Gear
The direction of the error
Most instrument errors here push one way: stretched tape reads long, tape thickness reads long, tab offset reads short, single-diameter calliper reads short. Knowing the sign helps.
Most of the instrument errors on this site are not random. They push in one direction, consistently, which means that once you know which instrument you used and how, you can say more than "this number has error in it" - you can say which way the true figure probably sits relative to what you read.
The table
| Instrument or fault | Typical direction | Why |
|---|---|---|
| Stretched cloth tape | reads long | a tape that has stretched shows more distance than actually wrapped, tested by laying it against a rigid ruler at 10 and 20 cm |
| Tape thickness | reads long | the printed scale sits one thickness out from the skin it wraps, adding roughly two pi times the tape's thickness to a girth reading |
| Tab offset (sewing tape) | reads short | anchoring the metal end tab as though it were the printed zero cuts the true reading by the gap between them, as with a ruler's own zero offset |
| Ruler zero offset | reads short | a scale that starts a little in from the physical end returns a bone-pressed length that is short by exactly that margin |
| Single-diameter calliper reading, converted with pi | reads short | a shaft cross-section is closer to an ellipse than a circle, and one diameter times pi under-states the true circumference |
| Steel retractable tape used for girth | reads high in some spots, low in others | the blade resists curving into a circle and sits away from the surface in places, so the error does not run one clean direction |
| Reading a ruler from above and behind | reads long | the viewing angle foreshortens the scale and shifts where the tip appears to land |
Why the sign is worth knowing
A random error tells you a reading could be off in either direction by roughly some amount, and there is not much more to say about any single reading than that. A systematic error with a known direction lets you do something more useful: if you know your tape has stretched, you know the true girth is smaller than what you read, not merely "different by some amount." That turns an uncertain number into a bounded one - a ceiling or a floor rather than a shrug. The general distinction between error that averages out and error that does not is worth having before this table means much.
Combining directions
Where two errors in the same reading push the same way, they compound rather than cancel - a stretched tape used to read a girth already inflated by its own thickness will read longer still, on both counts. Where they push opposite ways, they can partly offset each other by coincidence, which is worse in one sense: the reading can look accurate while both of its component faults are still present, uncorrected, and only luck cancelled them for that one measurement. Neither situation is a reason to skip a real check - testing your actual instruments against a fixed object, rather than assuming a direction from a table, catches whichever combination you happen to have. A single reading tells you nothing about which of these directions, if any, is present; a table like this one is a starting hypothesis to test against your own instruments, not a substitute for testing them.
What this table cannot do
Knowing a direction narrows a range; it does not hand you a corrected number, because the size of most of these effects varies with the specific instrument and how it is used. It also does not extend to anything outside a physical reading. An AI-derived estimate from a photograph does not have a "direction" in this sense at all, because it is not measuring against a scale with a known fault - it is inferring a figure from an image, which fails differently. A rating carries no directional error either, since a score is a position in a distribution of photographs rather than a length that can read long or short, and a person's stated impression is an opinion rather than a biased instrument - worth treating as a judgement rather than correcting for a sign. Rate Cock's output belongs in that same category: not a reading with a systematic direction to bound, because it was never claiming to be a length in the first place.