We have argued elsewhere that a specification clause is a sentence that must later be proved or disproved, and that four operations in a sliding window leave nothing visible in the finished product. This article starts where those two ended. A clause becomes enforceable only when somebody turns it into a number, and that conversion happens on exactly one document: the fabrication drawing.
Ask a supplier for a sliding window fabrication drawing and, nine times out of ten, you will be sent an elevation. It is the wrong sheet. An elevation tells you what the window will look like from one side. It cannot be cut from, it cannot be glazed from, and it cannot be ordered from. It is the cover page of a package, and the package has seven sheets.
Here is the definition we will use for the rest of this article, because almost every dispute about drawings comes from two people using different ones. A fabrication drawing is not a description of a window. It is a set of instructions that three departments read independently of each other: the saw, the glass supplier, and the hardware buyer. Anything that one of those three cannot read off the package was never specified. It was merely described, and a description cannot be failed.
A published Chinese fabrication-office standard for door and window packages lists the sheets and, usefully, the order in which they are bound. That order is itself a statement of priority.
| Sheet | Who actually reads it | The question it answers | What goes wrong when it is missing |
|---|---|---|---|
| Split elevation | Sales, site foreman, installer | Which opening is this, and how is it divided | Two bedrooms with mirrored handing get one mark and are built identically. One of them is wrong. |
| Section details | Factory engineer, installer, building control | What is inside the profile, and where the water goes | Drainage becomes a closed rectangle on paper. An anchor nobody drew bridges the thermal break. |
| Fabrication and assembly drawings | The workshop | How long is each bar, at what angle, with which milled holes | The saw operator makes the decision for you, per shift, without a record. |
| Glass schedule | Glass supplier | Exact pane sizes, build-up order, and every treatment | Glass is cut to the frame size instead of the rebate size. Every pane is rejected on site. |
| Accessories schedule | Purchasing, and the person who services the window in year eight | Every gasket, brush, corner key, screw and sealant, with its grade and standard number | The spare part you need has no part number, because it was never a line item. |
| Profile optimisation list | The saw | How many stock lengths, in what cutting order, with what kerf allowance | Cut lengths drift between batches and between shifts, and nothing on the elevation explains why. |
| Profile order list | Purchasing and the coating line | Which die, which stock length, which finish, which colour, and how much spare | A colour batch is re-ordered halfway through the project and does not match the first delivery. |
The published binding order runs cover, index, profile order list, accessories schedule, glass schedule, glass fit table, optimisation list, split elevation, sections, itemised list, fabrication and assembly drawings. Note what that order says. The two documents everyone asks for first, the elevation and the render, are the ninth and the absent. The three documents that actually determine what arrives are the first three.
The reason a fabrication drawing cannot be replaced by an email is that the word size does not mean one thing. GB/T 5824-2021, the Chinese standard for opening size systems for doors and windows, sets up a three-level dimension system that is worth learning even if you never build in China, because every window industry on earth has the same three levels and most of them have never written them down.
Now the sentence that should be printed at the top of every approval sheet. Clause 5.2 of that standard says the construction dimension may be greater than, equal to, or less than the marking dimension, depending on where the locating line sits. Three legal answers to the same question. A frame that is larger than its own name is not an error; a frame that is smaller than its own name is not an error either. Which one you get is decided by a clause most buyers have never read.
Clause 5.3 then lists what you need to get from the marking dimension to the construction dimension: the installation form, the installation method, the thickness of the wall finish, and the structural gap. Published readings of that clause give the two installation methods their own gap figures. Dry installation, where a metal subframe is set into the opening first and the window is fixed to it after the wet trades finish, is given as J not less than 15 mm. Wet installation, where the window goes straight into an unfinished opening and the gap is filled during finishing, is given as 10 to 20 mm. And the finish thickness is counted inside the total, so that J equals J1 plus J2.
Read that list again and notice who owns each item. The installation form belongs to the detailer. The installation method belongs to the builder. The finish thickness belongs to the tiler or the renderer. The gap is the arithmetic result of the three. The factory owns none of them, and the factory is the one that has to print a number.
So this is what actually happens. An enquiry arrives saying 1200 by 1500. No installation method, no finish thickness, no locating line. Somebody in a drawing office makes three assumptions, writes the resulting frame size on a sheet, and sends it for approval. From that moment the assumption looks like a fact, because it has a dimension on it and a border around it. A callout such as 36x80 entry door is a marking dimension in its own system, in exactly the same way: it is the one number that will never appear on a cutting list.
A published shop-drawing guide for aluminium windows puts the practical rule plainly. There is no single installation gap suitable for every aluminium window; the allowance must suit the selected system, the wall construction, the installation method and the expected movement. It also says to record who supplied the dimensions and whether they were field verified or taken off design drawings, and not to calculate manufacturing dimensions by scaling a drawing or by applying an unconfirmed deduction to every opening.
Suppose the marking and construction questions are settled, and the drawing says 1200. What does it mean? Not 1200. GB/T 8478-2020, the aluminium door and window product standard, gives the manufacturing tolerances in its Table 6, and they are not rounding errors.
| Item | Size range (mm) | Doors | Windows |
|---|---|---|---|
| Width and height construction dimension | Up to 2000 | Plus or minus 1.5 | Plus or minus 1.5 |
| Width and height construction dimension | 2000 to 3500 | Plus or minus 2.0 | Plus or minus 2.0 |
| Width and height construction dimension | Over 3500 | Plus or minus 2.5 | Plus or minus 2.5 |
| Difference between opposite sides | Up to 2000 / 2000 to 3500 / over 3500 | Not more than 2.0 / 2.5 / 3.0 | Not more than 2.0 / 2.5 / 3.0 |
| Diagonal difference | Up to 2500 / over 2500 | 2.5 / 3.5 | 2.5 / 3.5 |
| Frame to sash lap width | Not size dependent | Plus or minus 2.0 | Plus or minus 1.0 |
| Step between frame and sash members, same section | Not size dependent | Not more than 0.3 | Not more than 0.3 |
| Step between frame and sash members, different sections | Not size dependent | Not more than 0.5 | Not more than 0.5 |
| Assembly gap between frame and sash members | Not size dependent | Not more than 0.3 | Not more than 0.3 |
A drawing that says 1200 is therefore not a contract about a window. It is a contract about a band roughly three millimetres wide, plus a separate promise about squareness and a separate promise about the step at every joint. Two windows can both be compliant and differ from each other by more than the gap a fitter tries to shim.
Now look at the row nobody dimensions. The frame to sash lap width is the tightest tolerance in the whole table for windows, plus or minus 1.0 mm, and on a sliding window the lap is where the interlock, the seal line and the anti-derailment geometry all live. It is the dimension that decides whether a panel sits quietly in its track or chatters in a wind. It is almost never dimensioned on an approval drawing, because it is not a visible dimension, and because nobody asks.
Notice also the asymmetry with doors, at plus or minus 2.0. If the product you are approving is a sliding door rather than a sliding window, the applicable column changes, and a drawing that quotes the window figure has quietly claimed a tighter promise than the standard requires.
Here is the most expensive omission in a window drawing package, and it is almost universal. The drawing specifies the glass. The drawing does not specify the groove the glass sits in.
JGJ 113-2015 does not give a groove size. It gives two formulas, in clauses 12.1.2 and 12.1.3, and a table of minimum values to feed them.
For insulating glass, the published table gives two rows. For 4 plus A plus 4 and 5 plus A plus 5 builds, the front and back clearance is 5.0 mm with sealant glazing and 3.5 mm with gasket glazing, the engagement depth is 15.0 mm, and the edge clearance is 5.0 mm. For 6 plus A plus 6 and thicker, up to 12 plus A plus 12, the clearance is 7.0 mm with sealant and 5.0 mm with gasket, the engagement depth is 17.0 mm, and the edge clearance is 7.0 mm.
Do the arithmetic for one of the most common insulating builds in the world, 6 plus 12A plus 6. Its nominal thickness is 6 plus 12 plus 6, which is 24 mm.
The same standard sizes the blocks that hold the glass off the metal. A setting block must be at least 50 mm long, as wide as the glass plus both clearances, and as thick as the edge clearance. A location block must be at least 25 mm long. Published glazing practice adds that setting blocks are placed at the quarter points of the bottom rebate.
There is a published design review note from a Chinese curtain wall engineer that is worth repeating here, because it is the cleanest illustration of what goes wrong. A detail showed a laminated insulating unit around 60 mm thick with an engagement depth of 18 mm. The author's point: 18 mm is barely above the 17 mm floor that applies to a build a fraction of that thickness, the published table does not even contain a row for that glass, and beyond the table the engagement depth is a calculation against wind uplift and self weight, not a lookup. The detail had been built on many projects and had never visibly failed. That is exactly the problem. The groove is the one part of the window that fails late, fails suddenly, and fails in a way that nobody connects back to a number on a sheet.
This is the part where we talk about our own numbers, and the honest version is slightly uncomfortable. A catalogue is a collection of system properties and visible properties. A fabrication drawing is a collection of the things in between, and the things in between are almost all of them specific to your opening.
| What we publish | What it actually measures | What the drawing still has to add |
|---|---|---|
| Frame wall width, 90 to 182 mm across the ranges | How far the frame goes into the wall | The depth of the wall you actually have, including its finish |
| Glass sash thickness, 43 to 90 mm | The stack up of the sash section from front to back | The groove width and depth, which is a different measurement taken in a different place |
| Sash width, 44.5 to 96 mm | The visible face of the sash | The lap onto the frame, which is the tolerance item at plus or minus 1.0 mm |
| Frame width, 26 to 62 mm | One visible edge of the frame | The structural section behind it, and the anchor positions inside it |
| Track heights, 53 and 31 mm, or 43 and 8 mm | Two different tracks on one frame, not a tolerance band | Which panel rides which track, and the route the water takes to leave |
| Mullion, 75/21, 75/31, 58/22, 72/4 | Depth first, then visible face | Which of the two governs at your span, and whether reinforcement is required |
| Roller rating, 200 kg, 220 kg, 400 kg | Per sash, under an assumed glass build | The glass build itself, printed next to the rating on the same sheet |
| Thermal value, 1.3, 1.6, 1.7 | A system value on a stated reference configuration | A whole window value at your size, your split and your glass |
Read down the middle column. Every single published number is either a property of the system or a property of what you can see. Not one of them is a property of the joint, the groove, the cut, or the route the water takes. Those are the four things that fail.
This is also why a drawing package has to be written per product family and cannot be templated. A sliding window set and a custom bi fold doors set share a cover sheet and almost nothing after it: different load paths, different panel parking, different hardware schedules, different assembly sequences. In the same way, a vertical bi-folding doors manufacture package and a horizontal slider package do not use the same sheet list at all.
A published Taiwanese site engineering checklist for reviewing aluminium window fabrication drawings contains one line that deserves to be on every approval email ever sent, and to our knowledge it has never been written in English for an export audience.
The architect's door and window elevation is drawn as an exterior view. The aluminium fabricator's elevation is drawn as an interior view. The two documents that must agree with each other are mirror images.
The same checklist gives the method for catching the error, and it is not "look more carefully". It is to walk the floor plan clockwise or counter clockwise and check the left and right edges and the opening direction against it, rather than trying to compare two pictures that were drawn from opposite sides of the wall. Every handing error ever made was made by someone comparing two drawings that were both correct.
It also gives the rule for sliders, and it is the kind of rule that sounds like taste until you work out the consequence. The panel that is used most often should sit on the inner, interior side track. Which sounds ergonomic. Then remember that on a horizontal slider the insect screen is mounted on the outside of the operating panel. So the panel you touch every day is the one nearest to you, the screen is outboard of it on the outer track, and the fixed panel, which is the one most likely to take a knock from a ladder or a ball, is on the far side of both.
And the checklist says the quiet part out loud, in as many words: the dimension on the door and window drawing is not the dimension on the aluminium fabrication drawing. Everything in the previous section is the long form of that sentence.
The published section drawing convention from a Chinese fabrication office is in the same spirit, and equally easy to get wrong when the package crosses a language. Section details must show both a horizontal and a vertical cut; on the horizontal section, the bottom of the sheet is the exterior; on the vertical section, the left of the sheet is the exterior. A reviewer who assumes the opposite will read every drainage slope backwards and every thermal break upside down, while agreeing with every number.
Take a competent casement drawing set and ask what a sliding window adds to it. The answer is not fewer sheets. It is nine items, and six of them are invisible in the finished window.
| Item | Why a slider is harder than a casement here | What to demand on the sheet |
|---|---|---|
| Which panel is fixed | On a two panel slider this decision appears on no other document in the project | A written note plus hatching on both the elevation and the section |
| Is the fixed panel removable | Many fixed panels are direct set, meaning the glass is bonded into the frame and the panel cannot be lifted out at all | The words removable or direct set, in the section callout, not implied |
| Which track each panel rides on | Our published track heights on one frame are two different tracks, not a tolerance band | A per panel track assignment, named on the section |
| Travel direction | "The left panel moves right" is meaningless until you know which side you are standing on | An arrow plus the words exterior view or interior view |
| Drainage route | A slider's bottom track is designed to hold water and send it out; a closed rectangle on the sheet has answered nothing | The route in, the route along, and the outlet positions, drawn as a route |
| Roller rating and glass build | Maximum panel sizes are published against an assumed glass build; change the glass and the number moves | The glass build printed immediately next to the rating, on the same sheet |
| The lock keeper | In at least one major manufacturer's own service guide, the keeper on a horizontal slider is part of the frame or the fixed sash, not a separate hardware item | A callout, because replacing it is a profile job and not a parts job |
| Screen | It is a second panel on the outer track with its own cycle count, and it must exist before the frame is cut | Included or excluded in writing, plus which track it occupies |
| Anchors | A fixing that spans the thermal break turns a thermal break into a decorative stripe | Anchor positions drawn on the thermal break section, with a written note that no fixing, sub sill or clip bridges the break |
Nothing about a slider's finished appearance tells you this, and it is the single most consequential thing on the sheet.
Published service instructions for horizontal sliding windows give a fixed order of work. To remove the fixed panel you must first remove the operating panel: unlock it, slide it to the middle or beyond, lift it up and over the bottom track, and bring it into the room. Only then can you reach the fixings of the fixed panel, which on many products are accessed from the exterior. And some fixed panels cannot be removed at all, because the glass is set directly into the frame and the replacement path is a call to the manufacturer.
Published glazing guidance adds the two numbers that make any of this possible. The head channel usually has somewhere around 10 to 15 mm of clearance above the panel, and that clearance is the entire reason a panel can be lifted out at all. And on some systems the panel has to be slid to one specific position where a removal notch lines up before it will come out. If that notch is not on the drawing, it is discovered on a ladder, with a panel in two hands.
The same guidance covers the glazing beads, and the advice is a sequence rather than a rule. Start with the shortest bead, usually one of the verticals, because beads overlap at the corners and taking them out in the wrong order cracks the neighbouring piece. Pry at the middle first and work toward the ends, never at the corners. Set the blocks at the quarter points. Every one of those is a statement about order, and order is the one thing a photograph of a sample can never show.
A published Chinese fabrication office specification contains three lines that very few buyers ever see, that almost no supplier volunteers, and that are the clearest available test of whether you are looking at a real package.
The same source gives the kerf allowance: 25 mm for a 45 degree cut, and 15 or 10 mm for a square cut. This is the line that explains why a cutting list can never be read off an elevation. Every mitred corner on a sash consumes 25 mm of bar. Four corners is 100 mm of profile per sash that never appears in the window, never appears in the elevation, and is paid for.
None of these three lines is visible in a finished window. All three determine whether the window works. That is the practical answer to the question "can't I just approve from the sample I was sent?" No. The sample is evidence that a sample exists. The three lines above are evidence that somebody has done this before.
The most quoted sentence in a published aluminium window shop drawing guide is also the least followed. The release record should make it possible to answer one question: exactly which windows may be made, to which revision?
Everything else in that guide is a mechanism for answering it.
A published Chinese fabrication process note puts the same idea more bluntly, and it is worth quoting because it changes the tone of the whole exercise. All dimensions and technical parameters must be confirmed in writing by all parties, signed and sealed, before they become the sole basis for production, and that written confirmation is an important legal document. Not a step in the schedule. A legal document.
One more item belongs in that written confirmation, and it is the one most often left out: who performs the first adjustment on site. On a sliding window with rollers, keeps and an interlock, the difference between a window that seals and a window that nearly seals is a hex key and a quarter turn. If that is not assigned to a named party, it will be assigned by default to whoever is holding the tool, and it will not be recorded.
| What you were sent | What it actually is | The one question that exposes it |
|---|---|---|
| A catalogue section | A system illustration, drawn once, for everybody | Does your opening width appear anywhere on it? |
| A quotation drawing | A pricing geometry that has done its job once the price is agreed | Does it carry tolerances, or only overall sizes? |
| A three dimensional render | A sales asset, and a good one | Can a saw operator cut a single bar from it? |
| A photograph of a sample corner | Evidence that a sample exists, which is genuinely useful | Can you see the drainage route, the groove, or how the corner is made? |
| A PDF with no revision block | A document that cannot be superseded | What happens to it when the glass build changes? |
| A drawing with no dimension chain | A picture with some numbers printed on it | Can every cut length be derived from what is printed, or did somebody have to already know something? |
Where that leaves us is a short and honest list. We can issue a revision controlled package with drawing numbers, marks and quantities, in metric, with cut lengths to one decimal place. We can put your opening dimensions on the elevation rather than only ours. We can print the glass build next to the roller rating on the same sheet. We can draw the drainage as a route with outlet positions, and the thermal break section with the anchor positions on it. And we will say in writing which published figures are system values and which are not published at all. What we cannot do is sign for your opening, your wall, or your local code.
| The item | Status | Why the gap exists |
|---|---|---|
| Alloy and temper, 6060-T66 | Published | Identical across the whole range, so it is a floor and not a differentiator |
| Wall thickness, 1.8 to 2.5 mm | Published as a range | A range over a whole section; the drawing must name the member and the rule for measuring it |
| Insulation strip material, PA66GF25 | Published | The material is not the width. Ask for millimetres. |
| Thermal value, 1.3, 1.6, 1.7 | Published as a system value | It belongs to a stated reference configuration, not to your size, split or glass |
| Glass sash thickness, 43 to 90 mm | Published | It is a stack up, not a groove. The groove is a sum of three other numbers. |
| Frame wall width, 90 to 182 mm | Published | It is how far the frame goes into a wall, not the depth of the wall you have |
| Track heights, 53 and 31 mm, or 43 and 8 mm | Published | Two tracks on one frame, not a tolerance band around one track |
| Roller rating per sash, 200 kg, 220 kg, 400 kg | Published | Per sash, and under an assumed glass build that has to travel with it |
| Salt spray hours and shear resistance on one series | Published for a named series | They do not transfer to another series, and they are not air, water or wind ratings |
| Hardware cycle counts | Published as a rig result | A rig test on a component is not the service life of a loaded sash |
| Air, water and wind ratings on the sliding ranges | Not published | Without a declared threshold there is nothing to measure a complaint against, so every leak becomes a negotiation |
| Cut lengths, groove dimensions, anchor layout, drainage route, panel sequence | Not published | Because they are yours and not ours. They are produced, per opening, on the drawing. |
This is the part to copy into an email. Twenty lines. If a supplier can fill fewer than fifteen of them from the package they intend to build from, you do not have a fabrication drawing; you have a quotation with pictures attached.
| Line | Question | Answer |
|---|---|---|
| 1 | Drawing number, revision and date | ____ |
| 2 | Window mark and installation location | ____ |
| 3 | Marking dimension, opening construction dimension, unit construction dimension | ____ / ____ / ____ |
| 4 | Installation method, installation form, gap J | ____ / ____ / ____ mm |
| 5 | Is the wall finish thickness included in the gap | ____ |
| 6 | Manufacturing tolerance, diagonal limit, frame to sash lap width and its tolerance | ____ / ____ / ____ |
| 7 | Glass build, outside to inside, and nominal thickness | ____ / ____ mm |
| 8 | Groove width, groove depth, glazing method | ____ / ____ / ____ |
| 9 | Setting block length, location block length | ____ / ____ mm |
| 10 | Which panel is fixed, and is it removable or direct set | ____ / ____ |
| 11 | Track assignment per panel, and travel direction from which viewpoint | ____ / ____ |
| 12 | Roller rating per sash, and the glass build it assumes | ____ kg / ____ |
| 13 | Drainage inlet, route, outlet positions | ____ / ____ / ____ |
| 14 | Anchor positions, and written confirmation that none bridge the thermal break | ____ / ____ |
| 15 | Screen included, which track, cycle rating | ____ / ____ / ____ |
| 16 | Removal notch position, head clearance, bead removal sequence described | ____ / ____ / ____ |
| 17 | Corner construction callout, and kerf allowance stated | ____ / ____ mm |
| 18 | Revision block present, superseded documents listed | ____ / ____ |
| 19 | Release names revisions, marks and quantities | ____ / ____ / ____ |
| 20 | Named party for the first site adjustment | ____ |
| Ask | A useful answer | An answer that sounds reasonable and says nothing |
|---|---|---|
| Drawing number and revision | A number, a letter and a date | "The latest version" |
| Is this the fabrication drawing or a quotation drawing | Fabrication, with cut lengths and kerf allowance | "This is the drawing" |
| Tolerances | The table, including the lap width | "Standard tolerance" |
| Groove width and depth | Two numbers, plus the glazing method they assume | "It suits 6 plus 12A plus 6" |
| Is the fixed panel removable | Removable, or direct set, in those words | "It can be replaced" |
| Which track does each panel ride on | A per panel assignment named on the section | "Two track" |
| Drainage | The route in, along and out, with outlet positions | "Drainage holes are standard" |
| Corner construction | The named process and the corner component | "Seamless corners" |
| What happens if the glass build changes | Re issue of the affected sheets before anything is cut | "No problem" |
| Who signs the production release | A named person, a revision and a quantity | "Please confirm" |
Yes, and the reason is not the size. It is the mark. The moment more than one opening is involved, the drawing is the only document that connects a window to a location, a handing and a screen. A schedule of standard sizes without drawings produces a container full of correct windows in the wrong rooms.
Yes, and it is probably the best way, because email leaves a record. But the email has to name the drawing number, the revision, the marks and the quantity. A reply saying approved, with a drawing attached, has recorded that a conversation happened.
Because a gap is required, and its size depends on the installation method and the finish thickness, neither of which is the factory's to choose. If the drawing does not say which method and which finish thickness it assumed, the gap is a guess with a dimension on it.
Ask. It may be the marking dimension of the opening, in which case the frame is smaller and the number still has to be resolved through the installation method. It may be the unit construction dimension, in which case it is subject to the tolerance band. It may be the visible glass width, which is a third thing again. A drawing that uses one number without saying which of the three it is has not finished being drawn.
Before the saw, yes, and it costs nothing but a re issue. After it, the change runs through four documents at once: the groove width and depth, the roller rating and maximum panel size, the sash weight the hardware was chosen for, and the glass schedule. That is why a published guide treats a glass substitution as a coordinated drawing revision, not a note.
Put it in the order. Name the drawing package as a deliverable, with a revision number, and require a written production release naming revisions and quantities before anything is cut. The sequence you want is drawing, revision, release, production. The sequence that causes trouble is deposit, production, drawing.