Twenty-five problems found on paper before anything was cut
Reviewing a design on paper, and what that exercise actually caught.
A design review on paper works by making a small group of people walk through every dimension, every fit and every assembly step out loud before a single part exists, and it caught twenty-five real problems in that particular set of drawings precisely because a problem spotted in that room cost nothing more than a marked-up print, while the same problem discovered later would have cost a finished, and wasted, part.
Why a review works before anything exists
A drawing describes a part completely, every dimension, every tolerance, every surface finish, yet almost nobody reads a full set of drawings the way they read a novel, start to finish, checking that each page is consistent with the ones before it. A design review forces exactly that kind of reading, with several people looking at the same drawing from different angles, one checking that two mating parts can physically be assembled in the order specified, another checking that a stack of individually reasonable tolerances does not add up to an impossible total gap, a third simply asking what happens if a dimension on page four turns out to disagree with the same feature on page seven. None of these checks require a physical part to perform, which is exactly why doing them on paper first is so much cheaper than doing them for the first time at the workbench.
The reason a group catches problems a single careful person tends to miss is not that any one reviewer is more diligent, it is that each person walks into the drawing set with a different question already in mind, one thinking purely about how the part gets machined, another thinking purely about how it gets assembled, a third thinking about how it will actually be used once finished. A single designer checking their own drawing tends to check it against the same mental model that produced it in the first place, which is exactly the model most likely to contain the error, since a mistaken assumption rarely announces itself to the person who made it.
The wedding-invitation comparison
A wedding invitation proofread carefully on a screen, checked line by line for the date, the venue and the spelling of every name, costs nothing to correct the moment an error is spotted, a keystroke and a re-read. The same invitation proofread only after two hundred copies have already been printed and half of them mailed turns the identical typo into a much larger problem, an apologetic reprint at best and an invitation that quietly says the wrong thing at worst. A drawing carries exactly the same asymmetry, a missing dimension or a tolerance that leaves no room for an o-ring costs a red pen mark before anything is machined and a scrapped, re-ordered part afterwards, and the review exists purely to make sure the mistake gets caught while it still only costs a red pen mark.
Where the twenty-five problems actually came from
Sorting the twenty-five caught problems afterwards showed they clustered heavily around a small number of recurring categories rather than spreading evenly across the drawing set, which is itself a useful finding: a large share were tolerance stacks, several individually generous tolerances on parts that had to nest together adding up, in the worst case, to a gap or an interference nobody had checked for directly. A second cluster involved datum references that looked fine on a single part's drawing but did not agree with the datum scheme used on the part it had to mate with, so two drawings that were each individually correct still failed to describe a pair of parts that would actually fit together. The remaining problems were mostly assembly-order issues, a fastener specified in a position that no tool could actually reach once the surrounding parts were in place, the kind of mistake invisible on a single part's drawing and obvious the moment someone traces the full sequence of putting the assembly together by hand, on paper, before committing to it in metal.
The one number worth remembering
Roughly a third of the problems caught in that review were stack-up issues that no single drawing showed on its own, only visible once someone deliberately added the tolerances of several mating parts together and checked the total against what the assembly actually needed. That fraction matters because it is exactly the category of error a rushed, single-person check almost never catches, since checking one drawing at a time for internal consistency says nothing about whether several individually correct drawings still add up to a design that physically works together.
What this changes in practice
A review that only checks whether each drawing is legible and complete catches typos and missing dimensions but misses almost everything that made this particular review worth running, which means the useful version of a design review has to deliberately walk the assembly sequence and deliberately sum the relevant tolerance chains rather than simply reading each sheet in isolation and moving on. Building that habit costs an afternoon of several people's time before anything is cut, a cost that reliably looks large in the moment and small in hindsight once even one of the twenty-five caught problems is compared against what it would have cost as a finished, scrapped part.
Where this stops being true
A paper review only catches what the drawings themselves can show, so it says nothing about how a part actually feels in a hand, how a surface actually looks under real light, or how a mechanism actually behaves once friction, vibration and a real operator are involved rather than an idealised one, all of which stayed genuinely unknown until a physical prototype existed regardless of how carefully the drawings had been read. Treating a clean design review as proof that a design is finished, rather than as proof that it is internally consistent, is its own kind of mistake, and the twenty-five problems caught on paper were never a substitute for the separate round of problems a first built prototype went on to surface in its own right.