Lessons · Lesson 3 of 6
- 01 · What a pattern carries that a drawing does not
- 02 · A wrong rule is not a wrong pattern
- 03 · What the nester optimises, and what it was never told
- 04 · Automatic against a marker maker: where each one wins
- 05 · The block library nobody maintains
- 06 · What survives a handover, and what silently does not
What the nester optimises, and what it was never told
State a marker-making problem the way the software actually receives it, and find the constraints that are missing rather than wrong.
Lesson 3 of 6 · 18 min
2 April, the cutting order for LR-7714
A marker is the cutting plan: every pattern piece laid out on a strip of cloth of a stated width. A nester is the software that makes one automatically. The nest ran overnight and came back at 84.7%. The marker room released it, the cutting room spread it, and nothing went wrong for four months.
Course 4.5 covers what marker efficiency measures, what a point of it is worth, and how a cutting room turns it into metres. This lesson is about the other half: what the software was actually asked, and what it was never asked at all.
The problem, as the software receives it
A nester is given a strip of fixed width and a set of shapes, and told to make the strip as short as it can. Everything else you believe about the marker has to reach the software as either a piece attribute or a profile setting. The difference between those two matters more than anything else in this lesson.
| The instruction | Where it lives | What happens if it is blank |
|---|---|---|
| Which pieces are in the marker | the piece list | a piece is silently in or silently out |
| What fabric a piece is cut in | an attribute on the piece | the piece joins whichever marker is running |
| Whether a piece may be turned end for end | an attribute, or a profile default | the default decides, and the default is the last style's |
| How much a piece may be tilted | an attribute, or a profile default | the default decides |
| Whether a piece may be mirrored | an attribute on the piece | the default decides |
| Buffer around each piece | a profile setting | the last value used stays used |
| Stripe or check matching | matching points on pieces | it is not attempted |
An attribute on a piece travels with the piece. A profile setting belongs to the session, and it stays behind when the style changes. A constraint that lives in the profile rather than on the piece is a constraint you will inherit — from a style you have forgotten, on a fabric that behaves differently.
Three pieces that were never in the twill
CH-482 has 23 pieces: 19 in twill and 4 in pocketing — two pocket bags, and two front-pocket facings. Three of those four came in from the digitising bureau in lesson 1 with the fabric field blank. Nobody filled it in, because a blank field is not an error message.
A blank fabric field is not a question. It is an answer, and the answer the nester hears is include it. All three went into the twill marker.
Nothing looked wrong. The pieces are small and tidy and they fill gaps, so the marker's reported efficiency actually improved, from 84.6% without them to 84.7% with them. The marker room's own quality measure moved in the reassuring direction.
What it cost
The three strays add 0.090 m to the twill marker for every pair. In the pocketing marker, which is narrower, the same three pieces would have needed 0.105 m a pair.
in twill: 0.090 m at USD 3.95 = USD 0.3555 a pair
in pocketing: 0.105 m at USD 1.10 = USD 0.1155 a pair
difference = USD 0.2400 a pairOn 24,600 pairs that is USD 5,904.00, against an order margin of USD 23,616.00. A quarter of the margin, from three blank fields.
The physical trail is worth following, because it names the signal that was available in May. The twill store issued 2,214 m more than the marker was supposed to need. The pocketing store issued 2,583 m less than its booking — 17.4% under.
An overrun gets investigated. An under-consumption gets filed as good news. It is the same event seen from the other side, and it was the only visible evidence of the fault while the order was still in the building.
The five things it cannot be told at Karnaphuli
Every nester has a boundary. The honest way to work with one is to know exactly where that boundary is, rather than assume the software thinks like a marker maker.
- A nap or a directional face. This one it can be told, as an allowed-rotation set per piece: no turn, or a half turn, or free. It has no way to discover it for itself. Course 4.5 prices what one-way costs.
- A stripe or a check to be matched. Matching is a constraint between two pieces at a repeat, and Karnaphuli's nester cannot express it. A matched marker here is human work, and lesson 4 puts a number on that.
- A shade band or a roll width. The nester makes a marker for one stated width. Which rolls that width is true of is the cutting room's problem, and course 4.5 covers it.
- A fabric fault. It places pieces on a perfect rectangle. Cloth is not one.
- Which pieces must come off the same ply. Two front panels from different plies can read as two shades in daylight. That is a constraint about bundles, and it lives outside the marker entirely.
| What the nester was told | What it assumed | What nobody could tell it |
|---|---|---|
| the piece list, including 3 strays | fabric, where the field was blank | that the twill is a plain solid |
| the cuttable width | a half turn allowed, from the profile | which rolls run narrow |
| minimise length | the previous style's buffer | which pieces must share a ply |
Prompt · Write the constraint brief before the nest runs
Before an automatic marker run on a style whose fabric, pieces or size mix differ from the last one you ran.
Act as a marker room manager who has to state a nesting problem completely before it is run. Facts: style [STYLE], order [QTY], sizes and the quantity of each [LIST], fabrics in the garment and the price a metre of each [LIST EVERY FABRIC], nominal and cuttable width of each fabric [WIDTHS], and the cut list showing every piece, its fabric and how many are cut [PASTE IT]. The pattern file reports these pieces [PASTE THE PIECE LIST WITH EACH PIECE'S FABRIC, CUT QUANTITY, MIRROR FLAG AND ALLOWED ROTATION]. The fabric is [SOLID, NAPPED, PRINTED, STRIPED OR CHECKED], with a repeat of [SIZE] if it has one, and the buyer's fabric manual says [PASTE ANY CLAUSE ABOUT MATCHING, DIRECTION, SPLICING OR TILT]. Do the following. First, run a piece census: count the pieces in the file by fabric and reconcile against my cut list, and name every piece whose fabric, cut quantity, mirror flag or rotation is blank, since a blank field is an instruction and not a question. Second, list the constraints that must be attributes on a piece and the ones that will otherwise be taken from the session profile, and tell me which of those profile settings I should read out loud before the run because they were set by the last style. Third, state what this fabric forbids and turn each of those into the specific attribute that expresses it. Fourth, list what this nester cannot be told at all, and say for each one who owns the risk and where it will surface. Fifth, tell me whether the objective of minimising marker length is the same as minimising money on this garment, and if it is not, say exactly where the two diverge. Sixth, give me the three checks to run on the finished marker before the cutting order is released. Ask me for anything missing rather than assuming it.
AI can make mistakes — check anything you act on.
It minimises length, not money
The last thing to hold on to is that the software is chasing one number, and that number is marker length. Length is the same thing as money only when every metre in the marker costs the same.
On CH-482 it did not. A metre of twill is USD 3.95 and a metre of pocketing is USD 1.10. The software had no way of knowing that one of the strips it was filling was three and a half times dearer than the other. It made the total length of both as short as it could, and produced the more expensive of the two answers, correctly.
Check yourselfYour nester returns a marker at 86.1% against last season's 84.9% on the same style and the same width. What do you check before you congratulate anybody?Show the answer
Three things, in this order. The piece count by fabric, against the cut list — an extra piece raises efficiency and costs money, and a missing piece raises it and stops the line. The buffer setting, against last season's, because a smaller halo raises the number without changing the marker. And the allowed rotations and tilt on every piece, because a profile inherited from another style may have granted freedoms this cloth does not have. Only when all three match is the extra 1.2 points a real gain, and then it is worth exactly what course 4.5's arithmetic says it is worth.