# Site Layers-open

**Address in. Layered site drawing, four diagrams and a verification register out. Open data only — nothing to buy, no keys.**
AI for Architects · Assignment 3 · v1.0 · 23 Sep 2026

## What you get

| file | what it is |
|---|---|
| `site_<address>.dxf` | Layered drawing for Rhino or Revit. Metres, your address at 0,0, true north up. `CTX_` layers are the physical site; `AI_` layers are planning and climate. See `LAYER_SCHEME-open.md`. |
| `diagram_exploded` | The hero image: every layer on its own plate, stacked. SVG (editable) and PNG. |
| `diagram_plan` | Site and statutory plan: zoning, heritage, flood and bush fire over the physical base. |
| `diagram_statutory` | Eight small plans side by side: zoning, height, FSR, heritage, flood, bush fire, acid sulfate soils, context. |
| `diagram_climate` | Sun path, wind roses (annual and by season), monthly temperature and rain. |
| `register_<address>.csv` | **The verification register.** One row per fact the tool found, with its source, a link to check it, and an empty **verdict** column. |
| `layers_<address>/` | Every layer as GeoJSON, for QGIS or anything else. |
| `building_heights_<address>.csv` | Every building, its height, and where that height came from. |
| `prompt_suffix.txt` | Your style card as words, to add to any image-generation prompt. |

## Run it — pick one

**A · Google Colab (easiest).** Upload `Site_Layers-open.ipynb` at colab.research.google.com (File › Upload notebook). Run the cells top to bottom. Edit the style card cell, type your address, download the zip. Needs a Google account.

**B · Your own computer (if you can't use Google).** Install Python 3.10 or newer from python.org, then in a terminal, in this folder:
```
pip install ezdxf shapely matplotlib numpy requests
python site_layers-open.py "249 Oxford Street, Paddington NSW 2021"
```
Options: `--radius 500` · `--style my_card.toml` · `--out results` · `"-33.8847, 151.2268"` instead of an address.

**Not in the chat.** The code sandboxes inside ChatGPT, Claude and Gemini usually have limited or no internet access, so they may not reach these data services. Run the tool in Colab or on your computer, then bring the outputs to your AI with `INSTRUCTIONS-open.md`.

## Before your first run
Fill in `style_card-open.toml` — palette, line weights, hatches, typeface, three reference images. It decides how every diagram looks. The tool never picks the look.

## After the run
1. **Check the pin.** First row of the register. Is the lot outlined in red really your site?
2. **Fill the verdict column** for every row: *confirmed*, *wrong* or *not checked*, and say where you checked in *note*. The *how_to_check* column says where to look. Unfilled rows are not a failure of the tool; they are the assignment.
3. Open the DXF, compose your one or two diagrams, carry the geometry into Component 3.

## When something fails
The run always finishes. A layer that could not be fetched becomes a register row saying why, and the drawing goes out without it. Servers get busy — run again later and only the missing parts are fetched (everything that worked is cached in `_cache/`).

## What it can and can't tell you
- **Everywhere:** buildings, roads, paths, rail, water, trees and land use from OpenStreetMap (as good as local volunteers made it — the register says when coverage looks thin); contours; sun path; ten years of wind, temperature and rain.
- **NSW only (so far):** the lot boundary, and zoning, height, FSR, additional-control areas, heritage, State Heritage Register, flood planning, bush fire prone land and acid sulfate soils from the NSW Planning Portal.
- **Other jurisdictions:** those layers are marked *not available for this jurisdiction*. Find them in your local planning viewer and add them by hand.
- **Never:** title, easements, the DCP, or a planning certificate. It is not planning advice and it is not a survey.

## Data and credits
© OpenStreetMap contributors (ODbL) · NSW Spatial Services and NSW Department of Planning (CC BY 4.0) · Open-Meteo.com (CC BY 4.0, ERA5 reanalysis; free for non-commercial use) · Terrain Tiles on AWS (Mapzen/Tilezen: SRTM, LINZ and others). Keep the credit line on anything you publish.