How accurate is a PDF take-off?

Measuring from a PDF is only defensible if you can say where a number came from and show that it holds up. This page sets out how QS Takeoff produces a quantity, what stops a wrong one going unnoticed, where it can still be wrong, and how to check any of it yourself rather than take our word for it.

1. How a measurement is produced

Open

The PDF is opened and rendered on your own machine. It is not uploaded and it is not stored on our servers - only the take-off data (calibration, markups, your column values) is saved to your account. This is a privacy property, but it is an accuracy property too: what you measure is the original file, not a re-processed copy of it.

Calibrate

Calibration is what turns PDF geometry into real-world length. It is stored per page, so a multi-sheet set with a different scale on every sheet works correctly. There are three routes:

Once set, the calibration is shown to you in full: 1 PDF unit ≈ 0.03528 m, or whatever your page works out at. It is not hidden behind the interface, because a number you can read is a number you can question.

Measure

Length, perimeter, area, rectangle and count, with a live readout while you draw. Where the PDF contains real line work, the cursor snaps to it, so the measurement follows the drawing rather than your aim. Every measurement keeps the points you placed - it can be re-selected, inspected, and its vertices moved. Nothing is ever a number typed in on its own.

Total and export

Quantities are computed from the stored geometry every time they are displayed, so the markups list, the on-screen labels and the CSV export cannot disagree with each other. What you check on screen is what leaves in the file.

2. What stops a wrong number going unnoticed

These are implemented today and each one can be observed in the running app.

3. What can make a take-off wrong (known limitations)

Stated plainly, because a measuring tool that hides its limits is not a measuring tool.

Scanned and image-only PDFs

A drawing that is a photograph or scan of a drawing contains no line work to snap to. Measurement still works, but it falls back entirely on how precisely you can click, and how far you have zoomed in. The scale of that error is worth knowing:

On a vector drawing with snap engaged this table is irrelevant, because your click is moved onto the drawing's own geometry. On a scan it is the accuracy you have. Zoom in before picking points, and prefer longer measurements over chains of short ones.

One scale per page

A sheet carrying a 1:100 general arrangement and a 1:20 detail can only be calibrated correctly for one of them at a time. Measure one, recalibrate, then measure the other - and remember that recalibrating re-values everything already measured on that page. Where possible, measure details on their own sheet.

Printed scale notes can be wrong

A sheet replotted from A1 to A3 usually still carries its original scale note. Nothing on the page announces this. It is exactly why the suggested scale is cross-checked against measured geometry, and why the advice throughout is to verify against a printed dimension before relying on any calibration.

Dimensions edited without the geometry being redrawn

This is the one to watch for, and in our experience it is common. Where a drawing repeats something already drawn - a house type on a scheme, a repeated apartment, a standard detail - it is often quicker to copy the sheet and edit the figured dimensions than to redraw the geometry. The result is a drawing whose written dimensions describe the new unit while its line work still describes the old one.

Everything this tool measures is the line work. Snap makes it more faithful to that line work, not less - so on a re-dimensioned sheet it will trace the original plot precisely and report it with complete confidence. That is the failure mode most likely to put a wrong quantity into a priced schedule, because nothing about it looks wrong.

Which of the two is correct is not ours to decide, and by long-standing convention it is not in doubt: the figured dimension governs, and drawings are not to be scaled. Where the two disagree the written dimension is the authority, the geometry is the error, and a quantity measured off that geometry - by this tool or any other - is wrong. That is a query for the design team rather than something a measuring tool can reconcile.

What the app can do is show you the symptom. When it reads a page it takes a scale from each dimension string against the line it annotates, then reports how well those readings agree with one another - for example "7 dimension strings measure out at about 1:100, but they disagree by up to 3.4% - check it against a known dimension". A sheet re-dimensioned without redrawing produces exactly that disagreement. It is a symptom rather than a diagnosis, since a replot or a stray number on the page reads much the same way, but it is the signal available and it is worth reading rather than clicking past. Note that it is offered on pages that have no scale set yet, so it is worth a glance before you calibrate by hand. Check 5 below is how to confirm what you are looking at.

Curves are measured as the points you place

A curved wall or a radiused edge is measured as a chain of straight segments between your clicks. More points means a closer approximation. For anything where the curve itself is the quantity, place points generously.

Plan geometry only

Measurements are taken in the plane of the drawing. A pitched roof measured on a plan view gives plan area, not sloped area; a raking member gives its plan projection. Any adjustment for pitch, slope or thickness is yours to apply.

Very dense drawings

On extremely large sheets the snap index stops growing in order to keep the browser responsive. Snap becomes less helpful on those sheets; it does not become wrong. Measurement is unaffected.

Touch devices

The precision tools assume a mouse or trackpad. A tablet works, but pointing accuracy on a touchscreen is not good enough for take-off at typical scales.

AI-assisted detection

AI detection is in development and may be switched on or off during the beta, so it may or may not be available to you on any given day. Where it is available it sits outside the measurement path: it can propose where a room or an area is, but the quantity still comes from geometry and your calibration, and any proposal should be checked before it is used. No quantity in this product is produced by a language model.

What the tool cannot check for you

It measures what you point it at. It has no way to know whether you traced the right wall, used the correct drawing revision, or measured to the structure rather than the finish. Those remain professional judgements, and the take-off is your work product, not ours.

Product stage

QS Takeoff is in beta. It is used on real work, and it should be used with the same verification any new instrument deserves - which is what the next section is for.

4. Check it yourself on your own drawing

Five checks. None requires a confidential drawing, and the second is worth repeating on every new set you receive. The first four take about five minutes between them.

Check 1 - the scale chain, with any A4 PDF

Uses no drawing at all, so it isolates the conversion itself.

1. Open any A4 PDF. An A4 page is exactly 210 × 297 mm.
2. Set the drawing scale to 1:1, with paper unit mm and real unit mm.
3. Measure across the full width of the sheet, corner to corner.

Expected: 210 mm. The residual you see is your own click precision at the page corners - zoom in and it shrinks. If the reading is a different number entirely, the conversion is wrong and we want to hear about it.

Check 2 - calibration transfer, on your own drawing

The check that matters most, and the one to make a habit.

1. Calibrate against a printed dimension - the longest one available, snapping to each end.
2. Without recalibrating, measure a different printed dimension elsewhere on the same sheet.
3. Compare the app's reading to that dimension's printed value.

Expected: on a CAD-issued drawing with snap on, agreement to the millimetre or very near it. A consistent percentage error across several dimensions means the sheet is not at the scale it claims, or the calibration reference was mis-picked - recalibrate against a longer dimension and repeat. A scatter, where some dimensions agree and others are out by odd amounts, means either you are working on a scan or the sheet has been re-dimensioned without being redrawn; check 5 tells those two apart.

Either way, calibrate against a dimension you have already cross-checked against a second one. Calibrating against an edited dimension scales the whole page wrong, and every quantity on it with it.

Check 3 - internal consistency

1. Measure a rectangular room with the Area tool.
2. Separately measure its length and its width.

Expected: area equals length × width to the displayed precision. This confirms length and area are derived from the same calibration rather than by separate routes.

Check 4 - repeatability

1. With snap on, measure the same wall twice, approaching it differently each time.
2. Repeat with snap off, or holding Shift.

Expected: identical to the digit with snap on. The spread you see with snap off is the hand-aim error snap is removing, measured on your own drawing at your own zoom level.

Check 5 - repeated types

Worth running once on any scheme with a repeated unit - house types, apartment types, standard details. It is the check that catches a copied and re-dimensioned sheet.

1. Find the same element on two instances of what should be the same type.
2. Measure it on each.
3. Compare both measurements to the figured dimensions printed on each.

Expected: the two measurements agree with each other and with both printed dimensions. Two readings that are identical to each other while the printed dimensions differ means the sheet was copied and re-dimensioned rather than redrawn, and the geometry is still the earlier unit. Printed dimensions that agree while your measurements don't means the reverse - the geometry was edited and the dimensions were not. In both cases the drawing contradicts itself, and that is something to raise rather than a number to price.

5. Evidence and updates

This page describes behaviour that is live in the product today. It is reviewed whenever a change affects any claim on it, and the date below moves with the content rather than on a schedule.

If you run check 2 on a drawing and it does not agree, we would like the details - the scale, whether the sheet is vector or scanned, and the two dimensions involved. A reproducible disagreement is the most useful thing anyone can send us, and it is the only way this page stays honest.

Related: calibrating a drawing in the user guide, and how snap-to-geometry works.