Garbage In, Garbage Out
No method, however sophisticated, extracts truth from flawed inputs; the quality of a result is capped by the quality of its data.
The Oldest Rule in Computing
Garbage in, garbage out states that a computation's output can be no more reliable than its input. A flawless algorithm fed bad data produces confident nonsense. The sophistication of the method is irrelevant; it cannot manufacture information the inputs do not contain, and it will faithfully propagate their errors.
Forms of Garbage
- Measurement error and noise beyond the intended precision.
- Systematic bias: a miscalibrated instrument that is wrong the same way every time.
- Selection bias: data that do not represent the population of interest.
- Missing values handled carelessly.
- Wrong units, mislabeled fields, and silent format corruption.
Why Sophistication Makes It Worse
Powerful methods can be more dangerous with bad data, not less, because they produce polished, plausible outputs that hide the underlying flaw. A crude method may fail visibly; a sophisticated one may fail invisibly, giving a precise answer that is precisely wrong. Precision in the output must never be read as accuracy in the input.
The Defense
The remedy is upstream: validate inputs, check units and ranges, quantify measurement uncertainty, and understand how the data were collected before trusting any analysis built on them. Data cleaning and characterization are not preliminaries to the real work; they are much of the real work.
In Scientific Computing
A simulation is only as good as its physical inputs, transport coefficients, boundary conditions, and material properties, and its output uncertainty must carry the input uncertainty through. Kronos work propagates input uncertainty to results rather than reporting a single clean number, so a reader sees how much of the precision is real.