Cloth doing a job

A mended streak costs its length only where no lane can dodge it

A broken end is mended when somebody sees it, so a real warp streak runs from where it went wrong to where it was put right, not the whole bolt. Alone, it costs one lane of panels over its own length where no layout can dodge it, and nothing where one can, and a marker free to change its layout part way down the bolt changes neither. Freedom to break matters only for two streaks that can each be dodged and not together, and then it charges their overlap instead of the shorter streak. Over 300 bolts that is 17 of them; a charge of length where nothing dodges and overlap where something does ranks all 300 at 0.99.

Worth reading first: A streak's price belongs to the marker, not the bolt · The width of a bolt is worth what its faults leave it · A fault map is worth most where the grade is worst.

A streak’s price belongs to the marker, not the bolt found that a grade which charged each warp streak by the lanes of panels it really costs a cutter would order a population of bolts almost exactly as a cutting room does. Every streak in that account ran the whole length of its bolt. It ended on the case that assumption leaves out: an end that breaks part way through a piece and is mended when somebody notices, which spoils only the length between. A marker that could change its lane layout at the place where the streak begins would save the length before it. Should such a streak be charged in proportion to its length, or a whole lane regardless? It predicted the first, failing sharply once two streaks overlap along the bolt.

The prediction turns out to be right about the proportion and wrong about what the freedom to break buys. A mended streak is charged in proportion to its length by a marker that never breaks its layout at all, because a lane is already cut along the piece around its own faults. Breaking the layout does something else, and it does it to a particular pair of streaks.

A streak runs from the break to the mend

A warp end that breaks on the loom stops the loom if the stop motion sees it. One that is merely tight, or one whose break is not caught, weaves on and leaves a line along the cloth, and a missing end is a fault the length of the piece only when nobody ever finds it. Most are found, by the weaver walking the loom or by the inspector at the end of the piece, and the end is drawn in and the line stops. So a streak has three numbers: its position across the bolt, where along the bolt it started, and where it was mended.

The bolt here is the one the earlier essays used, 50 metres long and a metre and a half wide, cut into panels 450 by 900 millimetres, three across with 150 millimetres to spare and 55 along, 165 to a bolt. A lane is spoilt where a streak lies inside it, and it is cut along the piece around that stretch, so that a lane crossed by a streak running 20 metres loses only the panels those 20 metres touch.

Two markers are compared. One layout places its three lanes across the width once, for the whole bolt, wherever suits the streaks best. Free to break may cut the bolt across at any place where a streak begins or ends, or at the panel line on either side of one, and lay each stretch with its own three lanes. Both are exact for streaks: the lanes are placed by a short dynamic programme on a five-millimetre grid, and the breaks are chosen by another over the candidate places. Neither sees anything but the streaks.

Alone, a mended streak is charged by its length or not at all

One mended streak, against how long it ran. A single streak on a 50 by 1.5 metre bolt, cut into 450 by 900 mm panels, against the length it ran before it was mended. At 750 mm across no layout of three lanes can dodge it, and it costs 11, 22 and 44 panels at 10, 20 and 40 metres, one lane over its own length, whether or not the marker may break. At 525 mm a layout with its spare between the first two lanes dodges it for nothing at any length. The figure cannot show a streak that wanders across the width.
Fig. 1 One streak, running to the end of the bolt from a stated length before it, at two positions across. At 750 mm no layout of three lanes can avoid it and it costs one lane over its own length: 11 panels at 10 m, 22 at 20 and 44 at 40. At 525 mm a layout with its spare between the first two lanes avoids it for nothing.

The earlier account found that a full-length streak is free in four bands across the width, 40 per cent of it, because the spare 150 millimetres can be left at either selvedge or opened between any two lanes. Everywhere else it costs one lane of 55 panels. A mended streak keeps both halves of that. At 750 millimetres across, in the middle of the bolt where no layout can avoid it, it costs one panel of a lane for every 0.9 metres it ran: 11 panels for a 10-metre streak, 22 for 20, 44 for 40. At 525 millimetres, in the band between the first two lanes, it costs nothing whatever its length.

The freedom to break makes no difference to either. The streak at 750 millimetres spoils whichever lane covers it in any layout, so changing the layout where it begins only moves the loss to another lane. The streak at 525 needs no break, since one layout already dodges it for the whole bolt. So the earlier prediction, a charge nearly in proportion to its length, is exactly right for a lone streak, and a marker that never breaks already pays it. The lane is cut along the piece around the streak, and that was always so.

What the proportion depends on is where the streak lies, not how the marker is allowed to work. A mended streak in a free band is free; one anywhere else is charged by its length. The charge has no lane-sized lump in it at all, unless the streak runs the whole bolt.

The four-point grade gets the proportion right for once

The trade’s own scheme already charges a mended streak by its length. A grade charges by the length and a cutter pays by the panel set out how: every fault scores one to four points by how far it runs, and no linear metre may score more than four. A streak is longer than nine inches in every metre it crosses, so it scores four points a metre, and a streak that ran ten metres before it was mended scores forty points, wherever it lies.

That is the right proportion and the wrong condition. The cutter pays one panel of a lane for every 0.9 metres of a streak in the middle 60 per cent of the width and nothing for one in the free bands, so the scheme’s charge is exact in proportion for three streaks in five and wholly wrong for the other two. The full-length streak was the scheme’s worst case because a single streak filled its cap in every metre and nothing else on the bolt could be seen; a mended one leaves the metres before and after it free to score their own faults, which is better, and still charges a free streak as heavily as a costly one.

The repair is the one the full-length account proposed, applied to a shorter object. Record the streak’s position across, and its start and end, which an inspection frame that maps faults already writes down. A fault map is worth most where the grade is worst, and here the map is three numbers a streak.

There is a pleasing echo in what breaking means. That earlier essay compared a cutter who slides a rigid tiling along a lane with one who lets the tiling break around each fault, and breaking along the lane recovered twelve panels where sliding recovered two. Breaking along the lane is already in every marker here: each lane is cut around its own streak, which is why a lone mended streak costs its length and not a lane. Breaking the layout across the width is the same freedom in the other direction, and it is worth far less, because across the width there is only 150 millimetres of room to use it in. The width of a bolt is worth what its faults leave it found the width worth between nothing and two fifths of the length against every kind of fault; against mended streaks it is worth something on one bolt in eighteen.

Two streaks that can be dodged one at a time

A break matters only when one layout cannot do for the whole bolt what two layouts could do for its parts. That needs two streaks, each of which some layout can dodge, which no single layout can dodge together, and which do not both run the whole length.

Put one streak 100 millimetres from one selvedge and another 100 from the other. Each lies in a free band, and each can be dodged by leaving at least 100 millimetres of the spare at its own selvedge. Together they need 200 of a spare that is only 150, so one layout must leave one of them inside a lane.

Two mended streaks, a marker that keeps its lanes and one that may break them. A 50 by 1.5 metre bolt cut into 450 by 900 mm panels, three across, with one streak at 100 mm across running from the start to 30 metres and a second at 1400 mm running from a point along the bolt to its end. Either can be dodged alone and not both. Panels lost when the second begins at 10, 20, 30 and 40 metres: a marker that keeps its lanes, 33, 33, 22, 11; one that may change its layout where a streak begins or ends, 22, 11, 0, 0. The second follows the overlap of the two streaks, the dashed line, and the first follows the shorter streak. The figure cannot show what a break costs on the cutting table.
Fig. 2 One streak 100 mm from one selvedge, from the start of the bolt to 30 m, and a second 1,400 mm across from a stated point to the end. A marker held to one layout loses a lane along the shorter of the two: 33 panels until the second streak begins after 20 m, then falling. A marker free to break loses a lane only along their overlap, the dashed line, and nothing once they no longer overlap.

Let the first streak run from the start of the bolt to 30 metres and the second from some point along it to the end. Held to one layout, the marker loses a lane along the shorter streak. While the second begins before 20 metres it is the longer, so the marker dodges it and loses 33 panels along the first. After 20 metres the second is the shorter and the marker gives it up instead, and the loss falls with its length to nothing at the end of the bolt.

Free to break, the marker loses a lane only where both streaks run together. It lays the bolt with its spare at one selvedge while the first streak runs alone, cuts across, and lays the rest with its spare at the other. If the second begins at 20 metres the overlap is 10 metres and the loss is 11 panels, against 33 held to one layout. If the second begins at 30 metres or later the two never overlap and the loss is nothing, where one layout still pays 22 panels at 30 and 11 at 40.

Two mended streaks in plan, cut with one layout and with a layout for each stretch. A 50 by 1.5 metre bolt drawn in plan, its length across the page, with 2 streaks: 100 mm across from 0 to 30 m and 1400 mm across from 20 to 50 m. Held to one layout the marker dodges one streak and loses a lane along the other, 33 panels of 165. Free to change its layout, it cuts the bolt across at 19.8 m and gives each stretch its own spare, losing 11. Outlined blocks are runs of sound panels. The plan cannot show the labour of re-laying a marker part way down a table.
Fig. 3 The same two streaks, the second beginning at 20 m, drawn in plan with the length across the page. One layout dodges the second streak and loses a lane along the first, 33 panels. Free to break, the marker cuts the bolt across at 19.8 m, puts its spare beside the first streak before the cut and beside the second after it, and loses a lane only over the 10 m where both run.

The cut falls at 19.8 metres, on the panel line just before the second streak begins, rather than at 20. That is the reason the candidate places include the panel lines beside each streak’s ends: a cut that falls between panel lines strands the part of a panel on each side of it, and over a 50-metre bolt with 0.5 metres of remnant at its end, a cut on the panel line costs nothing at all.

So the earlier prediction’s second half has a precise form. Breaking changes a conflicting pair’s charge from the shorter streak to their overlap. The pair’s loss does fail sharply in one sense, since it is nothing until they overlap and then rises a panel for every 0.9 metres of overlap; but that is the breaking marker’s charge, and a marker held to one layout never shows it.

The streaks that cannot be helped are most of them

That is a large effect on the right pair. How often is a pair the right pair? A population answers it, and the one used here is chosen rather than taken from any mill. Three hundred bolts carry two or three mended streaks each, seven in ten with two, every streak at a position across drawn evenly, starting at a point drawn evenly in the first 90 per cent of the bolt and mended after a run drawn evenly up to the rest of it. The streaks average 14.4 metres long.

What freedom to break a layout recovers, over 300 bolts with mended streaks. The panels a marker free to change its layout recovers over one held to a single layout, bolt by bolt, for 300 bolts of 50 by 1.5 metres with two or three mended streaks each: nothing, 283 bolts; 1 panel, 1 bolt; 2 panels, 4 bolts; 3 panels, 2 bolts; 4 panels, 1 bolt; 5 panels, 1 bolt; 6 panels, 2 bolts; 9 panels, 1 bolt; 11 panels, 1 bolt; 12 panels, 1 bolt; 14 panels, 1 bolt; 18 panels, 1 bolt; 20 panels, 1 bolt. On average the one layout loses 21.6 panels a bolt and the breaking marker 21.2. The figure cannot show what a break costs in a cutting room's time.
Fig. 4 What a marker free to break recovers over one held to a single layout, bolt by bolt, over 300 bolts with two or three mended streaks. It recovers nothing on 283 of them; on the other 17 it recovers between 1 and 20 panels.

On 283 of the 300 bolts, breaking the layout recovers nothing. Held to one layout the marker loses 21.6 panels a bolt on average, and free to break it loses 21.2. Every one of the 17 bolts where breaking helps carries a pair of streaks that can each be dodged and cannot be dodged together, which was checked bolt by bolt, and there breaking recovers between one panel and twenty.

The reason most bolts gain nothing is the 60 per cent of the width where no layout dodges a streak. Of the 694 streaks drawn, 423 lie in a position no lane can avoid, so they cost their own length under any marker, and breaking cannot touch them. Of the rest, most either have no conflicting partner, or have one that a single layout can dodge with them, since a spare of 150 millimetres can be split among several gaps; two streaks at 60 and 525 millimetres, for instance, can both be dodged by lanes that start just past each. A conflicting pair needs its two streaks to want more spare between them than there is.

Two mended streaks in plan, cut with one layout and with a layout for each stretch. A 50 by 1.5 metre bolt drawn in plan, its length across the page, with 2 streaks: 919 mm across from 31 to 49 m and 1366 mm across from 3 to 30 m. Held to one layout the marker dodges one streak and loses a lane along the other, 21 panels of 165. Free to change its layout, it cuts the bolt across at 29.7 m and gives each stretch its own spare, losing 1. Outlined blocks are runs of sound panels. The plan cannot show the labour of re-laying a marker part way down a table.
Fig. 5 The bolt in the population on which breaking recovers most, in plan. Its two streaks, at 1,366 mm across from 3 to 30 m and at 919 mm from 31 to 49 m, can each be dodged and not together. One layout loses a lane along one of them, 21 panels; free to break, the marker cuts across near 30 m and loses 1.

The bolt on which breaking helps most is the clean case drawn whole. Its two streaks lie where each needs the spare in a different place, and they do not overlap at all: one runs from 3 to 30 metres and the other from 31 to 49. One layout loses 21 panels and a broken layout loses one, the single panel stranded at the cut. A cutter who could see that would re-lay the marker once, at 30 metres.

What a grade should charge

The earlier account’s grade charged each streak by the lanes of panels it cost a cutter who knew its position, and that put a population of full-length streaks in very nearly a cutting room’s order. With mended streaks the same idea needs two terms, and both can be read off the streaks without running a marker.

Charge each streak no lane can dodge by its own length, one panel of a lane for every panel length it ran. Charge each pair that can be dodged one at a time and not together by their overlap. Leave every other streak free.

300 bolts with mended streaks, lost panels against a charge read off the streaks alone. 300 bolts of 50 by 1.5 metres, each with two or three mended streaks at positions drawn evenly across and along, cut into 450 by 900 mm panels. A marker held to one layout loses 21.6 panels a bolt on average and one free to break 21.2; breaking recovers something on 17 bolts, 6 per cent. Against the breaking marker's loss, a charge of each undodgeable streak's own length and each conflicting pair's overlap ranks the bolts with Spearman 0.988, and the streaks' total length with 0.507. Each dot is a bolt; the diagonal is equality. The population's rates are chosen, not a mill's.
Fig. 6 The 300 bolts at the panels a marker free to break loses against that charge. Nearly all lie on the diagonal; the ones below it carry two streaks no lane can dodge that share a lane and spoil it once. The charge ranks the bolts at Spearman 0.988, and the streaks’ total length at 0.507.

Against the breaking marker’s loss, the charge ranks the 300 bolts at Spearman 0.988. The streaks’ total length, which is what a grade measuring the length of each fault would see, ranks them at 0.507. The bolts that fall below the diagonal are a fault of the charge’s simplicity rather than of its form. Two streaks that no lane can dodge and that lie in the same lane spoil that lane once over their overlap, not twice, as the full-length account found for streaks close together, and a charge that adds their lengths overcounts.

So a mended streak’s price is its position first and its length second. Position decides whether its length is charged at all; length is the price only where position says it must be paid; and the freedom to change layout down the bolt adds a third term that matters on about one bolt in eighteen.

How it was computed

A streak is a position across and a start and end along. A lane at a position across is spoilt by a streak lying in the 450 millimetres it covers, over the length the streak runs, and its panels are counted greedily along each clean run between spoilt stretches, which is optimal within a lane. A layout is three lanes on a five-millimetre grid that do not overlap, placed by a dynamic programme to maximise the panels; a broken layout is a sequence of stretches cut at the candidate places, each with its own layout, chosen by a second programme over the cuts. The candidate places are every streak end, and the panel lines immediately before and after each.

The population is drawn from a fixed seed so every figure shows the same 300 bolts. The charge counts each streak outside the four free bands at its length over 900 millimetres, and each pair inside the free bands at their overlap, if a single layout cannot dodge them both; that test runs the layout programme on the two streaks alone.

Required of it and not shown: that a lone streak at 750 millimetres running the last 30 metres cost 33 panels with or without breaks; that the pair above cost at least 33 held to one layout and at most 12 free to break; that a broken layout never do worse than a single one; and that a streak which ends before it starts be refused.

What a marker of streaks leaves out

Everything but streaks. Point faults and weft bars, the across-the-width shape a mispick or a stop mark leaves, were left off every bolt here, because the question was about streaks. On a real bolt they move the layout and they make the remnant at a cut matter, since a cut placed to suit a streak may land where a bar would have been free. The width essay’s marker handles both, and the breaking marker could be joined to it; that has not been done.

What a break costs in a cutting room. A cut across the bolt part way down a spread is cheap in a cutting room that lays one marker at a time and expensive in one that lays a long marker over many plies. Nothing here prices that, and a cutting room that charged a break at even a few panels would break only on the bolts at the top of the gain chart.

Where an end breaks. The population draws positions across evenly, and ends do not break evenly. A warp breaks in a tail no tensile test reaches, at its weakest places, and those meet the loom’s hardest work at particular ends rather than everywhere. A real population would weight positions, and how much of it fell in the free bands would decide the whole answer.

And the rates are chosen. Two or three streaks a bolt, each mended after a run drawn evenly, is a population built to have conflicting pairs in it, and a mill’s own records could hold far more or far fewer.

Who measured what

Mending broken ends, and finding the streaks that were not caught at the inspection frame, is ordinary weaving practice, and cutting rooms do lay markers around known faults. The four-point system’s treatment of a streak, its length bands and per-metre cap, is the trade’s own.

What is added here is the reading of a mended streak as a position and an interval; the result that alone it is charged by its length or not at all, whether a marker can break its layout or not; the exact charge of a conflicting pair, the shorter streak held to one layout and the overlap free to break; and the population result that breaking matters on 17 bolts of 300, while a charge of length where nothing dodges and overlap where something does ranks all of them at 0.99.

Still open: whether a mill’s broken ends fall where a lane can dodge them

Everything here turns on one share: how many streaks lie in a free band. With panels 450 millimetres wide on a 1,500-millimetre bolt it is 40 per cent of positions if ends break evenly across the width. They do not. Selvedge ends are commonly reported to break more often than body ends, since the temple and the weft’s turn work them harder, and a figure is harder on its warp for the ends a figured harness flexes most.

If a mill’s streaks bunch at the selvedges, most of them fall in the two outer free bands, and they conflict with one another only in pairs at opposite edges. The breaking marker’s share would then rise sharply, since opposite-edge pairs are exactly the pairs it helps. A record of where along the width a mill’s mended ends lay, which an inspection frame that marks fault positions already makes, would say which; and it would say, for that mill’s own panel widths, whether a cutting room gains anything by being allowed to cut across a spread.

Named objects

A flat tag is an object no other essay names yet.

Fault mapMarkerMissing endPanelYield