The Office Keeps Its Judgement. The Machine Keeps the Record.
Terms are defined, authority is hierarchical, and a document either says something or it does not. What breaks ordinary tools is that being approximately right is worthless here.
Judgments Translated to Filing Standard
Courts in India work across languages, and a translation that reads well but renders a term of art loosely is not usable in a filing. So the terminology is not left to the translator's discretion. A controlled glossary of legal terms is mandatory input, and compliance with it is checked rather than hoped for.
Scanned pages become text, with the structure of a judgment respected: cause title, parties, headnote, body, footnotes, and the running headers that must not be translated as if they were prose.
The whole document is read before a word is translated, and a written strategy is fixed: how this judgment handles its terms, its register, its citations. Translating passage by passage without that produces a document that drifts.
In passages, each carrying the strategy and what came before, so terminology stays identical from first page to last. The glossary is binding.
Every passage is scored against the source and the glossary. Anything below the bar is translated again rather than shipped with a caveat.
The glossary is the ontology. Several hundred terms, one binding rendering each. Two documents translated months apart then come out consistent, which is the difference between a convenience and something you can file.
- A term of art in the source judgment
- The rendering a translator might choose
- A second, equally reasonable rendering
- What a different translator used last year
Checked on the way out, not left to preference. Extend it and every future document follows the new entry.
This is the same alias problem an archive has with names, and it has the same answer. Left alone, one concept fragments into four, and a body of translated judgments stops being searchable as one thing.
What comes out is a formatted document, not a text file. It reproduces the conventions of a legal document in the target language:
- Party tables set out as a registry expects them
- Footnotes numbered correctly
- Headers carrying page numbers
- The fonts the registry requires
A translation a clerk has to reformat has not saved the office anything.
Case Files, Structured
A matter is a pile of documents and one lawyer's understanding of it. The ontology writes that understanding down:
- The parties, and the role each plays
- The issues actually in dispute
- Each side's position on each issue
- The authorities relied on for each position
- Every fact asserted, with the document that evidences it
- The procedural history, with its dates
Parties and the role each plays. Forum, and the procedural history with its dates.
What is actually in dispute, stated one issue at a time rather than as a narrative.
Our position and theirs on that issue, with the authorities relied on for each, and the facts asserted with the document that evidences them.
Three levels, and the bottom one holds the work. Positions are kept per issue rather than per document, so two authorities answering the same point sit together instead of being buried in separate files.
A junior joining the matter then reads a structure rather than a shelf. Ask what is our answer on limitation, and what supports it, and the answer opens the document behind it.
Contradiction Across a Body of Law
The analysis described for legislative drafting is useful to a practice in a different way. It finds:
- Conflicting authority on the same point
- A line of cases that has quietly split between benches
- A definition changed by statute, under precedent built on the old one
- An instrument still being cited after it was amended
Each one is shown with both texts side by side. The system never picks a winner, because that is the practice's work.
How It Sits in the Office
- Access is by invitation, per person
- Work is metered, so cost is visible per matter
- Every action is logged and auditable
- Deletion of stored material is deliberate and restricted
- The glossary is yours and you extend it
- Formatting matches the registries you file in
- The case ontology follows your areas of work
- It can run entirely on infrastructure you control
Privilege is the constraint that shapes the architecture. Client material is not a corpus to be pooled. Each practice gets its own instance, its own glossary and its own store, and nothing crosses between them.