Here I go again…. part 2.

I was pleased to see someone pick up on my first diary item. OSM weekly is hardly the New York Times, yet I know that the editors like posts which are constructively critical, and they did spot that I was hoping for something to happen which would let us all “do better”.

I decided to look in more detail at how the lack of detail was leaving validators with a lot of mapping to do. Evidence of this is seen here, where the mapper marking the tile as “completely mapped” is nowhere near being the main contributor of objects and the validator – DeBigC – has add or adjust 62% of the objects in the tile. This shows the last mapper to touch any object. Note: I do accept that this is one tile, but it’s not unusual to find this all over the Fingal task.

There is an age-old idea that validators are “second mappers”, who mop up the unnoticed details that mappers leave behind. There have to be limits to that, and there usually are. I have seen in HOTOSM activations where validation standards of completeness are quite strict and fussy with any more than a handful of missed objects getting a do-over request. That was never the intent with #osmIRL. That said the validator mapping more than the person marking the tile as done is the hallmark of needing to signal the tile cannot be validated.

The most useful tool for measuring how much is being left behind is the Heigit dashboard. Right here a query on the Fingal area from a date proximate to the last tile being marked as done yields up the size of this gap. The baseline before validation started on December 18th 2023 was 110k buildings, and up to November 10th 2024 this had risen to 118k. In context this is an extra 8k buildings, meaning 7.5% of the intended task objects were missed by mappers. On the face of it a 92.5% completeness rate would seem good, but then this relies on several factors, which I will set out here now and state that I may reveal some deeper problems.

The fact that many of the heavily developed and dense urban areas are unvalidated as of today’s date then this must mean that the 7.5% missing rate is going to increase significantly.

But this is far from the only issue to be revealed as validation continues. I have observed that tasks that are incomplete are being validated. And because a small number of mappers were involved that this issue was repeated and replicated.

• Some mappers validated their own tasks. This is literally a case of the blind validating for the blind. A mapper doesn’t see, cannot recognise or has misinterpreted the task instructions they are going to simply compound their error via error cosanguinity.

• Some mappers and validators had a common mis-interpretation of the tasks. This is literally the problem of group-think, assuming that the mapper’s take is correct, or not having the experience to assert that the mapping needed to be re-done.

On the basis of these last two points I am positing that a 92.5% completion rate is optimistic. I will post again demonstrating proof of the issue with validators.