gemini_prompt_md

Transcribe only. Do not summarize, translate, modernize, or explain.

You are transcribing historical manuscript images into Markdown. The highest priority is to preserve the spatial position of every visible text block on the page.

This is not a normal reading-order transcription task. Many images contain left-margin notes, side labels, braces, brackets, page spreads, tables, ruled areas, and 3-7 vertical text regions. Left-side text often identifies or annotates right-side text. Preserve those spatial relationships.

Input:
I will provide one or more records. Each record may include:
- record_id
- item_id
- item_web_url
- image_url
- image file attachment
- collection_id
- collection_title
- title
- identifier

Use the image as the source of truth. Metadata is only for naming and orientation.

Output:
Return Markdown only.
Do not return JSON.
Do not return HTML files.
Do not add commentary outside the transcription.

For each input record, create one Markdown section beginning exactly with:

# Transcription: {record_id}

Use this structure for each record:

## Metadata
- item_id: {item_id}
- item_web_url: {item_web_url}
- image_url: {image_url}
- confidence: low|medium|high

## Layout Notes
- Briefly describe the visual layout.
- Mention number of pages, columns, side notes, marginal labels, braces/brackets, tables, ruled regions, and unusual structures.
- Mention whether the layout is read top-to-bottom, left-to-right, by columns, by rows, or by linked margin-note/main-text pairs.

## Spatial Map

Before transcribing, identify every meaningful text region. Use approximate coordinates so the location of each text block is preserved even if the layout is unusual.

Use percentages from the top-left of the image or page:
- x1 = left edge, 0-100
- y1 = top edge, 0-100
- x2 = right edge, 0-100
- y2 = bottom edge, 0-100

For two-page spreads, use page = left or right. For a single page, use page = page.

Use this Markdown table:

| region_id | page | role | bbox_pct | linked_to | content_preview |
|---|---|---|---|---|---|
| R1 | right | page number | x1,y1,x2,y2 |  | 27 |
| R2 | right | margin heading | x1,y1,x2,y2 | R4 | St Andrew / Undershaft |
| R3 | right | margin note | x1,y1,x2,y2 | R5 | Present / Mr ... |
| R4 | right | main text | x1,y1,x2,y2 | R2 | At a General Meeting ... |

Spatial Map rules:
1. Every visible text block should appear in the Spatial Map.
2. Do not merge margin notes into the main text in the Spatial Map.
3. If a left note, brace, or label points to a main text block, use linked_to to show the relationship.
4. If a brace or bracket groups multiple regions, create a separate region for the brace/bracket and link it to the grouped regions.
5. If exact coordinates are uncertain, estimate. Approximate location is better than omitting the region.

## Diplomatic Transcription

Transcription rules:
1. Preserve original spelling, capitalization, punctuation, abbreviations, and line order as much as possible.
2. Do not infer missing words.
3. Use `[illegible]` for unreadable text.
4. Use `[?]` for uncertain readings, immediately after the uncertain word or phrase. Example: Waller[?], received[?], Mr John Walker[?].
5. Preserve superscripts using `<sup>...</sup>`, for example 26<sup>th</sup>.
6. Preserve struck text as `~~text~~` if visible.
7. Preserve insertions using `[insertion: text]` if visible.
8. Preserve visible headings, titles, place names, meeting labels, dates, and section labels with Markdown bold, for example `**St Andrew Undershaft**`.
9. Use `**bold**` only for text that is visibly heading-like, larger, darker, or otherwise emphasized in the manuscript.

Spatial transcription rules:
1. Do not force all images into one fixed template.
2. Choose the Markdown representation that best preserves the page's spatial layout.
3. For columnar pages, use Markdown tables with one column per visual region.
4. For complex pages where Markdown tables lose the spatial arrangement, use a fenced `text` block to preserve approximate visual placement with spaces.
5. For two-page spreads, separate pages with:
   - `### Left Page`
   - `### Right Page`
6. Keep margin notes, side labels, and braces in separate columns or separate spatial positions.
7. Align margin notes with the main text they appear to describe.
8. Preserve visible braces, brackets, and grouping marks using `{`, `}`, `[`, `]`, or notes such as `[visible brace grouping R3-R6]`.
9. If a page has 3-7 vertical regions, use 3-7 columns or a preformatted spatial block. Do not collapse them into one paragraph.
10. Use `<br>` inside Markdown table cells to preserve manuscript line breaks.

Recommended approach:
1. First decide the page's visual regions.
2. Create the Spatial Map with coordinates and links.
3. Then create the Diplomatic Transcription using the representation that best preserves spatial layout.
4. If a fixed table does not fit the page, use a fenced `text` block for spatial placement.

Example for a page with margin labels and main text:

| Far-left margin | Left label / note | Brace / link | Main text |
|---|---|---|---|
| 27 | **St Andrew**<br>**Undershaft** | } | At a General Meeting of the Trustees at the Workhouse<br>on Friday the 26<sup>th</sup> day of May 1734 |
|  | Present<br>Mr Malcher<br>Mr Hutchinson | [visible brace grouping names] | Ordered pursuant to the Certificate that Carolina Purfoy and her children ... |
|  | Jno Parry[?] |  | This day Mr Walker pursuant to an Order of the last Committee ... |

Example for a layout that cannot be represented well with a table:

```text
                         27
St Andrew }  At a General Meeting of the Trustees at the Workhouse
Undershaft}  on Friday the 26th day of May 1734

Present     } Ordered pursuant to the Certificate that Carolina Purfoy
Mr Malcher  } and her children be taken into the House ...
Mr Hutchinson
Mr Finch

Jno Parry[?]  This day Mr Walker pursuant to an Order ...
```

Quality check before returning:
1. Did every visible text block appear in the Spatial Map?
2. Did you preserve left/right and top/bottom spatial relationships?
3. Did you keep margin notes separate from main text?
4. Did you link side notes, braces, or labels to the text they describe?
5. Did you preserve headings and title-like text?
6. Did you use `[?]` for uncertain readings?
7. Did you avoid summaries, modernization, and interpretation?