Notes from 'Annual UX Coordination' (AUX-Coord) 2026-01-07

[trexfeathers]

Between @trexfeathers @hsteptoe @pp-mo @bouweandela @ukmo-ccbunney @ESadek-MO @jrackham-mo @stephenworsley @tkknight @valeriupredoi

What happened last year

• New label - Type: User Experience
• New discussion - Iris User Experience #6594
• A public commitment in the docs - Reviewing the Iris User Experience
• Cube components #6854 - add/remove coords, cell_measures etc with a single method, instead of a specific one for each
• WIP: revamped User Manual - demo
  • Diataxis philosophy
  • Better page discoverability

Hopes for 2026

• Finish revamped User Manual
• Public roadmap/dev-diary
• Start prototyping more disruptive improvements for future major release
  (e.g. better lazy data API, merge/concat refresh, etc. No promises!)
• Over to you!
  • What stories can you share?
  • What's on your wishlist?
  • For inspiration (thanks @ESadek-MO): https://scitools-iris.readthedocs.io/en/latest/further_topics/ux_guide.html#ux-guide

Wisdom of the Crowd

• Getting involved
  • Easy to raise a PR, but hard to get reviewed. Need to 'jump up and down'
    • Recommit to dedicated 'community time' (e.g. Thursday Thrills)?
    • How to apply the Surgery model externally? 'office hours'
    • Monthly community meeting e.g. ESMValTool, Xarray
      • Careful of language e.g. 'developer' excludes a lot of people
      • Ephemeral is important - encourages max engagement
      • Consider timezones. E.g. sometimes early morning sometimes late afternoon
  • Encourage life in GitHub Discussions somehow
  • Opening issues
    • Stale bot
      • Most stale bots are anti-user and anti-contributor, but they don't have to be - Jacob Tomlinson
      • Supposed to be an honest reflection of our capacity
      • How do we make it less of a barrier/less depressing/etc?
        • Longer window
        • Better text?
        • Policy on how to treat the closure warning?
      • How many get re-opened or not?
        • There is a mix of stories
    • Public backlog, public management
    • Summary page "how to get something actioned"
• Discovering Iris
  • Most people discover because their institution chose it (so there are political aspects)
  • We don't know the stories of others
  • Some contributions seem to come from 'Good First Issue' - e.g. university students
• It's important for a repo to show honest signs of life
  • Including staying obviously up-to-date with conventions
  • What does it look like from the outside when we spend a month converting legacy tests?
• Type hinting!
• Velocity
  • Can seem slow
  • Can seem fast
• Bugbears
  • CDO's merge-time is a better experience than merging in Iris
    • Very customisable
    • Faster AND easier
    • 'Correct' metadata handling is probably getting in the way for Iris
  • Constraint API
    • Limiting
    • Fussy
    • Awkward
    • Compare to Xarray's sel and isel
  • CF strictness
    • Files in the real world are a mess
    • NetCDF4 has none of these problems!
    • Xarray has fewer of these problems
    • Simple operations slower than expected
      • Parsing metadata?
      • Chunks?
  • Gallery discoverability
• Docs
  • Different ways of slicing/extraction and why you'd want each
  • How to make a Cube from scratch (or can we solve that with better API?)
• API
  • Cubes from scratch
    • Something like from_dict? E.g. Xarray, Pandas
  • Reshaping a Cube
  • Intuitive naming
    • merge/concatenate
    • Slices and slices_over
• What is Iris getting right (what should we keep)?
  • Quickplot is useful from the command line
  • Masking!!
  • Specificity to weather/climate - is more convenient
  • CF-awareness
  • Flexible chunking control
  • Interoperability (e.g. ncdata, pandas, etc.)
  • Mesh support 💖
• Better ncdata integration for better customisability
• Pip installability (maybe!)

[bouweandela]

Another user experience issue that came to my mind: if you have an iris cube called cube, the code cube.data will load the data, but something like cube.data = cube.core_data() * 2 will not. It might be nice to explain this in https://scitools-iris.readthedocs.io/en/stable/userguide/real_and_lazy_data.html. Typically, you want to tell users to avoid cube.data when working with large data, but assigning to cube.data is fine so that makes the story more complicated.