Vectors, Logos and Things That Get Printed

Sundeep runs a signage shop with three staff and a small CNC router. On Wednesday afternoon a jewellery showroom books him for an opening on Sunday: a six-metre by three-metre flex hoarding for the approach road, and 1.2 metre acrylic letters for the fascia above the shutter.

He asks for the logo. What arrives on WhatsApp is a JPEG, 480 pixels wide, forwarded twice. The owner says it was made in 2019 by his nephew, who has since moved to Dubai. There is no other file, nobody knows what typeface the name is set in, and the owner is confident he has supplied a logo, because he can see it on his phone.

Two jobs, two different problems.

The flex hoarding will accept a raster file, because the large-format RIP will take whatever it is given. At six metres wide, each of those 480 pixels becomes 12.5 mm across. Even at the generous 30 ppi a hoarding seen from thirty metres can get away with, the file would need to be about 7,000 pixels wide. It is fifteen times short. An upscaler will invent detail, and the detail it invents will be a guess about somebody else’s curves.

The acrylic letters are worse, because the router does not read pixels at all. It follows paths. Somewhere between the JPEG and the machine, somebody has to produce geometry.

Sundeep has two routes and he has to choose on Wednesday evening.

Trace it. Twenty minutes in PowerTRACE. The output is roughly 2,400 nodes of contour found along a staircase of pixels, with the corners rounded where the tracer decided a corner was noise. On the hoarding the wobble on each letter is about a centimetre. On the routed letters it is worse, because the bit follows every wobble at a constant feed rate and the edge comes out rippled where it should be straight. The traced file also becomes the showroom’s logo for the rest of its life, because it will be the only vector anyone has.

Redraw it. Identify the typeface, rebuild the diamond mark from two circles and a rotated square, and draw the wordmark over a locked, dimmed template. Sundeep has done this before and it takes him between two and three hours. Those hours are not billed: the client believes he has supplied artwork, and Sundeep has not raised the question of charging for it.

The cost of redrawing is not really the three hours. It is that the flex file then misses Thursday night. The printer he uses has one machine, Friday is already full with municipal election work, and the next clear slot is Saturday morning. If anything goes wrong on Saturday there is no float before the Sunday opening.

The cost of tracing is nothing on Wednesday and something every year afterwards.

Sundeep’s brother-in-law, who does the installations, has a view: nobody stands close enough to a hoarding to see a wobble, the showroom is paying for a sign and not for archaeology, and Sunday is Sunday. That argument is half right, which is what makes it hard. At thirty metres the wobble genuinely does not read. At the fascia, where customers wait for their car with their nose a metre from the acrylic, it does.

There is a third consideration that neither of them raises on Wednesday. Whoever produces the vector will be the only person in Nashik who has it, and the showroom will need it again for a visiting card, a bill book, a shutter sticker and a box for every purchase.

A filter-coffee brand ordered a die-cut star label, 70 mm across the points, for the lid of a 500 g tin. Eight points, each one about 22 degrees at the tip, because the eight points stand for the eight blends the company sells. The brand manager chose the sharp points deliberately and said so twice in the approval email.

The designer built the artwork in Illustrator. Around the star she wanted a white keyline 5 mm wide, so she selected the star, applied Offset Path at 5 mm, and used the resulting outline as the die-line. It previewed correctly on her screen. She exported the PDF, put the cut path on a spot swatch named Dieline, and sent it.

The die-maker telephoned the next morning, before cutting anything.

The die-line he had received was not a star 80 mm across. It was a star 122 mm across, with each point drawn out into a long thin needle. Nothing was corrupt. The offset had done exactly what an offset means: the two edges leaving the tip run parallel to their originals, and parallel lines from a 22 degree corner do not meet 5 mm out. They meet at the offset divided by the sine of half the angle, which at 22 degrees is 5 divided by 0.191, or 26.2 mm from the original point.

So the label would not fit the tin, would not fit two-up on the sheet the press runs, and would have needed a die twice the size of the one quoted.

The die-maker had seen it before and could have simply rounded the corners himself. He rang instead, which is why this case is worth reading, and put the decision back where it belonged.

The designer now had two hours before the press slot on Tuesday, and two options.

Round the joins. Five minutes. Illustrator’s Offset Path has a Joins setting; changing it from Mitre to Round produces a border that arcs across each tip at a 5 mm radius. The result is safe, the die fits the quoted size, and every one of the eight points the brand manager specifically approved becomes a 5 mm blunt curve. She would have to send that change for approval, and the brand manager is in Chennai until Thursday.

Build the border as geometry. Forty minutes. Draw a second star, constructed the same way as the first but larger, with the same 22 degree points, so the border sits 5 mm from the artwork along the flats and narrows gradually toward each tip. The points stay sharp. The border is no longer a uniform 5 mm, which is a real difference: at the tip it tapers to nothing, and if anybody measures it, it is not what the specification says.

Both are honest. One changes the shape the client chose; the other changes the specification the client was given. Steel rule dies are cut by hand and a wrong one costs money and a week, so guessing was not among the options.

One more thing surfaced in the phone call, and it is the part worth carrying away. The designer’s own preview had shown blunt points, not needles. Her renderer was applying the PDF default miter limit of 4, which gives up and bevels at around 29 degrees, so at 22 degrees it had quietly bevelled the spikes on screen. The die-maker’s software was not applying the limit and drew the true mitre. Two programs, one file, both following a specification, and a 26 mm difference between them.

Nandini works freelance and sets forms for three hospitals and a co-operative bank. She does not read Marathi.

The district hospital asked her to reset its surgical consent form, two pages, Marathi on one side and English on the other. The administrative officer sent the text as a Word file. The Marathi in it had originally been typed in Kruti Dev by a clerk in 2011 and converted to Unicode by somebody at some point since, using a tool nobody remembered the name of.

She set it in Mukta, which is an Ek Type family designed for Devanagari first, and built the layout in Scribus. On her own machine it rendered correctly. The shop that prints for the hospital runs CorelDRAW, on a version whose conjunct handling is known in the trade to be uneven, so she supplied a PDF rather than a working file.

The shop sent back a laser proof on Friday afternoon. Two things on it were wrong. One three-consonant conjunct had come apart into separate letters with a visible virama between them. One vowel sign was sitting on the wrong side of its consonant. Neither is a typo in the ordinary sense; a non-reader looking at the page would see nothing unusual at all, and Nandini could only see them because the shop’s proofreader had circled them in pencil.

The shop now wanted curves. Convert the text to outlines, send it back tonight, and they would make plates at seven on Saturday morning, print over the weekend and deliver on Monday. The Health Department inspection was on Tuesday and the administrative officer had already told the district office the forms would be in place.

The hospital’s matron is the person authorised to sign off the Marathi text. She reads it, she uses the form every day, and she was at a family function in Pandharpur until Monday morning. The ward clerk, who also reads Marathi, offered to look at it, but the officer was uneasy about a clerk signing for a consent document.

So the decision was this.

Outline tonight, print six thousand, and accept that if anything in the Marathi is wrong, it is wrong on a document a patient signs before surgery, and it is wrong permanently, because outlined text contains no characters for anybody to proofread. Six thousand forms is a reprint cost, an inspection finding, and a document in active clinical use that says something nobody intended.

Or hold. Plates move to Tuesday, delivery to Wednesday, and the inspection happens with the old forms on the desk. The officer has given an assurance he would then have to withdraw.

Both costs are real, and the second one is the kind that gets a freelancer replaced.

The thing worth noticing is that the shop’s request and the actual risk are not the same thing. The shop is afraid of one failure: that they open a file, the font is missing, and the layout reflows. That fear is entirely reasonable and it has a second answer.

The co-operative packs table grapes for two markets. Domestic cartons are printed at a five-unit press in Sangli, forty thousand a season. Export cartons need a different board and a longer run, and go to a larger shop in Pune, about ninety thousand.

Their brand green was chosen in 2016 on somebody’s laptop. It is a vivid emerald, a little towards blue, and it is on every carton, every crate label and the co-operative’s board outside the collection centre.

The problem surfaced when the Pune shop’s prepress operator soft-proofed the artwork against the press profile and rang to say the green would not hold. The operator did the useful thing and sent a printed sample rather than a screenshot: the nearest CMYK build on the export board, beside the current Sangli carton. Under the office tube light they looked similar. Under the D50 lamp the Pune operator had put on the desk, they did not. The export sample was flatter and about four Delta E away, which is a difference anybody sees when the two are side by side.

That is the worse case for a fruit exporter, because retail buyers do see them side by side. The co-operative’s cartons stand on a pallet next to each other at the mandi and, in season, on the same shelf in two different cities.

The managing committee had three options and money on all three.

Specify the green as a spot ink. A mixed ink is a recipe, so the same green comes off both presses, and it reaches a colour four process inks cannot. It costs a fifth plate, a fifth unit and a wash-up on every run at both shops. On the domestic carton, which is a four-colour job, that is a real per-thousand addition, and during the season the Sangli shop’s five-unit press is the one everybody wants, so a five-colour job is also a scheduling risk.

Convert to the nearest CMYK and accept it. No extra plate anywhere. The green goes flat, and, more importantly, the two shops will not match each other, because a CMYK build is a simulation that depends on the press, the board and the day. The co-operative would be buying a colour that is different in two cities and calling it their brand colour.

Change the brand green to one sitting near the cyan primary, which reproduces almost untouched everywhere. Cheapest to print, forever. It means a colour change on the board, on the crate labels, on nine thousand printed cartons already in the shed, and a conversation with a committee of growers who chose the green in 2016 and like it.

The secretary’s instinct was that spot inks are for large companies. The Pune operator’s instinct was that the export carton is the one seen beside a competitor’s on a shelf in Rotterdam and the domestic carton is not.

What nobody had was a number anybody could act on. The brand sheet, such as it was, held a hex value and a CMYK build typed by a designer in 2016, and neither means the same thing in two factories.

The trust running the school approved a mark in February. It is a peacock’s eye feather: fourteen fine barbs radiating from an eye, with the school’s name set below in a light sans. At 80 mm on the prospectus cover it is genuinely good, and it won partly because the chairman’s daughter draws peacocks.

The designer had shown three concepts and would have been content to ship any of them, which is the rule he works to. This was the one chosen.

In May the uniform order went out to a supplier in Rajkot: eighteen hundred shirts, chest pocket embroidery, to be placed within eleven days so the garments arrive before the school year starts.

The supplier’s digitiser looked at the artwork and made three observations, none of them a matter of opinion.

At the pocket size of 45 mm, the barbs are 0.7 mm wide. A satin stitch works between about 1.2 mm and 12 mm; below 1.2 mm the needle simply perforates the fabric into a line, and on a knit polo the fabric pulls inward as the thread is drawn, so the drawn width is not the finished width. The school’s name sits at 3.8 mm cap height in the lockup, and embroidered text below about 5 mm turns into a smudge. And the mark uses five colours, each of which is a machine stop on every single garment.

He could digitise it. He said so. He also said what would come back.

The designer had three options and eleven days.

Simplify the primary mark. Remove the barbs, thicken the strokes, drop the name to a heavier weight. The result works everywhere, including on the shirt. It is also visibly not the mark the trust approved in February, and the chairman will be looking at a different peacock.

Draw a second mark for embroidery and small use. Keep the approved mark for the prospectus, the board and the website, and draw a separate single-thread version at 45 mm: fewer barbs, thicker strokes, no tagline. Two days of drawing, a fee the trust has not budgeted, and two marks in the world, which the stationery shop near the gate will certainly confuse.

Run it as drawn. The supplier will do it if instructed. Eighteen hundred shirts, and the smudge arrives on every child’s pocket for the year.

The trust’s position, stated by the honorary secretary on the phone, was that the logo had been approved, paid for and finished, and that a designer asking for more money in May for something approved in February looked like a designer who had not done his job in February.

That is an uncomfortable position to argue against with words, because it is half true. The surfaces list should have been collected before anything was drawn, and embroidered pockets on eighteen hundred shirts is not an obscure requirement for a school. Had the question been asked in January, the peacock would have been drawn with four barbs from the start and nobody would have experienced it as a loss.

There was also a quieter argument against the second mark, and it came from the school’s office rather than from the trust. The stationery shop outside the gate prints the diaries, the receipt books and the identity cards, and whoever is in a hurry there opens the folder and takes the first file. A school with two peacocks in one folder will get the wrong peacock on something within a term. Large organisations do ship graded sets of marks, drawn separately for each size, but they also have somebody whose job it is to police them, and a six-hundred-pupil school does not.

The designer’s own view, which he kept to himself during the call, was that he would rather have one mark that worked everywhere than two that were correct and confusable. What stopped him proposing that outright was that the chairman had chosen the fourteen-barb version in a room full of people.

The polytechnic redesigned its convocation certificate because the principal wanted each student’s name printed in Marathi as well as in English. Fourteen hundred and six students, one certificate each, personalised.

The designer built the template properly. One page, at trim size, with the student name, the Marathi name, the programme, the roll number and the year as merge fields. The registrar’s office exported the data from the student information system to a spreadsheet, and the clerk sent it across on Thursday evening with an apology for the delay.

She ran the merge at eleven that night, which was already later than she wanted, and previewed before creating the merged document.

Record one was fine. Record two was fine. From record three onwards, every Marathi name was a row of question marks.

The cause took four minutes to find and could not be fixed in the layout. The student system had exported the file in a legacy single-byte encoding rather than UTF-8, so every Devanagari character had been replaced on the way out. The English names were intact because they are ASCII. There was nothing in the layout file to repair; the data had lost the characters before it arrived.

While she was in there she found a second fault. Three names overflowed the frame, one of them badly, because she had built the template around a record she had picked at random from the top of the file rather than around the longest one.

The digital press was booked for ten the next morning. The convocation was on Saturday, the Governor was attending, and the certificates are handed over on stage.

Three options, at eleven at night.

Drop the Marathi line. Print English only. This can be done immediately and the registrar could authorise it by phone. It also removes the one thing the principal asked for, on the day it was meant to appear, in front of the Governor.

Get the data re-exported as UTF-8. Somebody with access to the student system has to do it. The registrar’s office opens at ten on Friday, which is when the press slot starts. Asking the press to slip a day puts the printing into Saturday morning, which is the convocation.

Retype fourteen hundred Devanagari names from the original register overnight. Nobody seriously proposed this, and it is worth noticing why: it would introduce fourteen hundred fresh opportunities for a name to be wrong on a document a person keeps for life.

The press slot mattered more than it looks. The polytechnic had booked a four-hour window on a digital press that was running an election job either side of it, and the shop had made it clear that a slot given up was a slot gone. Digital printing has no plates, so a short notice change is cheap in materials and expensive in scheduling, and the scheduling was the whole constraint.

Dropping the Marathi line was also not as simple as it sounded. The certificate had been redesigned around two names, and the English name alone would sit in the middle of a space designed for two lines, which anybody on stage would read as a mistake rather than as a decision. Fixing the layout for a single line was another half hour she did not have, on top of a change the principal had not agreed to.

The clerk who had sent the file had gone home at half past five. The designer had his mobile number, because he had rung her earlier in the week about the column headings, and she had saved it.

The job was a folding carton for a single-estate tea, three thousand pieces, printed four colour on 260 gsm uncoated board with a die-cut window and a spot varnish. The trade fair the cartons were for opened on Thursday morning. The printer’s platemaker starts his run at eight in the evening to hold that delivery.

The designer exported the press PDF at ten past six and ran the five-line preflight script he keeps on his machine. It takes about thirty seconds. He had been sending files to this shop for two years and had never once had a file rejected, which is the state of mind that makes people skip the check.

Four things came back.

The boxes were right: pdfinfo reported a TrimBox at the agreed carton size with the BleedBox 3 mm outside it on every edge.

The fonts were all embedded, every row showing yes, though one carried an unfamiliar name with an axis value appended, which is what an instanced variable font looks like in a listing.

Ink coverage was not right. Ghostscript’s inkcov reported a page total of 341 per cent. The printer’s stated limit on that uncoated board is 300. The excess came from a dark photograph of the estate at dawn, which had been converted with a coated profile because that was the last setting the document had been given.

And the separations were wrong in a way that would have been invisible on screen. Running tiffsep produced five files where four were expected: cyan, magenta, yellow, black and a spot named Dieline, which is correct, because the die-line should travel as its own separation. But the die-line objects were set to knock out rather than to overprint, which means the cut path would punch a white outline straight through the artwork it sits on. In a normal render of the PDF it looks like a thin line on top of the picture. On the sheet it is a hole.

It was now twenty-five past six.

The die-line fault is certain: it will print, on every one of the three thousand cartons. The ink coverage is a judgement: 341 per cent on uncoated board may dry given time, or may set off onto the sheet above it in the stack, and the job is going straight into a finishing line.

The options were the obvious three.

Send it and add a note asking the RIP operator to force the Dieline separation to overprint. Many RIPs can. It puts the fix in somebody else’s hands, at eight at night, on a job he has not seen.

Hold the plates and fix both faults properly: convert the photograph with the uncoated profile and set the die-line objects to overprint, re-export, re-preflight. Perhaps forty minutes of work, but the platemaker would then be working past ten, which is overtime the shop charges for, and the two-hour slip eats the float before Thursday.

Fix only the die-line and let the coverage ride, on the grounds that one is certain and the other is a risk.

A hotel group opening a property in Amritsar ordered uniforms: eighteen hundred garments, split between woven cotton shirts for the front office and knit polos for housekeeping and the restaurant. Both carry the same chest logo.

The mark is a stylised lotus, drawn well, with a tagline beneath it. The group’s marketing manager approved it on a screen and on a printed presentation, where it is 90 mm wide and looks excellent.

At the chest size the supplier works to, 55 mm, the lotus stem measures 0.9 mm and the tagline sits at 4 mm cap height. The mark carries five colours, one of which is a pale gold.

The digitiser quoted 11,400 stitches and five colour stops. Embroidery is priced per thousand stitches, and every colour change is a machine stop on every single garment, so both numbers are in the price on all eighteen hundred pieces.

The unit’s manager asked for a sew-out before the run: one stitched sample on the actual fabric, seven hundred rupees, and a day and a half, because his machines were booked and he would have to slot it between orders.

The hotel’s project manager said no. The artwork had been approved. The property handover date was fixed. There were three days of float in the schedule and he did not want to spend half of them on a sample of something already agreed.

The disagreement is a real one and both men have a reasonable case.

Against the sew-out: a day and a half out of three days of float, a small cost, and a conversation with a client who believes the decision was already made in a meeting they attended.

For the sew-out: satin stitch works down to about 1.2 mm and the stem is 0.9. Embroidered text below about 5 mm cap height smudges and the tagline is at 4. The knit polo will pull far more than the woven shirt, so the same digitised file will come out narrower on one than the other. And the failure mode is not a slightly disappointing garment: it is eighteen hundred garments, the thread, the machine time, and a property opening with no uniforms.

The digitiser can compensate for some of this. Pull compensation means widening shapes by two or three tenths of a millimetre on each side so the finished stitch lands where the drawing says. But compensation is per fabric, and nobody in the room knew how much this knit pulled, because nobody had stitched on it.

The split order made that worse rather than better. Nine hundred woven shirts and nine hundred knit polos is not one job with two fabrics; it is two digitising jobs, because the file that comes out right on poplin comes out narrow on a knit. Somebody was going to have to decide that, and the only way to decide it with a number rather than a guess was to stitch one of each.

The economics are worth stating plainly, because they are what made the argument possible at all. Seven hundred rupees is roughly the cost of four garments. Eleven thousand four hundred stitches at five colour stops, on eighteen hundred pieces, is the largest single line in the embroidery quotation, and every one of those numbers was set by decisions in the artwork rather than by anything the unit could control. A conversation about the artwork was therefore also a conversation about the price, and neither man had framed it that way.

So the decision at that table was not really about seven hundred rupees. It was about whether the unit would run eighteen hundred garments on an assumption nobody had tested, on fabric nobody had stitched, to hold three days of float.