Bring a statute. Read it clause by clause.
Upload an Act, a Bill or a Finance Act. PRISM parses its structure, tags every clause, and builds a workspace where each finding links back to the words that support it.
Upload
Uploads are kept on the analysis server. Signed in, they are listed only for you; anonymous uploads and the demo are shared. Model extraction runs on whichever backend the server is configured with, which may be a hosted service.
Documents
DashboardWhat happens to a PDF
- 01
Structure
The statute is parsed into chapters, sections, sub-sections, provisos, explanations and schedules, each with its exact character offsets and page position.
- 02
Clauses and entities
Each clause is tagged for obligations, penalties, rights, thresholds, actors and beneficiaries, and for condition → consequence patterns, by rules you can read.
- 03
Meaning
Clauses are embedded and projected to a map, so similar provisions sit together and can be searched by meaning.
- 04
Rules, read by a model
On request, a language model reads clauses into structured rules. Every amount and rate it states is checked against the clause's own words; one the clause does not contain is flagged for review.
- 05
Provenance
Everything is linked in one graph from document to clause to entity and rule, so any output can be walked back to its text.
- 06
Explanations
Each extraction can be probed with LIME and attention views. These are labelled as commentary: the research found cited spans are no more faithful than random ones.
The research behind PRISM tests whether these readings lead to the same policy conclusions as the law. Read the story or open the results.