Version 1 Files in DataStore

(IATI Technical Team) #1

Hi all,

As you are no doubt aware, a new IATI DataStore is in the process of being built. This topic has been raised to update and inform the community about what will happen with files in version 1 of the Standard when the new DataStore is launched.

The IATI standard has undergone a number of upgrade processes with one major upgrade - moving to Version 2 - back in 2015. Version 1 contains outdated rules and guidance on publishing development and humanitarian data on resources and results. The Members’ Assembly and consequently the Technical Advisory Group made the decision to deprecate IATI v1 by 30th of June 2019 (see this post).

In the context of developing the new IATI Datastore, the Tech team suggested (see this discussion thread) that a one one-off upload of v1 files will take place, after which v1 files would not be ‘allowed in’. In April 2019 the IATI tech team met with the vendors of the new datastore and validator service (Zimmerman and Zimmerman along with Data4Development) and scoped out how the two systems will integrate. Following investigation of various options, the teams came to the conclusion that it will not be possible to do a one off upload of v1 files. We will share more information on the system integration of the two products later in May with an update on the project timeline.

Analysis of publishers using v1:

Following this confirmation, the technical team engaged in analysing the makeup of organisationswhich still have at least one file published in v1 of the Standard, to understand how publishers might be affected.

Please see the full list of v1 publishers and summary stats in this sheet (https://docs.google.com/spreadsheets/d/156WZ9j8X8NzKzCJCyVuJSHCAluk2LvhDIfKkuC_kI5g/edit?usp=sharing). The headline stats from the analysis listed below show that the majority of v1 files have not been updated in the past 12 months and the majority are organisation files:

  • 201 publishers have at least one file in v1. Out of these:

    • 81 (40%) have all of their files in v1 of the Standard

    • 120 (60%) have at least one file in v2 of the Standard

    • 61 (30%) have ‘active’ Activity files - updated their data in the last 12 months.

    • 140 (70%) have activity files that have not been updated in the last year

  • 132 publishers have only a single file in v1. Out of these:

    • 41 (31 %) have updated their Activity file in the last 12 months.

    • 83 (62%) of the publishers with a single file in v1 have an Organisation file.

    • 107 (81%) out of the 132 publishers with a single file in v1 are using AidStream (including both activity and organisation files)

Tech team’s support for v1 publishers:

Since the decision to deprecate v1 of the IATI standard, the technical team has been in touch with all 201 publishers to notify them of the timeline of deprecation and the impact on tools when keeping their files in v1. We will continue engaging with publishers and take the following actions:

  • Continue engaging directly with the 20 publishers listed as ‘targeted publishers’, including IATI members and those with a high number of files who still have at least one file in v1. We have been liasisng with them over the last 6 months and all except two are actively engaging with us about their transition plans. Some are very close to fully transitioning to v2.

  • Directly contact the 52 ‘active’ publishers who have updated their data in the last year and have at least one file in v1 of the Standard. This is to follow up on an email that was sent to all v1 publishers in November and again in January, informing them of our recommendation to transition. We would again offer to provide 1-2-1 support to help them move their data to v2.

  • Continue to support other publishers who are getting in contact following the bulk email.

What happens to data that remains in v1?

While we hope to help to ensure that as much v1 data is transferred to v2 as possible, it is inevitable that we will be left with a number of v1 datasets. For instance, while we hope to hear back from the publishers we get in touch with, there will be cases where we receive no response.

  • All v1 data will continue to be accessible via the ‘old’ IATI Datastore until the end of 2019.

  • All v1 data will continue to be available via the IATI Registry, despite not being consumed by the new Datastore.

  • D-portal will also continue to utilise v1 data in the immediate future; the Technical Team are currently looking into potential impacts of this.

2 Likes
IATI Datastore - what data should go in?
(Andy Lulham) #2

Really useful bit of analysis – thanks @IATI-techteam.

The above post doesn’t say whether v1.0x datasets will be included in the v2 (“new”) IATI Datastore. Can I take this opportunity to again flag IATI transformer (described in this discuss post) which I think could be of use here. One option would be for the v2 (“new”) IATI Datastore to run any v1.0x IATI data through IATI Transformer as a pre-processing step. This is really trivial to do – you can even do it on the fly. So for instance, instead of using:

http://aidstream.org/files/xml/icauk-tg.xml

You’d use:

https://iati-transformer.herokuapp.com/transform.xml?url=http%3A%2F%2Faidstream.org%2Ffiles%2Fxml%2Ficauk-tg.xml

Does that sound like a viable option? In that way, every effort would be made to ensure v1.0x data isn’t being dropped completely.

2 Likes
(IATI Technical Team) #3

Hi Andy,

Apologies if it was not clear from the above post; we can confirm that v1.0x datasets will not be consumed by the new DataStore.

It is important to bear in mind that the new Datastore is not a database; rather it is a curation and display of files that are hosted on the IATI Registry. As such, unless publishers themselves host the transformed file after using your transformer tool and link back to the Registry, the dataset will not be picked up by the DataStore.

It is for this reason that we cannot ourselves run all v1 data through the transformer tool. We do however appreciate its availability as a public tool that can be used by publishers who are undergoing the v2 transition and we have certainly raised it as a potential transition option for the publishers that we are in touch with.

(Andy Lulham) #4

Apologies, you’re absolutely right. It was clear – I’m afraid I didn’t read it carefully enough.

Okay, great! This is a good decision in my opinion. I don’t see how the one-off upload would have worked (in terms of data ownership) so I’m relieved to hear this has changed.

I’m not sure I understand what this means. The datastore performs an ETL (Extract, Transform, Load) cycle. I think the “L” bit means it is a database. The “T” bit means it already performs a transformation/normalisation step. I’m just suggesting adding to that transformation/normalisation step.

(Matt Geddes) #5

Is there an easy (or even a known) way to check how many of those v1 files contain active activities, i.e. activities that have not yet reached their predicted end date, as those are the things that would be most key for me to keep in the new datastore - if you could share a list of the urls for the from the great analysis above then I might be able to figure this out myself?

(Amy Silcock) #6

Hi @matmaxgeds, if I was to do the above (someone more technical than me probs has a better approach) I’d use the current IATI datastore.

  1. download the activities for all publishers we know are using v1, and that have activities with an end date greater than today.
  2. filter out all v2 activities (ones with numeric codes).

That should provide you with a list of v1 activities that are still active. Or, use the activity-status attribute as a filter as there’s some debate whether it’s updated more regularly than the activity dates.

1 Like
(Matt Geddes) #7

Thanks @amys - If I get time to have a go I will be sure to share

(Andy Lulham) #8

@matmaxgeds you can use iatikit to get the answers you’re after.

Here’s a jupyter notebook demonstrating how.

In summary:

  • Total v1.0x organisation datasets: 116
  • Total v1.0x activity datasets: 594
  • Total v1.0x activities: 68,698
  • Total v1.0x activities with activity-status of “pipeline/identification” or “implementation”: 24,125
  • Total v1.0x activities with no end date, or an end date in the future: 28,057

Hope that helps.

1 Like
(Matt Geddes) #9

Hah, @andylolz, I ‘hope it helps’ is underselling it somewhat :slight_smile:

Based on this, I feel that not having v1 data in the new datastore would be a substantial loss until we can get this figure down substantially.

I am aware that DI are working hard with many of the key publishers still using v1

Suggested ways forward that I am aware of in the interim (until the publishers update their systems):

  1. Not implement the v1 aspects of validation yet
  2. Keep the existing datastore online
  3. Have the new DS use the iati-transform tool to convert them to v2 data
  4. Republish them all using iati-transform and a secondary publisher (although they still wouldn’t be available in d-portal and several other tools)
  5. Explore how many of these activities are actually closed but where the data has not been updated

In light of my recent commitments to being a good discuss citizen - how about we have a skype/similar meeting for anyone who is interested - we could discuss:

  1. What are the strategic/political decisions that are affecting this?
  2. How this fits in/alters current Secretariat/DI workplans - and what scope there is for tweaking them?
  3. What is a reasonable number to aim for in terms of the abandoned data - or whether we should just drop this data?
  4. What the community can contribute to making this work - and how best to do that in a coordinated way?
  5. Whether the different technical options really work?

So - if keen, please let me know on this thread, and I will setup a doodle to find a time that works for everyone - or maybe this process is better managed by DI/UNDP - please say if so.

(IATI Technical Team) #10

Hi Matt,

Thanks for your thoughts; our response is below. We’re happy to set up a call if that would be helpful.

A few specific responses to your points:

The way the system has been built is that the Validator checks if a file is 2.0x compliant; if not then it doesn’t parse it onto the DS. This means there is no way to “turn it off”.

We would have to turn off all validation to allow v1 data into the DS or we’d have to delay the launch of the validator and the DS in order to build a v1 validation service, which would add considerable cost.

There is an agreement to keep the existing DS in place until the end of 2019.

We discussed with our board focal points the problems of having data that doesn’t have an owner; this gets into the core of how IATI is designed.

The way the whole IATI data system functions is that a publisher owns and hosts their own data. If we (as the tech team) start to transform and host people’s data this is a fundamental change to how IATI publishing works - a big change of mandate for the IATI secretariat and something that was agreed with IATI Board focal points wasn’t the right direction for us to take.

One example of where it would get complicated is when a publisher chooses to update their data. We would then be left with two versions in the system: the data transformed and held under a fake account as well as the data owned by the publisher that they have later handled.

The IATI-transform tool will work in some cases but not in all. Not all mandatory elements and attributes in version 2 of the IATI Standard are required in version 1. The IATI-transform tool can turn existing version 1 data into version 2, but it cannot create data.

For example, it cannot provide an activity status, or an iso-date for an activity where none was originally given. That information needs to be added by the publisher.

Hopefully the more detailed explanation above makes it a bit clearer as to why the IATI secretariat can’t re-publish data that we don’t own and how there isn’t something we can “turn off”.

In terms of what we are doing behind the scenes; we have been in touch with all the publishers that this pertains to offering our assistance for them to upgrade.

From our analysis only 61 publishers in v1 are still active. We have reached out to all of these organisations and have been pleased with the level of engagement so far. The vast majority of these publishers only have their Organisation file in v1 and/or use AidStream to publish their data, meaning that the transition process is not particularly strenuous.

Getting the organisations to engage is the biggest barrier! We would really welcome your thoughts and ideas on how we can get better engagement.

Also, it’s important to note that version 1 data will also still be available from the Registry.

As above, we’re happy to have a call with you and any other community members if you think it would be helpful.

2 Likes
(Andy Lulham) #11

Agreed that neither of the approaches described here are desirable. One alternative approach would be: if the dataset is v1.0x, attempt to transform it (i.e. pass the dataset through an XSLT.) Then continue as before i.e. pass the transformed version through the validator.

(Mark Brough) #12

Thanks for this great response, @IATI-techteam!

Personally I am pretty happy with this change. The analysis indicates that all major (“targeted”) organisations will hopefully be migrating to v2 so we will not miss their data. There are two exceptions: Scottish Government and JICA. I think it is OK that JICA’s data won’t be migrated as my understanding is that it is CRS data anyway – so it would be preferable to move them to a better publication process instead. That leaves only the Scottish Govt.

On @matmaxgeds’ point, I think the number of activities overstates a bit the issue here – thanks to @andylolz’ IATI Kit, I did a quick analysis based on the great analysis spreadsheet @IATI-techteam shared at the top of this thread – see here for details (here for code), but in summary – looking only at activities:

  • 84% of activities are from targeted publishers
  • of the 62,056 activities from targeted publishers, all but 2,886 (for Japan and Scotland) should become available in v2.
  • there are only 12 untargeted V1 publishers with over 100 activities
  • I guess there could be other effects in terms of enabling traceability if the data from some of these smaller organisations is no longer so easily accessible, but I think that is OK if the data is normally quite stale.

Maybe I have got something wrong in my analysis, but I think this indicates that the vast majority of activities, and almost all from targeted organisations, should be covered by organisations that will convert soon.

So maybe need to keep an eye on the remaining targeted publishers, but in the meantime, I think this looks good @IATI-techteam!


(As an aside: I think it might be good to have a conversation about what we do in the future in similar cases, where there are a large number of organisations with stale data.)

3 Likes
(Steven Flower) #13

Thanks @IATI-techteam for the thorough analysis and explanations. Thanks @matmaxgeds @andylolz & @markbrough for your support analysis, commentary and pivot tables :slight_smile:

We should remind ourselves of how productive it is to drill into the issue we first started to discuss in January, whilst also minded of the need for this key infrastructure to be delivered and available to us as planned.

Thanks to this latest discussion, I think we now have further evidence to advise the @IATI-techteam that this decision is productive. As @markbrough also highlights, there are further issues to discuss and deliberate on - but let’s make sure we consider those in away from this decision.