Ie, I say the wrong outcome, then change my mind.

If you complete all the facets and reload the page without a facet hash in the URL it will show a broken state for "Save facet as"

And the title is broken too.

The "Save facet as" and title of the facet above the PDF viewer should be hidden.

See screenshot. "27th" has not been detected by OCR. Dropdown box offers January 1999 only.

When considering outcome variable, for example, can we add a function that lets the researcher select the relevant text that justifies their conclusion in relation to the facet. Ie, with the "Candy" decision, highlight text in last sentence of final paragraph ("claim for weekly compensation must fail"), and have that added to data for that facet.

May also need some way of recording whether claimant is appellant or respondent, although is this already recorded?

At the moment the height is calculated as if all PDF pages are the same. This means that in cases where they aren't, the overlay positioning is broken.

Once I finish the final variable, there is no obvious transition to the next case. Purely a UI thing unlikely to be an issue for a competent user.

If you select all of the Fixed Columns on the Export page and choose to export all records to CSV it will compile a query that causes the DB server (RDS Postgres t3.medium) to timeout. This DB type should be able to handle reasonable load, so it's likely the query has to be optimised.

The code that compiles the query is here: https://github.com/openlawnz/openlawnz-research/blob/master/index.js#L394

Task:

Amend the queries and the code so that any combination of Fixed Columns, UGC Columns, and Keywords exist doesn't cause the server to crash.

e.g. Auth0 or Cognito.

Add the ability to pass in a mock flag that means fixed JSON responses and PDFs are returned to the client. This should help front-end developers get started without having to set things up.

Improve page layout for smaller screens (but not mobile as it's not supported).

in "Manawatu Knitting Mills" case (decision 2 of 1999), the word 'declined" in the final paragraph isn't detected by the parser in relation to the "was the appeal successful" facet. It's not clear if this is an issue with incomplete keywords or whether it is due to the text not being recognised.

When searching the text manually, the PDF viewer can find the date string. This should be added to the date facet parser.

Make it work

Give researchers the graphiql user interface to perform queries with.

Use @media (prefers-color-scheme: dark) {

Include Postgres and restore from DB backup script.

Polish it up a bit to make it nicer to use.

Gives people an idea of what it is.

On the homepage.

Make the PDF viewer standalone:

• Clear up the code into the control being its own class that can handle itself being reset (when changing cases). Resetting will also reset the "Find in PDF" control
• "Find in PDF" to handle clicking the "x" for clearing the input, and pressing "Esc" closes it
• Add loading state (ideally with PDF download progress bar)
• Fix dragging minimap behaviour so that if you move off the minimap while dragging it won't select elements on the page

Words in para 16 should be highlighted, unless they have been overlooked by us in the list of words.

The page currently has a scrollbar when it doesn't need to.

pdfjs splits the text into segments at random points and currently we are not yet accounting for inter segment matches. That means some results will not surface.

Across the board.

• Naming consistency
• Investigate using CSS variables

Add another layer above so that case sets are tied to projects. This will involve adding another table and foreign key to the case sets.

Projects can have:

• Title (e.g. My Research Project)
• Lead (e.g. William)
• Categories (optional, but can have several)
• (case sets that can be regenerated)

This way we can have multiple projects each with their own case sets.

Description

When a user loads a PDF and selects the facet they wish to search by, the webworker processes the case and returns bounding boxes with highlighted words associated with that facet. If however the user changes cases before this is completed for the previous PDF, the bounding boxes can sometimes be stale from the old PDF as the web worker processing the new case data can return the bounding boxes before the previous web worker has finished, effectively overwriting the new PDF bounding boxes.

Proposed Solution

When a user changes their case selection, terminate the currently running web worker, or interrupt it to re-start processing the new PDF from the beginning, (cancelling work on the old one)

If a case set is regenerated the old URL breaks. It should show a 404 page that looks up the current case set that the file belongs to and provides a link to it.

Current facet types (for human refinement) are limited to

• boolean
• date

Task

Add a third facet type to store any dollar figures awarded in the judgment

Minimap highlighting should search for any '$' in the case text

Occasionally, the court addresses multiple decisions within the same case + judgment. An example of this is Aalderink v Accident Compensation Compensation.

When this occurs, it may be useful to be able to enter multiple dates for the same facet.

Interested to discuss whether this will cause downstream problems for the data.