Quantitative Error Correction
also Numerical Error Correction
Reducing a measurable deviation from a correct value with numerical techniques (e.g. averaging repeated measurements), where error is typically documented with error bars rather than perfectly eliminated.
Quantitative error correction is CF’s name for the math-flavored mode of fixing errors: you are off by some amount, and you reduce or bound that amount with numerical techniques such as taking many measurements and using their average. Elliot Temple distinguishes it from explanatory error correction, where you instead explain what went wrong and devise a different solution that lacks the flaw. Quantitative errors arise from variance; explanatory errors arise from thinking mistakes.
CF’s key observation is that the standard quantitative tool — error bars — does not remove error, it merely documents it (“we’re 95% sure the value is x±y”). Because the error is only labeled, it accumulates across steps. Worse, it compounds non-linearly: adding two ranges widens the result modestly, but multiplying them widens it far more, and exponentiation explodes it. Across the billions of operations a computer or a thinking life performs, undocumented-but-uncorrected error would grow without bound — so genuine correction, not mere labeling, is eventually required.
This connects to CF’s preference for digital error correction and a binary solution space. When only some values are valid (e.g. integers, or just on/off), measurements can be rounded back to the nearest valid value, actually eliminating small errors — but only when errors stay smaller than half the gap between valid values. If every real value were legitimate, rounding helps nothing and you are stuck with whatever you measured. Restricting the solution space thus turns vague quantitative drift into detectable, correctable error, which is why CF favors yes/no questions and small steps over one large, error-prone leap.