Feature requests

[chrisbrownlie]

Prework

• Read and abide by pointblank's code of conduct and contributing guidelines.
• Search for duplicates among the existing issues (both open and closed).

Proposal

Thanks for the great package! I've been getting to grips with it recently and had some things I wanted to do that don't currently appear to be possible - or at least are not straightforward to do. I've also checked the issues list for all the below and couldn't see anything. For info I'm pretty much solely using the VALID-I workflow as described in the docs.

For ease I've compiled them here in one issue and would greatly appreciate if you could:
(a) let me know if there is already an easy way to do this with {pointblank} that I'm not aware of,
(b) if not, would you consider adding it as a feature, and
(c) if yes to (b), would you welcome a pull request for it, as I'm happy to have a go at doing so

Let me know if you'd prefer me to open separate issues for anything! :)

Validation functions

• An opposite/complement to conjointly(): i.e. 'if any of these individual checks pass, the overall check passes'. I figure its possible with preconditions/specially() but I would find a separately()/alternatively()/or() type function useful
• Control over data extracts. One use case is to add a data extract to specially(), but in general it would be good to be able to specify 'this is how to represent failing checks' via an extract_fn argument (or similiar) to validation functions. For another example, I am trying to validate some very wide datasets and it would be useful to only include relevant columns in data extracts so that I can easily see why/where the check failed.
• Custom metadata for each validation step, i.e. a metadata list argument or similiar to all validation functions which would be added somehow to the agent - either to the validation_set tibble or as a separate element. My use case is to add a custom 'type' to each check, to group them when doing custom reports/investigations.

Reports

• keep and arrange_by arguments to get_multiagent_report(), behaving the same as the arguments for get_agent_report()
• Control over display of checks in agent reports. I'd quite like to group some checks together - particularly when its the same check as in the example below - or group passing checks together. For example, below it would be nice if all columns that passed could be combined into a single row in the report - for some wide datasets I want to check >100 columns are all numeric but am only interested in those that fail when I look at the agent report:

library(pointblank)
agent <- iris |>
  create_agent() |>
  col_is_numeric(starts_with("S")) |>
  interrogate()

^{Created on 2025-02-05 with reprex v2.1.0}

Other

• Some functionality for using a database to track validations. My use case is that we have validations run on survey data - lets say we get tens of responses each week and at the end of each week 'interrogate' the responses we've received. We have easy access to a SQL server database and ideally I'd like to run the {pointblank} interrogations in a CI pipeline which outputs an email of all the failing checks, so my dream is to have a way to: save an agent specification to a table or row in a table; and/or save the results of interrogations as row(s) in a database table (to track changes over time). This is maybe less of a feature request as it might be difficult/beyond the scope of the package but I'd very much welcome any thoughts on how you might go about this!