TGIS v0

Up next

Two tracks run at the same time. One widens what TGIS can find and describe. The other puts v0 in front of a small group of users. Both lead to a private beta.

Now

v0: 100 sources, 58 of them global; 30 concepts; discovery seeded from Addis Ababa.

Build

Wider discovery, more concepts. Find more of what exists, and describe more of what a planner asks about.

Test, in parallel

Short sessions with a small group on the concept and how it works. What they find steers the build.

Then

A private beta.

Build: wider discovery and more concepts

The registry answers well for what it already holds. The next work is about reaching what it does not.

Find more sources at question time

Today: Open data portals and satellite catalogues are searched live. Statistics agencies (OECD, Eurostat) cannot be, so 30 tables were listed in advance and wait for a question.
Next: Search more catalogues live, keep listed tables to the country they cover, and add a helper that reads long result lists and hands back a shortlist.

Grow the list of concepts

Today: 30 concepts, written and reviewed by hand. A run may not add one.
Next: Import the concept lists statistics agencies already publish, defined and linked, so a new sector arrives ready to search.

Rank sources better

Today: Source descriptions often restate the place and the subject, which the search already handles.
Next: Rewrite them to say what tells sources apart: vintage, method, resolution. Test giving concept matches a boost; in a test set it lifted the right source into the top three from 56% to 81% of questions.

Read documents

Today: Transport master plans, corridor designs and zoning rules are gaps, because nothing reads a PDF.
Next: Treat a document as one more kind of layer, so a plan that already commits a corridor can sit beside the population and the stops.

Test: short sessions with a small group

A few people at a time, on v0 as it stands and on these pages. Each session is short.

What the sessions are for:

Is a gap with a named holder useful? Or does “here is who has it” read as a failure?
Do the labels change trust? Official statistic against modelled estimate, Provisional, coverage under 100%.
Can people follow the trace? Do they open it, and does it explain the answer?
Do two disagreeing figures help? Or would they rather be handed one?

How it feeds the build

Each session runs on the live system, so its runs are recorded like any other. A question that ends in a gap points at the next source or concept to add.

Then: a private beta

The beta’s scope comes out of the two tracks: what the sessions show people need, and how far discovery reaches by then.

The engineering plan behind the build track, and the rules every change has to pass, are in the evolution plan.