How We Built the Forms Dataset, and What Measuring It Taught Us

A matter plan tells you which document a task needs. The forms dataset is the document itself, ready to fill. Building it meant collecting forms from courts, tribunals, state agencies and councils, then doing the slow part: deciding, field by field, what each blank on each form actually asks for.

Collected From the Source

Every form in the library comes from the authority that publishes it. We republish an authority's own form rather than a copy someone else has altered, and we record where it came from. The library now holds more than 700 templates: fillable PDF forms published by courts and councils, and Word precedents for the documents that have no official form.

Every Field Mapped to the Standard

Each template carries a manifest: a list of its fields, and for each one, which defined term in our field standard fills it. That is what lets a form be completed from a matter's data instead of retyped. It is also where quality is easy to fake, so we measure it.

In August we audited the whole library and did not like everything we found. Every field reference resolved to a real term, so a simple health check came back green. But the mappings themselves were thin: too many forms leaned on a handful of generic terms, and some litigation forms had inherited terms from conveyancing. A green check is not the same as a good mapping. We built a snapshot and rollback first, so no change could be made that could not be undone, and then rebuilt the mappings under written rules. By October the library carried over 7,000 bound fields, every date field had an explicit format, and every stored file matched its recorded hash.

Watching for Changes Upstream

Authorities revise their forms without telling anyone. A monitoring job fetches each source form on a schedule, hashes it, reads its field structure, and classifies what it finds as unchanged, content altered, structure altered, or failed to fetch. A changed form is flagged for review before the copy we serve goes stale, and a form an authority withdraws is retired from the library rather than left on offer.