The two-sheet problem
Drawing sets don't fail one sheet at a time. The expensive mistakes live between sheets — and cross-referencing them doesn't scale with headcount.
Sheet A-411 shows a two-hour fire barrier. The wall type is tagged, the rating sits in the schedule, the detail is drawn correctly. Sheet M-201 runs a supply duct straight through that wall. No damper, no penetration detail, no note.
Each sheet, taken alone, is fine. A reviewer could stare at either one all afternoon and sign it in good conscience. The set is wrong — and the error costs real money, because it surfaces as an RFI at best and as a failed inspection behind closed drywall at worst.
That is the two-sheet problem. Almost every expensive drawing mistake has this shape.
Review is built for one sheet at a time
Watch a reviewer work. They open a sheet, check the title block, walk the notes, verify dimensions, scan for the obvious. This is what people are good at: reading a page closely. Markups, comment logs and checklists are all organized the same way — per sheet.
But look at what actually gets projects in trouble:
- A door scheduled at 45 minutes sitting in a wall the plan rates at 90.
- Detail callout 5/A-501, referenced from A-201, pointing at a detail that no longer exists in the current revision.
- A roof tag with no matching schedule row.
- A structural opening that moved in rev C while the mechanical sheet still penetrates where it used to be.
None of these live on a sheet. Every one is a relation between two sheets — often between two disciplines, drawn by two people, in two different months.
Why headcount doesn't fix it
The instinctive fix is more reviewers, or more senior ones. The arithmetic says otherwise.
Single-sheet checks grow linearly with the set: 300 sheets, 300 passes. Cross-sheet checks grow with the pairs. A tag on one sheet can be contradicted by any sheet that touches the same wall, door, duct or datum. Nobody reviews all pairs — reviewers carry an index in their heads and spot-check the relations they remember. That index is experience, it is real, and it does not transfer, scale or survive a busy week.
Splitting the set among more people makes it worse in a specific way: the misses concentrate exactly where the split is. Architecture reviews A-, mechanical reviews M-, and the fire wall with the duct through it belongs to neither. Divide the work by discipline and you have rebuilt the two-sheet problem at the level of the org chart. The seam between reviewers is the same seam between sheets.
Checklists don't close the gap either. A checklist tells you what to confirm — "penetrations through rated assemblies are protected" — but not where the other half of the fact lives. To run that check honestly you must find every rated wall on every plan, then find every element on every other sheet that crosses one. The checklist takes one line. The lookup takes the afternoon, per rated wall.
The set is a database that nobody joins
A drawing set pretends to be a stack of pages, but it behaves like a database: tags, callouts, schedule rows, ratings, revision marks — keys and references everywhere. Coordination errors are broken joins. A callout with no target is a dangling pointer. A schedule row that disagrees with its plan tag is two records that failed to reconcile.
Nobody would audit a database by printing the tables and reading them one at a time. That is roughly the state of the art in drawing review.
The fix is mechanical, not heroic: extract every reference and every rated claim from every sheet, resolve them against each other, and surface the contradictions — each one carrying its two source locations, so a reviewer can open both sheets and verify the finding in under a minute. The judgment stays human. The lookup should not be.
Reviewers are expensive because their judgment is. Spending that judgment walking a 300-sheet set hunting for the second half of a fact is the most costly way anyone has yet devised to do a join.
The first pass over a set should read all of it at once. That is the standard we build to — and the next post in this series is about a regulator that has started demanding, in law, that the result be documented.