# Swiss Apartments — research snapshot, 18 October 2026

Companion to Reading Swiss Apartments: From Floor Plans to Design Parameters.
Source: Swiss Dwellings v3.0.0, Standfest et al., Archilyse AG.
https://doi.org/10.5281/zenodo.7788422 — CC BY 4.0.
See ATTRIBUTION.json for all dataset creators and modification notices.

## Quick reproduction from this snapshot

Extract the ZIP and open a terminal in this directory. With Python 3.10 or newer:

    python check_snapshot.py

No third-party packages or original CSVs are needed for this check. It verifies
all included file SHA-256 values, recalculates conditional area quantiles and
median corridor shares from cohort.csv (42,242 keys), and recounts connectivity
at all three tolerances from topology_sample.csv (400 cases). It also checks
composite join totals recorded in JSON. It does not re-extract the source census.
Open explorer.html directly in a browser for the existing interactive findings.

## Full source reproduction (separate, heavier workflow)

Download Swiss Dwellings v3.0.0 from the DOI above. Extract geometries.csv,
simulations.csv, locations.csv and location_ratings.csv beside the four scripts.
The original CSVs total approximately 2.69 GB; allow additional memory and disk
space for wide CSV parsing, Parquet caches and derived outputs. No timing or
minimum-RAM benchmark is claimed. All new outputs go under research/.

Original study runtime: Python 3.13.9, pandas 2.3.3, NumPy 2.2.6,
PyArrow 21.0.0, Shapely 2.2.0, plus Matplotlib. In an isolated environment:

    python -m pip install pandas==2.3.3 numpy==2.2.6 pyarrow==21.0.0 shapely==2.2.0 matplotlib
    python analyze_dwellings.py --self-check
    python analyze_dwellings.py
    python research_relationships.py
    python research_layouts.py
    python build_research_report.py

zip-tail.bin is the saved publisher archive central-directory fragment used for
member size/CRC comparisons, not a complete archive. Full scripts and original
source hashes are retained. Only the report writer's internal handoff and prior
UI-verification footer are removed; analytical scripts are otherwise unchanged.
The original machine used a local _deps Shapely installation. The scripts import
Shapely normally, so readers can install it in their environment; that clean
environment installation and full workflow were not repeated for this article.
Do not copy the kit's compact snapshot into research/ as if it were full caches.
The scripts regenerate single_floor_reference.csv and other large outputs.
Fresh extraction should start without research/ caches. When changing source
data or extraction logic, use a fresh directory; see METHODOLOGY.md for caveats.

## Contents and limits

FINDINGS.md, METHODOLOGY.md and DESIGN_GUIDE.md contain the dated research
findings, definitions and proposed design pipeline. cohort.csv preserves just
five columns for portable statistical checks; it is not a complete generator
program library. JSON summaries and six exact WKT examples retain source keys.
example_apartments.local.geojson uses engineering metres, not WGS84 lon/lat.
The example PNG is a rendering of selected source layouts, not generated plans.
Some research notes name large Parquet/CSV audit artifacts excluded here; those
are outputs of full source reproduction. README references to the original
study folder in those notes refer to that full workflow.

No generator, CAD export, accessibility, construction feasibility or national
representativeness was validated. Full census claims use the original dated
study; article preparation reruns compact checks, not the full extraction.
See VERIFICATION.md for what was actually checked for this companion.
