sktime "fork" - what is happening, rumour/misinformation debunk

[fkiraly]

sktime "fork" - what is happening, rumour/misinformation debunk

Explainer for the sktime community fork.

Various narratives are circulating, and with it a number of rumours (or fake news).

This is an attempt to summarize the situation and provide some fact-based debunk.

what is happening

fork = attempt by a few actors in one academic network to take control of sktime assets, while keeping appearances of an "open" community.

Relevant key points of the story below.

Recommended way to read: first without following links, until appendix; then re-read and check links for matching facts.

Background:

• sktime was on path to success in 2022!
• on track with NumFOCUS to become fully sponsored (i.e., a charity) in December 2022
• awarded ca 640.000 dollar worth of taxpayer funds in April 2022 (UKRI grant)

Situation:

• Tony Bagnall (tonybagnall) - former treasurer - removed the funds from strategic control of the sktime community, using "grey patterns" (i.e., taking control without necessarily breaking laws), see point 4 and 5 of approved CC meeting notes of May 31, 2022
• proceeds with spending, e.g., hire associates on positions funded by the grant, such as former PhD student Matthew Middlehurst on a senior PDRA position
• inside sktime leadership, this leads to disagreement, and patterns of harassment, bullying, and mob(bing) tactics by forkers, e.g., mobbing off uninvolved contributors such as in this public PR; more links below
• only days before sktime incorporates as charity via NumFOCUS (Dec 2022), Martin Walter (aiwalter), one of the associates, claims "governance breakdown". the escalating conflict, pushed by the forkers, leads to NumFOCUS dropping sktime
• as harassment/bullying reaches extreme levels, Tony Bagnall and perpetrators are banned in mid-Dec 2022
• forking group subsequently makes attempts to take over digital assets of sktime, see issue 4203
• as attempts to take over full release pipeline fail, strategy changes to: (a) destroying sktime assets, e.g., sktime brand, webdomain, documentation, etc; and (b) trying to convince/confuse onlookers that fork is the "legitimate branch", this is currently underway

key parties involved

original project:

• the "open", distributed, international contributor community
• community leaders in favour of the open model and transparent governance
• groups of users dependent on sktime

the fork:

• Tony Bagnall and associates, around University of East Anglia (UEA) and the taxpayer money pot they control

Rumour/misinformation fact checker

Summary of common pieces of misinformation circulating. Detail fact check in the next section.

Misinformation	Reality
1. NumFOCUS supports the fork! See this letter of NumFOCUS !	The letter only refers to escalating conflict which is correct (see above). If NumFOCUS were supporting either side, they would have said it. Rather, it's a "solve your problems yourselves" message.
2. The fork has a diverse cast of developers!	Developers in the fork are almost all tied to UEA, a UK university (and/or the taxpayer funded money pot under their control).
3. The fork has all the important developers!	The fork has none of the release managers or architects. The majority of the developers in the fork did not make substantial code or non-code contributions over the last year; some mostly contributed on political GitHub issues or trolling. Patterns of actual contributions by forkers are mostly single-issue (e.g., favourite algos).
4. The fork wants to create a better open source community!	Inconsistent with past behaviour: mobbing off contributors; politicizing technical issues; vehement resistance to financial transparency and accountability - see examples in appendix

Appendix: fact check details

1. NumFOCUS stance

Reading the actual letter carefully should debunk the claim.

2. fork developers

Fact: almost all linked to UEA, directly or via grant money pot

UEA grant link - academics

• tonybagnall - academic at UEA, Norwich, UK
• matthewmiddlehurst - hired by tonybagnall on the grant; before was PhD student at UEA, Norwich, UK
• chrisholder - PhD student of tonybagnall; UEA, Norwich, UK
• patrickzib - close academic associate of tonybagnall; Berlin, Germany

UEA grant link - commercial

• ltsaprounis - formerly grant liaison at GSK (GlaxoSmithKline), no longer works at GSK; UK based
• aiwalter - formerly grant liaison at Mercedes-Benz, no longer works at Mercedes-Benz; German expat, now resident of Tanzania

compare the above with the public grant page, it lists:

• UEA, Tony Bagnall
• Mercedes-Benz (compare aiwalter's linkedin)
• GSK (compare ltsaprounis' linkedin page)

other, or indirect link

• guzalbulatova - now works at same company as ltsaprounis, Russian expat (Moscow/St Petersburg area), moved to London, UK in mid 2022
• lmmentel - commercial data scientist; Oslo, Norway

3. Developer capacity

Summary of contributions in last year (as of Mar 2022 - Feb 2023)

"low code contribution capacity" can be gleaned from public GitHub profiles.

The majority (not all) of forkers have an "inactive" profile, or participate mainly in "political drama" mode. The rest have a single- or few-issue profile.

GitHub names: tonybagnall, patrickzib, guzalbulatova, lmmentel, aiwalter, ltsaprounis, chrisholder, matthewmiddlehurst

(of course, there is increased activity right now for appearances sake)

Lack of non-code contributions can be gleaned from:

• meeting notes/agendas, for admin
• general de-facto policy of non-participation in, and not helping out with events for the wider community (e.g., developer sprints)
• help-desk channel on discord or slack, non-engagement by forkers

4. the forkers' history of non-collaborative behaviour and bad governance

mobbing off contributors - hcrystalball

Context on mobbing off contributors:

• hcrystalball is/was a framework package for forecasting run by Michal Chromcak and team. We wanted to collaborate and integrate (following sktime's vision of integrated open source ecosystem)
• aiwalter, out of the blue, opens the PR above to remove an existing framework adapter
• pushes for a vote and also ignores the voting rules (compare governance rules, see link in issue)
• the forkers pile on, even though before and after most of them have never been interested in forecasting
• results in removal of estimator, discouragement of Michal to contribute, his self-removal from community

in hindsight, this could be seen as a form of covert bullying and pushing off productive contributors

politicizing technical issues - numba

Context on politicizing technical issues

• architecturally, sktime was following sklearn with dependency minimization as an aim (see issue for tech context)
• when the discussion moves to numba and enabling python 3.11 support, this becomes highly politicized
• the forkers engages vehemently in preventing numba to move to soft dependency status
• arguments in the issue only make partial sense technically
• but it causes a lot of political heat in internal discussion channels
• aiwalter again tries to ignore the voting rules
• the issue eventually sees a mob vote by the forking group

This came only a few weeks before the NumFOCUS charity incorporation deadline (early Dec 2022), for "sponsored status".

Despite the heat, the governance rules were respected in the end - numba did not become soft dep, python 3.11 support was delayed, despite python 3.11 being one of the most popualr user requests at the time.

The forkers could not make a claim that sktime governance rules were disrespected, while the NumFOCUS deadline was only days away.

The forkers then proceeded to escalate further by attacking governance directly, via "governance breakdown" claims, with coordinated multiple complaints sent to NumFOCUS leadership.

This is also extremely curious as two of the forkers at the time (aiwalter, guzalbulatova) were on the leadership committee that they were accusing of breaking down.

resistance to financial transparency

Context on resistance to financial transparency and accountability

• this was an effort to sharpen financial governance given influx of donations and the 640.000 USD worth taxpayer/grant award
• content are rules on how to handle resource allocation decisions, e.g., advertising jobs transparently, not handing money to associates, etc
• originally opened as PR 3095
• feedback on 3095 was solicited in multiple CC meetings, see repeated actions in notes/agendas
• after months of being open and only positive feedback, fkiraly merged this
• aiwalter immediately unmerged, with a framing that fkiraly was breaking rules, see PR 3673
• heated discussion ensues, sees mob behaviour and stalling by forkers (see discussion on 3673)

More generally, Tony Bagnall, while treasurer, pushed away from public budget reporting:

• general refusal or evasion to provide a budget despite being treasurer (see notes/agendas from Feb 2022 on)
• attempt to funnel donations to UEA without transparent reporting (in this case 80.000 USD from the oil company Shell), and
• push to abolish the public budget completely, see point 4 of approved CC meeting notes of May 3, 2022

[astrojuanlu]

Thank you for collecting this information @fkiraly

[ViktorKaz]

It is really sad to read all of this. I have not been an active contributor to the project for quite a while but I can definitely testify that sktime is the brainchild of @fkiraly. Back in 2018 he set up the initial team of developers and invited Tony Bagnall to join the project.

In my personal opinion, sktime became a successful project mainly due to the efforts and knowledge of @fkiraly and @mloning.

As I wrote, I have not been an active contributor for a number of years and I cannot know what caused this fallout. However, it does seem quite unfair to me that, based on the post above, Tony Bagnall is essentially trying to steal the idea of @fkiraly and capitalise on it.

In addition to this, sabotaging the NumFOCUS application and stealing the sktime digital assets, if true, does appear to be quite insidious.

I do hope that the sktime project will continue.