# Filling a missing building and checking the palace outlines

September 7, 2026. This work returns to the fair’s larger building layout. The new shapes are still plain exterior studies.

## A fourth building behind Fine Arts

Our first model included the main museum and two temporary wings, but left out the sculpture annex between them. The [official guide, printed page 64](https://archive.org/details/ajq5466.0001.001.umich.edu), describes four pavilions and explains where the sculpture building stood. Both the [general ground plan](https://www.loc.gov/item/2007633932/) and [Sanborn sheet 18](https://data.library.wustl.edu/record/108214?ln=en) show the detached annex.

It now has its first footprint, with rounded ends, two lower side ranges and a taller central hall. Sanborn marks the side ranges at 26 feet and the center at 60 feet. Select the annex to see a small detail of that plan beside the model. It is a crop of an existing source, not another independent confirmation; the full-resolution scan remains in the research archive.

The fit is still approximate. At our current map scale, the modeled footprint is about 8% larger than Sanborn’s stated area. Roof profiles, entrances and materials need more evidence, and the other three Fine Arts buildings still have rough placeholder architecture. A photograph of the annex would be especially useful.

## Education’s court was enclosed

The [official guide, printed page 63](https://archive.org/details/ajq5466.0001.001.umich.edu), explains that Education’s planned central court was later enclosed for exhibits. [Sanborn sheet 13](https://data.library.wustl.edu/record/108214?ln=en) agrees: its center is roofed, with a lower 22-foot height inside the surrounding 50-foot hall. Corner and entrance pavilions rise higher. Education and Social Economy shared the building; there was no separate Social Economy palace.

This is why the dates and purposes of our sources matter. An early plan can describe a proposal clearly and still be wrong for the fair visitors experienced. Our old generic courtyard and approximate outline both needed correction. The revised source fit also exposed errors in the nearby rough waterways and waterside path, which we checked together. Three transverse bridges now cross the local channel, replacing one old deck that ran along the water. The approaches meet the banks, and their shared paving is drawn once at junctions. Bridge widths, profiles and the hidden water beneath them are still estimates; the rest of the basin has not received the same close review.

A [Library of Congress photograph](https://www.loc.gov/item/2013649652/) now appears beside Education’s model. It shows the long facade beyond the Main Cascade, but cannot settle the central roof arrangement because that roof is hidden from view.

## Mines: a hall behind a covered promenade

The [official guide, printed pages 83–85](https://archive.org/details/ajq5466.0001.001.umich.edu), describes walls set back behind an outer screen that provided covered promenades. [Sanborn sheet 14](https://data.library.wustl.edu/record/108214?ln=en) distinguishes the solid hall from that perimeter and labels the main building 66 feet high. Four small tower footprints have explicit 150-foot labels. Those do not establish the heights of every other entrance or roof feature.

The [Library of Congress stereograph](https://www.loc.gov/item/2013649990/) shows a colonnaded facade and tall paired obelisks beside the lagoon. It helps explain the arrangement, but it is not a measured elevation. The first model now separates the 66-foot hall and those four tall tower masses. Their flat tops are placeholders: the photograph shows how much the colonnade, obelisk shapes and roof details still need work. The covered promenade and the other entrance pairs remain unfinished.

## What this pass taught us

A proposed Mines fit extended beyond the building on the ground plan. Checking a fresh grid against the native image caught the error before publication. Matching a book’s overall proportions is a useful cross-check; it cannot replace tracing the actual map. A later review also caught an internal roof line mistaken for an outside wall, an over-wide recess and oversized tower footprints. Enlarged source grids helped correct each before integration.

The annex also exposed a small testing mistake: the renderer’s vertex count was zero until a camera rendered the scene. The check now renders before measuring. A separate roof-height check needed to account for the model’s existing thin placeholder roof caps. Neither adjustment makes the historical reconstruction more certain, but both make the software checks more meaningful.

GPT-6 Astra and GPT-5.6 Sol worked on separate source and modeling tasks, followed by integration and independent image review. That division made it possible to investigate more than one palace at a time. The rejected fits and corrected readings are part of the record of using AI, alongside the added buildings. We have not established a capability ceiling for any model.

The aim remains a connected, useful exterior reconstruction with visible evidence and gaps. Interiors are not being built in this pass, and the overall completion estimate remains unchanged.