Eleven of fourteen sheets are clean.The set is still not measurable.

It is a $99 deterministic gate that grades a construction drawing set before anyone measures it, and returns TAKEOFF-READY AS DESCRIBED, PARTIAL, or NOT MEASURABLE. You mark six properties on each sheet and six on the set as a whole; it returns the verdict, the sheets holding the set back, and the one thing to fix first. It ships a zero-dependency Python engine, a live-formula Excel workbook, two playbooks and a worked 14-sheet example. It grades the set as you describe it — it never opens your drawings.

Because a takeoff is only as good as the weakest drawing it reads. One sheet at an undeclared scale invalidates every quantity derived from that sheet no matter how clean the other forty are, and an average would let thirty-nine good sheets hide it. In the worked example that ships with the kit, the set-level controls score 92 and read TAKEOFF-READY AS DESCRIBED, eleven of fourteen sheets are READABLE, and the verdict is still NOT MEASURABLE — because one elevation sheet that itself scores 83 has no declared scale. Averaging is exactly how that set gets measured anyway.

Those tools measure. This one decides whether measuring is safe yet, and it is not affiliated with any of them. An AI takeoff tool will return quantities from a set it should have refused, because it has no way to know an elevation was rescaled when someone printed it to PDF at half size. This runs before that tool does, on the same set, and tells you which sheets will produce numbers you cannot stand behind. It makes takeoff software more useful, not less — most sets come back PARTIAL, which means proceed and hand-check the flagged sheets.

The gate is worsen-only and dispositive: it forces NOT MEASURABLE regardless of both scores when any measured sheet has no declared scale, or when the set has neither a controlled source nor a revision log — meaning nothing establishes which version you are holding. A high score plus a fired gate is the most common and most useful result, because it says the set is nearly fine and names the two or three sheets standing between you and a defensible number. The gate releases the moment the trigger clears, and it can only ever make the verdict worse.

Not on the three that matter. Scale declared on a measured sheet, revision log present, and single controlled source cannot be marked not-applicable, and the engine refuses the input rather than quietly dropping it — because marking one N/A would remove it from the divisor and from the gate at the same time, which is how a gate gets waived into silence. Elsewhere a blank genuinely means not applicable and is renormalised out. A schedule or general-notes sheet you never measure can be marked measured = N, which takes it out of the scale gate, and every run prints how many sheets you took out of scope.

That the verdict is a statement about the set as you described it in the file, not about a set anyone inspected. Nothing is uploaded, nothing is scanned, and no drawing is opened — every input is an answer you supply about a set you are holding. The suffix travels with the verdict because the verdict travels: it is the kind of line that ends up in an email to a GC or an architect, and whoever reads it there did not supply the inputs and should not have to guess.

No, and it is checked rather than asserted. The workbook's formulas are read back out of the shipped .xlsx and compared against the engine across 7,645 cases with zero mismatches, including the tie-break that decides which single item is named as fix-first. There is one deliberate difference: a blank scale on a measured sheet is a state the engine refuses outright, and neither Excel nor a browser can refuse, so both treat it as missing and fire the gate. All three err in the same direction.

It is for the general contractor, subcontractor or estimating team that owns the margin on a bid — the firm, not the seat. It pays for itself the first time it stops a set from being measured twice. It is not for anyone who wants a quantity, a cost, or a bid recommendation; it produces none of those. It is not a scanner, not a takeoff, not an estimate, and not legal or contractual advice. If you want the commercial read on whether your bidding process learns from its own history, that is a different instrument.