Backsights and default accuracy estimates

[Olly Betts]

OK, I've now got backsights implemented, and they seem to work correctly on
a few test cases I've made up.

If anyone's got text format survey data with backsights in which they're happy
for me to have to play with, that'd be very handy.  If you want it to be on
a "don't pass it on" basis, that's fine.

In the process of doing this, I noticed that the default standard deviations
aren't the same as those in bcra5.svx, which I thought they were.

The default is to set the sds to the limit values in the BCRA definition.
But bcra5.svx sets them to a third of these, so that 3 sds (which covers
99.74% of readings for a normal distribution) is the BCRA limit.

It seems that the correct thing to do is to change the default to match the
values in bcra5.svx.  Otherwise the warning about foresight and backsight
differing only fires above about 4.2 degrees, rather than above about 1.4
degrees with the values in bcra5.svx.

This would change the distribution of misclosures when a survey partly used
the default standard deviations, and partly used its own, or those in
bcra3.svx and bcra5.svx.

But perhaps the "3 sds = BCRA limit" is acutally too strict, and both should
change.  The BCRA limits are worded such that *all* readings must be within
the limit, and I believe Bryan Ellis wrote that this was intended.  But that
doesn't work when you want to apply them statistically as normal distributions
tail off rather than stopping.  I believe I intended to choose the "99%" limit
value, and misremembered this as approximately 3 (it's actually 2.565).  So
perhaps that's the most appropriate value.  That would mean warning about a
foresight/backsight mismatch of over 1.6 degrees or so for BCRA grade 5.

Thoughts?

Cheers,
Olly