Document the OCR demo dataset and PDF Bucket

README.md

@@ -19,16 +19,65 @@ configs:
 
 # OCR demo: Food for Space Flight
 
-Seven scanned page images (PDF pages 3-9) from NASA's *NASA Facts: Food for Space Flight*.
-Load with `load_dataset("uv-scripts/ocr-demo", split="train")`.
-The original PDF and matching extract are in the [OCR demo Bucket](https://huggingface.co/buckets/uv-scripts/ocr-demo).
+Seven scanned pages from NASA's *Food for Space Flight* booklet, with headings,
+columns, photographs, food lists and tables. This small dataset is an input for
+trying OCR recipes and inspecting their results.
+
+The images are PDF pages **3-9** (printed pages **2-8**) of the original booklet.
+The selection omits the reproduction disclaimer and dark cover. The complete
+original PDF and a matching seven-page extract are in the
+[OCR demo Bucket](https://huggingface.co/buckets/uv-scripts/ocr-demo).
+
+## Use as a dataset
+
+```python
+from datasets import load_dataset
+
+pages = load_dataset("uv-scripts/ocr-demo", split="train")
+```
+
+There are seven rows in the `train` split:
+
+| Column | Meaning |
+|---|---|
+| `image` | Page image, rendered at 300 DPI and saved as JPEG at quality 85. |
+| `page_id` | Stable ID containing the original PDF page number. |
+| `source_page` | One-based page number in the original PDF. |
+| `printed_page` | Page number printed on the scanned page. |
+| `source` | NASA's source-record URL. |
+
+The images use the [ImageFolder format](https://huggingface.co/docs/datasets/en/image_dataset).
+`pages/metadata.jsonl` associates each image with its source-page information.
+There is no verified ground-truth transcription in this dataset.
+
+## PDF and file inputs
+
+The [OCR demo Bucket](https://huggingface.co/buckets/uv-scripts/ocr-demo) contains:
+
+- `original/food-for-space-flight.pdf`: the unmodified nine-page NASA PDF.
+- `demo/food-for-space-flight.pdf`: a seven-page extract matching the dataset rows.
+- `pages/`: the same seven page images and metadata.
+
+Use one prefix, such as `hf://buckets/uv-scripts/ocr-demo/demo`, as your input to
+avoid processing both a PDF and its page images.
 
 ## Source and licence
 
-Source: [NASA NTRS record 19830005543](https://ntrs.nasa.gov/citations/19830005543).
-[Original PDF](https://ntrs.nasa.gov/api/citations/19830005543/downloads/19830005543.pdf).
+- [NASA record: NASA Facts: Food for Space Flight](https://ntrs.nasa.gov/citations/19830005543)
+- [Original PDF](https://ntrs.nasa.gov/api/citations/19830005543/downloads/19830005543.pdf)
+- [Document metadata](https://ntrs.nasa.gov/api/citations/19830005543)
 
 NASA's document record states **Work of the US Gov. Public Use Permitted**.
-See [NASA's usage guidance](https://www.nasa.gov/nasa-brand-center/images-and-media/).
-The pages retain their source rights; the recipe repository's code licence does not relicense them.
-NASA does not endorse this demo. No verified ground-truth transcription is included.
+The source pages retain that status; the repository's code license does not
+relicense the NASA document. See [NASA's usage guidance](https://www.nasa.gov/nasa-brand-center/images-and-media/).
+Attribute these source pages to NASA and identify any generated transcription as
+model output. NASA does not endorse this demo or verify its OCR results.
+
+The archival scan has damaged lettering and an imperfect existing OCR layer.
+Inspect generated text against the page images, especially tables and captions.
+
+## Preparation
+
+Pages 3-9 were rendered from the original PDF with Poppler's `pdftoppm` at
+300 DPI and saved as JPEG at quality 85. No OCR model was run during preparation.
+The Bucket's `manifest.json` records source URLs, file checksums and rendering settings.