[Talk-GB] Features which move...apparently spurious edits: iD bug or "finger trouble"?

[Dan Glover]

If there's a better place, please direct me appropriately...

I've been using Robert Whittaker's Post Box tool to help fill gaps and 
fix anomalies in the CT postal area. I think I've now found a pattern 
which leads to "ghost" entries in locations where there has never been a 
post box and leaves the actual post box either unmapped or with a new 
node.  I have three examples where the general scenario seems to have 
been:

1. Mapper "A" creates a node with amenity=post_box.  Other details such 
as reference and collection time may or may not have been entered at 
this point.

2. Time passes, possibly with edits to the node, but no change of 
position.

3. Mapper "B" does something apparently unrelated. In the examples I 
have seen it involves multiple ways/nodes, though not necessarily vast 
numbers.

4. The node created at (1) is re-positioned in a fairly random manner as 
part of the same changeset.

5. [possibly] Mapper "C" spots the missing post box and creates a new 
node for it.  The node from (1) is still "out there", in one case it was 
1.3 km from the original (correct) position.

Note: it transpires Mapper "A" in the three examples is the same user.  
Three different "B"s.

I suppose the first questions are:

- Has anyone seen something similar?  Presumably this could happen to 
nodes generally, there is no reason to think amenity=post_box is a 
factor.

- Is there any way to identify features which have moved by more than 
whatever distance might be considered "normal" for someone correcting a 
previously incorrect position?  This might be key to further 
investigation and correction.

I can provide the node and changeset numbers if anyone wants to look 
into the details and perhaps can spot a pattern,  The edits are by three 
different users on widely spaced dates and the iD versions are all 
different.  The most recent example was in September 2018, so it's not 
likely the mapper would remember anything.

Robert's tool shows the distance between OSM node and Royal Mail data, 
which is how I found one of the examples - but it is "normal" for RM 
data to have discrepancies, sometimes fairly significant.  The other two 
had relatively minor offsets and were picked up through "local 
knowledge" - but analysis of old/new position could have highlighted 
them.  Unfortunately post boxes do get moved whilst retaining their RM 
reference but I'd expect that to be done in OSM by creating a new node 
and deleting the old.

There are some philosophical question here: which features are "allowed" 
to move and by how much?  Natural features probably shouldn't.  Man-made 
ones are probably something new when they do move.  Boundaries, however, 
are subject to revision, roads and footpaths get re-aligned. Also what's 
an acceptable margin for a correction?


Dan