Every length in the takeoff is wrong by the same percentage

If your counts are right and every length is wrong by roughly the same proportion, you have a scale problem. Nobody mis-measured. The tool converted page geometry into real-world dimensions using a factor that does not match the drawing, and it did so consistently, which is why the error is invisible on any single measurement and obvious across a set.

Areas make it visible first, because area error is the square of linear error. A 3 percent linear error is a 6.09 percent area error, and a 5 percent linear error is a 10.25 percent area error, so finishes and slab quantities drift out of plausibility while door counts stay perfect. That asymmetry is the signature.

Confirm the signature before you touch a calibration

Three symptoms have to hold together. If any one of them fails, the cause is not scale and recalibrating will make things worse by hiding it.

• Counts are unaffected. Devices, doors and fixtures come out right, because counting does not use the scale factor at all. If counts are also wrong, you have a different problem — most likely the sheet-level extraction issue behind an automated count that finds devices on some sheets and none on others.
• Lengths are wrong by a constant ratio, not a constant amount. A fixed offset in metres points at a start-and-stop rule, not at scale.
• Areas are wrong by approximately the square of that ratio. If areas are off by the same proportion as lengths, something is applying the correction inconsistently and the calibration is already partially in place.

The measurement that names the cause: 3 dimensions, 3 sheets

1. Pick 3 sheets from different disciplines — an architectural floor plan, a structural framing plan and a services plan. Sets are assembled from several consultants' files and the failure is often confined to one of them.
2. On each sheet, measure 3 things whose true dimension is knowable independently: a dimensioned structural grid bay, a standard door leaf at 900 mm or 3 feet, and the printed bar scale from end to end.
3. Compute measured divided by stated for all 9 readings. Record them; do not average them yet, because the spread is the diagnosis.
4. Compare within each sheet first. If the 3 readings on a sheet agree to within about 0.5 percent, that sheet has one clean scale factor.
5. Then compare across sheets. Identical ratios across all 3 sheets means a global problem with a global fix. Different ratios per sheet means the set is mixed and no global fix exists.
6. Finally, measure one long horizontal and one long vertical dimension on the same sheet. If those two disagree, the sheet is non-uniformly scaled, which changes the answer entirely.

What the ratio itself tells you

The number you computed usually identifies the cause on its own, before you investigate anything.

The three cases where a global recalibration is the wrong answer

A single calibration is tempting because it is one action and it makes the first number you check come out right. Three situations punish it.

• Mixed scales in one set. Plans at 1:100 with enlarged toilet plans at 1:50 and details at 1:20 is normal practice, not a mistake. Calibrating globally to the plans silently multiplies every detail measurement by 5, and detail sheets are where the expensive assemblies live.
• A revised sheet reissued at a different page size. The set is now half at one factor and half at another, and the revision cloud draws attention to the changed content while nothing draws attention to the changed page. Any sheet issued after the original set needs its own reading.
• Rotated or cropped sheets. Cropping does not change scale, but a page with a non-zero rotation entry can produce x and y factors that differ if the tool stored them separately at calibration. That gives correct lengths along one axis and wrong areas everywhere, which reads like a measuring error rather than a scale error.

A global calibration on a mixed-scale set does not fix the takeoff. It makes the plans right and the details wrong, and moves the error somewhere nobody is looking.

Decide: one calibration, or a per-sheet scale table

1. All 9 readings agree within 0.5 percent: apply one calibration, record the factor and the reference dimension you used, and re-run the affected conditions.
2. Readings agree within a sheet but differ across sheets: build a per-sheet scale table — sheet number, page size, stated scale, measured factor, who verified it and when — and store it with the job rather than in the tool's settings.
3. Readings differ between the horizontal and vertical on one sheet: stop. Ask the design team for the vector file or a fresh export at 100 percent. Measuring a stretched scan produces defensible-looking numbers that cannot be defended.
4. Readings drift across a single sheet: same answer. This is a scanned original, and no calibration models the distortion.
5. In every case, re-run the quantities rather than scaling the previous results. The measurement rules that survive a recalibration are the ones held on the takeoff condition record — waste, deductions and unit conversions — and applying a factor to a finished quantity applies it to those too.

Record the factor even when it is 1.000. A verified scale note per sheet is what lets a second estimator reproduce your numbers, and its absence is one of the quieter contributors to two estimators returning different quantities from one set — although boundary definition is the larger one, and the two are worth separating before anybody is blamed.

What this costs downstream, and what to do about it

If the discovery happens before submission, the work is a repricing of affected conditions only, run the same way as repricing an estimate after an addendum: identify what the change touches, reprice that, and keep an audit trail explaining the movement line by line. If it happens after submission, you have a qualification to write and a decision to make about which number stands.

There is also a bidding consequence worth naming. A set that cannot be measured reliably — scanned sheets, inconsistent page sizes, missing bar scales — depresses subcontractor response, because a sub who cannot quantify quickly simply moves on to the next invitation. When you are diagnosing why invitations go out and almost nobody bids, the measurability of the documents belongs on the list of causes alongside reach and reputation.

The durable fix is procedural: a pre-measurement check that reads page size, stated scale and measured factor for every sheet and flags the ones that disagree, run before an estimator starts rather than after. That is a small tool, not a platform, and it is the kind of narrowly scoped internal system we describe in MVP and product builds. The rest of this silo sits under preconstruction, takeoff and estimating, inside our wider construction and contracting work.

• takeoff
• drawing scale
• calibration
• estimating