No Fudged Numbers
We never adjust a number to reach a nicer answer; when a result is inconvenient, we report the result and explain the inconvenience.
The one rule that underwrites the rest
A fudged number is any figure nudged toward a desired conclusion rather than derived from the model and data. It is the failure that makes all other standards worthless, because once one number is adjusted for effect, no number can be trusted. Our rule is absolute: the analysis decides the value, and we publish what it says.
How fudging usually happens
- Choosing an input at the favourable end of its range without saying so.
- Stopping a convergence study at the point that gives the wanted answer.
- Widening a tolerance after the fact until a failing check passes.
- Reporting the best of several runs as if it were typical.
- Rounding in the convenient direction.
Each of these is easy, tempting, and forbidden. The defenses against them are the other standards in this section: stated assumptions, convergence to a criterion set in advance, fixed tolerances, pre-registered predictions, and a reproducible record anyone can rerun.
Inconvenient results stay in
The burner gates are the proof of practice. A plug coil overstressed 3-3.9x, He-3 demand about 400x domestic supply per commercial unit, availability short of Tier III by 30-100x -- these are the opposite of flattering, and they are published exactly as computed. Keeping them is not modesty; it is the guarantee that the flattering numbers, like Q_sci 3.424, were computed the same honest way.
No fudged numbers is the promise that lets every other page in this section mean something. The rule is easy to state and hard to keep, because the pressure to fudge is strongest exactly when a number is inconvenient, which is exactly when keeping the rule matters most.