← All work

Maps & public data

Deterministic public-records computation

Millions of public records fused into one number — validated to the penny.

The system

A build-time pipeline reconstructs a real-world financial figure that is never published anywhere, by fusing fragmented public records with incompatible formats, vintages, and update cycles.

There is no model in the serving path. Every number on screen is deterministic, auditable math — and the pipeline proves it against reality before anything ships.

The hard parts

Penny-exact reconciliation

Computed outputs are validated against real source documents; the anchor case reconciles to the penny across three separate document types.

QA as build gates

Nineteen build, audit, and validation scripts. Coverage, naming, and outlier checks exit non-zero and fail the build — bad data cannot ship.

Scale without a database

The entire corpus ships as 34 MB of static, versioned data behind a map that resolves millions of entities with level-of-detail loading.

In numbers

7.36M
entities computed
79
regional datasets
19
audit + validation scripts
0
AI calls at runtime

Next.js · MapLibre GL · Python pipelines · static corpus

Happy to walk through this one properly — what it does, how it's put together, and what it would take to build something like it for you.

Vague is fine.

What kind of help? optional — pick any