 per cent or more.
The reason is that a stretch which only just fits its limit forces its neighbours. A pair of free cells on an end that happen to be the same value — two faces running — must be stopped at both edges, so the blind cells above and below them are each forced to the other value. Each of those blind cells sits on a pick, inside that pick’s blind stretch of four, and the pick has its own run to keep under four. Forced values arriving from four different ends can line up into a run the pick cannot hold, and then the surface is lost. A limit equal to a stripe is not slack: it is the limit at which the stripe’s edges start dictating to the other system.
Where the two limits differ
A limit is set by what a float will do in use, and the two systems rarely do the same thing. The float decides most of how a cloth wears, and floats and abrasion found that a long float fails badly rather than quickly, and it is the face that is rubbed. In a warp-faced cloth that face is the warp’s, so a designer limits the ends tightly and can be more generous with the picks, whose long floats lie mostly on the back. A cloth with a fancy weft — a slub, a bouclé, a lofty wool — reverses it, because the weft is the yarn that snags. Designing to a float limit priced a single limit’s cost to a catalogue; a limit that differs by system is the more ordinary specification, and it is the one under which a check has a better and a worse way round.
The structures that carry their asymmetry in the weave itself make the point without any colour at all. A cord is a stripe with no colour in it: a warp rib’s ends run in pairs while its picks alternate singly, so its own runs are already longer along one system than the other, and a colour order laid over a rib starts from a catalogue that is unequal by direction before any limit is written. A stripe is a partition of the warp treated the stripe as a fact about ends. Here it is a fact about the picks as well, and that is the half a float limit reads.
What this changes for a designer
The earlier essay gave a designer working to one limit a third lever besides the blind count and its arrangement: keep every stripe inside the limit and prefer the regular ones. With separate limits, the lever splits in two and the rule for each half is crossed.
A coarse stripe belongs in the system whose crossing threads may float longer. If the weft is the yarn allowed long floats — a lofty weft on a firm warp, or a cloth whose face is the warp and whose back can carry weft floats nobody sees — then broad stripes of colour should be put into the warp, where they are charged to the weft’s generous limit, and kept out of the weft, where they would be charged to the warp’s strict one.
The finer stripe should not be set at the tight limit. Under a limit of two on the ends, a weft of two-and-two loses half its drafts to competition even though it fits; a limit of three costs it three per cent. A specification that sets the tight limit at exactly the narrower stripe’s width has chosen the one value where the stripe dictates to the other system.
Turning a check is not neutral. A check is two stripes and a tartan is one, and as colour that is symmetric; under two limits it is not, and the check reads the same to an eye either way round while its catalogue differs by a factor of three. A mill that weaves a check sideways to fit a width, or turns a cloth for cutting, has changed which drafts the colour order can carry.
How the catalogue was sampled
The sampler is the one the float-limited counts used, extended so that a pick’s run is held to the picks’ limit and an end’s run to the ends’. A surface is a random filling of the free cells, at eight hundred surfaces a point; it is tested thread by thread against the free cells already known, and then a search fills the thirty-two blind cells one by one, abandoning a filling as soon as any pick or end breaks its own limit. A surface counts if some filling survives. Runs are counted cyclically round the eight-thread repeat, as everywhere in the catalogue these counts are taken from.
The closed form is computed exactly: the count of binary strings of each length with no run longer than each limit, multiplied over every thread’s free stretches. Required of the sampler: that with equal limits it return what the earlier counts returned, 12 per cent for four-and-four under three; that it never exceed the closed form by more than its sampling error; and that the four-and-four by two-and-two check keep nearly all its bound one way round and about a third the other.
What the counting does not reach
The repeat is eight threads. Every check here is a single repeat of eight by eight, so a stripe of four is half the repeat. At a longer repeat, stripes of the same width cross more threads, the product has more factors and the loss compounds: a four-end stripe under a limit of three across a sixteen-pick repeat costs (7/8)¹⁶ on its own system. The turning effect only grows.
Only regular orders are drawn. Four-and-four, two-and-two and one-and-one are the orders whose blind cells fall most evenly. The earlier essay found irregular orders losing drafts to competition even when their stripes fit, and under separate limits that competition will move with the limits in ways the four checks here do not sample.
A float limit is still a straight line through a judgement. A weaver’s real tolerance for a float depends on its yarn, its position and whether it is on the face, and a limit that is one number per system is the coarsest version of that. The arithmetic says what a designer gives up under a stated rule, not what the rule should be.
Counting is not seeing. Colour and weave is about what an eye sees, and two surfaces differing in one free cell count as two here. Whether the drafts a turned check loses are drafts anyone would have chosen is not a question a count can answer.
Who worked out which part
Float limits and colour-and-weave effects are old working knowledge, and checks woven sideways are ordinary mill practice. The blind count, the effect of a float limit on it and the four-and-four’s closed form are the earlier essays’ own.
What is added here is the separation of the two limits and its consequence. A stripe’s stretch lies along the threads that cross it, so each system’s limit charges for the other system’s stripes; the share kept is then a product with one exact factor per crossing thread, which is a ceiling on the true share; and the true share falls well under the ceiling only where a limit is as tight as the stripe it governs. None of this is known to have been written down before, which is not the same as saying nobody in a design room has noticed that a check weaves better one way round.
Still open: whether a longer repeat keeps the turn
Everything here is at eight by eight, where a stripe of four is a large part of the cloth. At sixteen threads a four-and-four order has four stripes in each direction, every crossing thread carries two stretches, and the closed form for one system becomes (7/8)³². The turning factor between a check and its quarter-turn should then grow from about three to about eight, since the losing orientation’s factor is squared.
Whether it does is a sampling question, and it is a harder one than the eight-thread case: a sixteen-by-sixteen draft has 128 blind cells under these orders, and the search that fills them abandons most fillings early only when the limits are tight. A sampler fast enough to run it would also answer the question these counts have carried since the four-by-four census, which is whether the competition between blind cells, the part the closed form cannot see, grows or fades with the repeat.
What links here
Computed from the collection rather than written here: the essays that point at this one.
Shares its objects with
Essays naming at least two of the same things, that neither author linked.
- A blind cell costs half the cloths at any real repeat — both name blind intersection, census, colour-and-weave, colour order, draft
- A colour-and-weave look costs its cheaper order — both name census, colour-and-weave, colour order
- A weft stripe is counted in pairs of picks — both name census, colour-and-weave, colour order
- No weave draws an unbroken line one thread wide — both name census, colour-and-weave, colour order
- The finest colour-and-weave effects need the rarest loom — both name census, colour-and-weave, colour order
- A balanced twill need not have two equal faces — both name census, float length
Named objects
A flat tag is an object no other essay names yet.
Blind intersectionCensusColour-and-weaveColour orderDraftFloat length