Legal Data Utilization Award at the Digital Agency Law x Digital Hackathon

After a disaster, a household in Japan may be eligible for a considerable amount of support. Some of it is national, some prefectural, some municipal. Each program is defined in its own statute or ordinance, with its own eligibility conditions, its own filing window, and its own office.

The information is public. All of it. And it is unusable at exactly the moment it is needed, because assembling it requires a person to read across several bodies of law and work out which conditions their own situation satisfies — while displaced, without documents, and often without power.

The municipal officers on the other side of the counter are in a similar position. They know their own programs well and the adjacent ones less well, and they are being asked hundreds of questions a day by people whose circumstances all differ in the details that matter.

Fukko Compass takes a household’s actual situation and returns the programs that apply to it, in the order they should be acted on. Three parts made that possible.

Legal text is not a list of benefits; it is conditions, actors and consequences woven together. Before anything can be matched, provisions have to be decomposed into tagged components — who is affected, what condition triggers the provision, what follows from it. That tagging is what turns a body of statutes into something a program can reason over. It is ontology work with a different name and much higher stakes.

The patent-pending Dango interface presents a sequence of decisions in a form usable on a phone by someone under acute stress. This mattered as much as the retrieval. A correct list of eligible programs presented as a wall of text is not a usable answer for a person who has just lost their house, and evaluating the system on retrieval accuracy alone would have hidden that.

Connecting to the national identity portal removes the step where a person has to prove circumstances they can no longer document, because the paperwork proving them was in the building that is gone.

This is the same problem as every enterprise retrieval project we take on, in its most consequential form. The knowledge was public. It was accurate. It was maintained by competent people. And it was unusable, because it had never been structured for the question anyone actually had.

Nothing about the model mattered here. What mattered was decomposing provisions into tagged conditions, defining the relationships between them, and building an interface honest about what it did and did not know. That is the ordinary work, and the award was for doing the ordinary work on a problem where it counted.

The method is written up in general form in Ontology design in practice, and offered as a service on the ontology page.