[Therion] Aven colour by loop error

[Olly Betts]

On Thu, Feb 23, 2017 at 12:08:09AM +0100, Stacho Mudrak via Therion wrote:
> Hi, I agree with Bruce, that it is really a very useful feature. I have
> just pushed it on GitHub (see commit 2ebc7ed).
> 
> The only problem is, that survex calculates error which "is the ratio of
> the observed misclosure to the theoretical one". I have implemented more
> simple export (it was much easier), currently therion .3d export contains
> error equal to 2 * relative error of a loop. (because of 0-12 aven fixed
> error scale). It should be possible to extract survex relative error from
> either thTMPDIR/data.err or data.3d file, but it will take some time.

I'd definitely use the .3d file (no rounding, and a file format that's
intended for machine parsing).

> But when I compared 3D files with relative loop error and error with survex
> ratio, I received two quite different pictures. Are these two measures not
> comparable? Is a relative error of a loop useless number?

tl;dr: No and yes!

By "relative error of a loop" I'm assuming you mean "misclosure / length
of traverse * 100%".  If so that's not really a useful measure, though
at least historically it's been a fairly popular one, I suspect because
it's an obvious one to calculate.

The problem with it is that what's a "good" value depends on the length
of the traverse.  For random errors, the squares of the errors add, so a
traverse which is 100 times as long would be expected to only have 10
times the error, which means the relative error would be expected to be
10/100 = 1/10 as much.

As well as not providing a reliable measure of the random error, a
blunder in a long traverse will tend to be smoothed out in the relative
error, as it gets divided by the length of the traverse.

Observed error divided by theoretical error should be agnostic to length
of traverse.  It's simply telling you which traverses don't seem to be
surveyed to the accuracy that was claimed.

> Has anybody tried to go to a cave and resurvey "red shots", whether
> they are really so bad?

I've done the armchair equivalent and used Aven's error colouring to
successfully find mis-ties, reversed legs, etc.

Cheers,
    Olly