Stone Guides

On a memorial, the letters are the finished product

On a memorial the stone is the carrier and the wording is the finish. What has to be true of a line of text before anyone is allowed to cut it.

15 November 2026  ·  22 min read  ·  GranTerra Stone, Hui'an

Light grey granite panel in the workshop showing two worked surface treatments, raised marks and cut grooves, side by side on one slab

On most stone orders the drawing is the authority. It carries the shape, the edges, the finish, the positions of the holes, and the count. Every argument that follows can be referred back to it. On a memorial order the drawing does all of that too, and then a second document quietly takes over: a few lines of text, usually arriving as an email attachment, sometimes as a photograph of handwriting, occasionally typed straight into the body of a message. That text is what the finished piece will be judged against. The stone is the carrier.

The reversal is easy to miss because the rest of the order looks familiar. There is a block to choose, a surface to work, a crate to build, a container to book. A buyer who has handled a hundred cladding orders will handle this one the same way, and will spend the enquiry on colour, finish and lead time. Those things still matter. But the complaint that comes back, if one comes back, is almost never about the colour. It is that a name has lost a letter, or a date has been written in the wrong order, or a mark above a character was left off because somebody thought it was decorative.

What follows is about that second document. Not about how stone is carved, which is a workshop subject, but about what has to be true of a line of text before anyone is allowed to cut it. Most of the cost in this category is decided at a desk, by people who never see the slab.

The text is the finished surface

Consider what an inspection actually checks on an ordinary order. A panel is measured against a drawing, looked at in the right light, compared with an approved sample, counted, and packed. The questions are all about material: is it the right stone, is the face right, is it square, is the edge what was agreed. A piece can fail any one of those tests and still be a panel. What nobody does is read it.

On a memorial, reading it is the test. A piece can be cut to the right shape, from the right block, with a perfect polish and a clean edge, and still be wrong, because the surface people actually look at is the lettering. The polish is the background. The words are the finish. This is the only category we handle where the thing that is inspected is a sequence of characters rather than an area of stone.

That changes where the risk sits. On an order that starts from a drawing, the risk is spread across dimensions, tolerances and quantities, and a mistake can often be absorbed by a joint or a trim. Here there is nothing to absorb it. A character is either right or it is not, and there is no tolerance on a surname. Every other property of the piece is tolerant of some variation; the wording is not.

It also changes what the order date means. In practice a memorial order does not start when the drawing is agreed. It starts when the text is frozen. Until then, everything else is provisional, because the layout of the face depends on how long the lines are, and the length of the lines depends on characters that have not yet been confirmed. We have seen buyers hold a whole run while a single mark above a single letter was being checked. They were right to. It is cheaper than recutting.

So the first thing to establish, before colour or finish or delivery, is who owns the text and in what form it exists. If the answer is “it is in an email somewhere”, the order is not ready. A line of wording is a specification with an unusually small margin, and it needs to be treated with the same formality as a dimensioned drawing, then with a good deal more suspicion.

The wording arrives as a transcript

A design brief can be interpreted. That is what a brief is for: it states an intention and leaves room for somebody skilled to realise it. A transcript cannot be interpreted at all. It has one correct version and every other version is an error, including the ones that read better. This distinction is the source of more trouble in this category than any cutting problem, because the people handling the text are used to briefs.

Look at how it actually arrives. A photograph of a handwritten card, taken at an angle, with a thumb over one corner. A paragraph copied out of a PDF where the line breaks are part of the layout and not part of the wording. A typeset funeral sheet with a decorative capital, a rule underneath, and a date in a small ornamental face. A note in the body of an email that has passed through a phone’s keyboard. A spreadsheet column, where a leading zero has already been removed. Each of these is a rendering of the text, not the text itself, and each rendering has made decisions on the buyer’s behalf.

Those decisions are silent. Optical character recognition will turn a handwritten character into the nearest common letter. A layout program will substitute a missing glyph without saying so. A mail client will normalise a long dash to a short one, or turn two hyphens into one. A spreadsheet will decide that a number is a number. None of this is visible in the final file, which looks clean and ordinary, and that is precisely the danger: a clean file invites trust, while a messy handwritten card invites a question.

What we need is a plain transcript, and it is worth being specific about what that means. Characters typed out as characters, one line per line as it is to appear, with no automatic capitalisation, no smart punctuation, no field formatting, and no styling doing any work. Marks above or below letters written as marks, not approximated by the nearest unmarked letter. Spaces where spaces belong, including the ones that look accidental. If a character cannot be typed, it should be described in words rather than replaced.

This is also the point where what a handover quietly drops becomes expensive. A drawing that passes between six people loses its assumptions; a line of text that passes between six people loses its least conspicuous characters. The two failures look similar on paper and behave quite differently in the workshop. A missing assumption produces a question. A missing character produces a finished stone with a name misspelled on it, and nobody notices until it is installed.

Spellcheck is a silent editor

Of all the ways a text can change, the ones done by software are the hardest to catch, because they improve it. A surname that is uncommon gets offered a common replacement and quietly accepted. A word in a language the checker does not know gets flagged and re-spelled into a word it does know. Two words written apart get joined. A capital in the middle of a family name gets moved, because the software has a rule about capitals and the family has a history. Every one of these edits makes the line look more normal.

That is why ordinary proofreading does not find them. A proofreader reads for sense. A corrected spelling still makes sense; in fact it usually makes better sense than the original, and the eye rewards that by going straight past it. The classic failure is not a typo that survives. It is a typo that was fixed.

Dates deserve their own paragraph, because they are the most frequently altered thing in the whole document. A date written day-month-year in one country is read month-day-year in another, and both readings are perfectly reasonable, and both produce a plausible date. Reordering the parts does not produce nonsense; it produces a different day. There is nothing on the finished stone to indicate which convention was intended. The only defence is to write the month in words at the approval stage, or to state the convention explicitly, and then to keep that statement attached to the job.

The practical answer is to stop treating the text as prose and start treating it as data. Freeze one version, give it a reference, and compare the next version against it character by character rather than by reading it. Do not accept a version that is “the same but tidier”. Do not let a file be re-typed when it could be copied. And be suspicious of any change that makes the line look more conventional than it did when it arrived from the family. In this category, an improvement is still a change, and a change is a defect until somebody with authority says otherwise.

There is a small habit that helps more than any checklist: read the approved line backwards, one character at a time, aloud. It breaks the brain’s habit of completing words from their shape, which is exactly the habit that lets a silent correction through. It takes a minute. It is cheaper than the alternative, and the alternative is not a rework.

Diacritics are not decoration

A mark above a letter, a ring over a vowel, a cedilla, a stroke through a character, a pair of dots, a ligature that has become a single shape over time: on a printed page these can look like typographic ornament, and in an industrial component drawing they would be irrelevant. On a memorial they are part of the name. Removing one does not simplify the spelling; it changes it, in the same way that removing a letter changes it.

The trouble is that these are exactly the characters that get lost in transit. A font that lacks the glyph will substitute something close. A system that cannot encode it will drop it. A keyboard that has no easy way to type it will encourage the person copying the text to leave it out, often with the reasonable thought that the reader will understand. And a buyer reviewing a proof on a small screen may genuinely not see whether a mark is present, because at that size it is two or three pixels.

Scripts that do not run left to right add a different set of problems. The direction of the line, the way characters join, and the position of marks relative to the baseline all have to be stated rather than assumed. A layout program may render a line correctly on screen while storing it in an order that a cutting file will read backwards. None of this is exotic; it is ordinary text handling, and it becomes critical only because there is no undo once the surface has been cut.

What we ask for is not a technical file. It is a plain statement alongside the transcript: which characters are intentional, whether any of them carry marks, which direction the line runs, and which forms must not be replaced by a near equivalent. If a character is present in the source and absent from the proof, that is a stop, not a query to be resolved later. And if the final rendering depends on a specific digital font being available at our end, that is worth knowing early, because the workshop does not cut a font. It cuts a shape that someone has to describe.

The font name is where the trouble starts

“Use this typeface” is a perfectly good instruction for a printer and a very incomplete one for a stone workshop. A font name identifies a set of outlines that live in a file, on a machine, in a particular version. The same name can produce different letter widths in different releases. A weight described as bold on screen has no direct equivalent in a groove cut into granite. And the shapes that look elegant at a large size on white paper can close up completely when the same outlines are cut shallowly into a dark, veined surface.

What actually determines how lettering reads on stone is a handful of physical things. The width of the stroke, the profile of the cut, whether the ends of strokes are finished square or tapered, how a corner resolves, how deep the recess sits relative to the surrounding polish, and above all whether those things stay consistent across every character in the line. A letter whose stem is slightly wider than its neighbours will announce itself under raking light even when nobody can say why the line looks wrong. Uniformity is the whole craft, and it is invisible in a PDF.

This is why a screen proof is not enough. A proof shows the arrangement: where the lines sit, how much space they occupy, how the block of text balances on the face. It does not show how the strokes will hold shadow, or how a thin serif will survive on a coarse-grained material, or what a shallow cut will look like when the stone is wet. Those are properties of the surface and the light, not of the layout, and how the surface is worked changes every one of them.

The cheap answer is a small physical sample with a few characters cut into the actual material, in the actual finish, looked at in the sort of light the piece will eventually sit in. It is the same logic as the panel that gets approved before a run starts, applied to a groove rather than a face. It settles questions that no amount of correspondence can settle, and it settles them while changing the answer still costs almost nothing.

A related point: “make it bolder” is not one instruction but two. Widening a stroke and deepening a cut both make lettering more visible, and they do not look the same. One throws more shadow, the other changes the width of every character and therefore the length of the whole line. If the line has already been balanced on the face, that single word can move it. It is worth asking which of the two is meant before anybody touches the stone.

Blank space is a reservation for a later name

Here is a question that has no equivalent anywhere else in the catalogue: will somebody come back, some years from now, and ask for another line to be added? On a memorial the answer is often yes. A second name, a second date, a short addition of the same kind, cut into the same face, by a workshop that may no longer hold the original files. It is a second order that is already implicit in the first one.

Which means the empty area on the face is not unused space. It is a reservation. A layout that fills the whole stone beautifully today has spent something the buyer did not know they were spending. When the later request arrives, the options are to squeeze a line into a gap that was never sized for it, to reduce the letter size so the new line does not match, or to replace the piece. All three are visible for as long as the stone stands.

So the first layout conversation should include a question about the future: is another line expected, roughly how much text, and on which side of the existing block does it belong. None of this requires a commitment or a date. It only requires that the blank be deliberate. A deliberate gap can be left in the right place, with the spacing and the letter size chosen so that the later work can match. An accidental gap is just whatever was left over after the first text was centred.

The second part is record-keeping, and it is the part buyers rarely think to ask for. Whoever does the later addition needs to know the exact typeface treatment, the stroke width, the depth, the finish of the recess, and whether any filling was used, along with the original approved wording. Without that, matching is guesswork. With it, the second line can be cut years later and still look as though it was always there. Ask for those details to be kept with the job, not just the drawing.

This is also where the point where a stone order stops being editable has an unusual twist. Normally the freeze is the end of the story. Here the first freeze has to leave room for a second one, which means deciding, at the outset, what is intentionally unfinished. Very few orders ask you to plan for work that has not been commissioned yet.

Where the stone fights the lettering

Granite is not a blank sheet. It carries colour lines, mineral concentrations, veins and patches that move across a slab without regard for where a line of text needs to sit. A fine stroke cut across a dark band can disappear from one end of a word to the other. A pale cluster behind a name can make a letter look chipped when it is not. The material is doing nothing wrong; it is simply not a uniform background, and lettering is the least forgiving thing you can put on a non-uniform background.

This is a selection problem before it is a cutting problem. How a slab is read before anything is cut normally concerns where a vein will fall across a visible face, and which end of a run should take the most prominent piece. On a memorial, the critical area is much smaller and much more specific: the rectangle where the letters will go. A slab can be handsome overall and unusable for this purpose, because the one quiet patch the wording needs is not where the layout wants it.

Filling the recess is the usual way to gain contrast, and it is worth being clear about what a fill is. It is a surface treatment sitting in a groove, not part of the stone. It is affected by cleaning, by weather, by the texture of the cut it sits in, and by time. A letter that reads well because of a fill will read differently when that fill has aged, and the difference will not be even across the line. None of that makes filling a mistake. It is a reason to choose the base colour and the cut depth so the lettering keeps some presence even without the fill, and to expect the line to change character slowly over the years rather than fail all at once.

Two further checks are worth making on the actual material rather than on a photograph. Look at it wet, because the same face does not return light the same way in every condition, and a piece that sits outdoors will be seen wet for a good part of its life. And look at it in low, angled light, because that is when a shallow cut either holds a readable shadow or vanishes. What counts as a natural feature rather than a flaw is a separate question; this one is only about whether the characters stay legible against whatever the stone happens to be doing behind them. If the answer depends on one granite holding one colour, the answer is no.

Five pairs of eyes, five quiet improvements

Between the family and the chisel there are usually several people: the person who writes the wording down, the retailer or cemetery office that collects it, the importer who places the order, the sales desk that turns it into a job, the person who sets out the face, and the person who cuts. Each of them sees a line of text that looks slightly wrong in some small way, and each of them is trying to be helpful.

One adds a full stop. One changes a capital because it looks like a mistake. One joins two words that have always been written apart. One converts a date into the local convention. One assumes a mark above a letter is a printing artefact. None of them thinks they have changed anything substantive, and no record of any of these decisions survives, because each was made in passing while doing something else. The finished wording is a composite of five small acts of tidying, and no single person has seen the whole chain.

This is different from a simple transcription error. An error can be found by comparing against a source. An improvement has no source to compare against, because the person who made it believed they were restoring the source. That is why a version number matters more here than a signature: it says which rendering is authoritative, and it makes “I tidied it up” visible as the change it actually is.

The confirmation that matters, then, is not a general approval of the design. It is a specific, character-level agreement that this exact sequence is what will be cut, and that no further changes will be made on the way to the machine. What a signature actually certifies is worth being precise about at this point: approval of a layout is not approval of the spelling, and a signature on a delivery note has nothing to say about either. A single named person has to confirm the characters, and that confirmation has to travel with the job rather than sit in a mailbox.

It is worth adding one unglamorous control. Ask whoever sets out the face to send back the exact string they intend to cut, copied from their own file, not retyped, and have it compared against the approved version by someone who did not prepare either one. That single exchange catches almost everything in this section. It costs an email.

A letter taken out of stone stays out

The reason a wrong character is so expensive is not that the workshop is busy or that the block has been allocated. It is that cutting removes material. Grinding a letter away lowers that part of the face below the surrounding plane, and the depression remains visible under raking light however well the surface is repolished. Filling it leaves a patch with a different colour and a different way of taking a polish. Recutting over the same area makes the stroke wider than its neighbours, which is visible for a different reason again. All three fixes are detectable on the finished piece, at exactly the place everyone looks first.

So “we can correct it later” is not true in this category, and it is worth saying plainly. A wrong character is not a rework; it is a replacement. The original piece may still be perfectly good stone, with a perfect edge and a perfect finish, and it is still scrap for this purpose. The limits of putting a surface back are usually discussed in terms of chips and cracks; lettering reaches the same limit by a shorter route, because the damage is deliberate, clean, and exactly where it cannot be hidden.

There is a useful way to think about the cost curve. Before the text is frozen, a correction is a few keystrokes. After it is frozen but before the face is set out, it is a revised layout. After the layout is approved but before cutting, it is an awkward conversation and a delay. After the first stroke, it is a new piece of stone, a new place in the schedule, and a new crate. Nothing about the mistake gets bigger; only the number of things that have already been built on top of it.

This is also why a physical sample is worth doing early. It is the last moment at which the wording, the stroke width, the finish and the actual material are all in front of one person at the same time. After that, each of them lives in a different file. If something is going to be wrong, that is the cheapest possible place to find out.

The approval that cannot be delegated

Here is the division of labour, stated as plainly as we can. We will cut the exact sequence of characters you confirm, keep it identical across however many pieces the order contains, set the face out so the lines sit where the approved layout puts them, produce a small sample in the actual stone and finish when that is useful, photograph the cut work before it is packed, and keep the approved wording and the treatment details with the job so that a later addition can be matched. Those are workshop commitments and they are ours.

We cannot decide whether a name is spelled the way the family wants it, whether a date follows one convention or another, whether a mark above a character is intended, or whether a line should be left for text that has not been written yet. We do not know, and guessing would be worse than asking. Nor can we sign on behalf of the person who does know. Where our part of the job ends is a general question on most orders; on this one it ends at a very specific line, which is the line of text.

Three things are worth settling before anything is cut. Who supplies the final transcript, in a plain form, with the characters written out rather than rendered? What is the reference for the version that has been approved, so that every later file can be compared against it? And is a second line expected in the future, with the space and the treatment details reserved for it? None of these needs a form or a formal document. A short note to us answering them is enough to make the rest of the order routine.

Every other stone order we handle is judged by looking at an area of material. This one is judged by reading a line, and a reader does not forgive a missing character the way a joint forgives a millimetre. The wording is the only part of the job that cannot be adjusted afterwards, and the only part that arrives from someone we will never meet. Give it a version, give it an owner, and give it one careful minute before the first stroke. That is the whole discipline, and it costs less than a single replaced piece.

Written in the yard

Every figure on this page comes from our own order sheets. Where something is a trade norm rather than a number we can hold, we say so — send a spec and we will confirm it against a real block before you commit.

Next step

Send the drawing. We do the rest.

Reply within one working day, in English or Chinese. Send a spec, a drawing or just a photo of the sample you are matching — we will tell you honestly whether we can hold it.

MOQ 1 × 20FT  ·  FOB XIAMEN  ·  SAMPLES IN 7 DAYS

Chat with us on WhatsApp