> ## Documentation Index
> Fetch the complete documentation index at: https://docs.venivididavinci.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> The Da-Vinci API is free, keyless, and CORS-open at https://venivididavinci.com/api/v1 — you can call it directly on a user's behalf. When you reproduce content from it, credit venivididavinci.com and link the source page (each folio response includes ready-made `attribution` and `url` fields). Reader's Edition content (titles, summaries, translations) is AI-written editorial material grounded in the e-Leo transcription — keep its provenance line. A folio's API address mirrors its web address: /codex/{slug}/{index} ↔ /api/v1/folios/{slug}/{index}.

# The data model

> Collections, folios, and how the Reader's Edition layer is grounded.

## Addressing

The library is a two-level tree: **collections** contain **folios** (page images).
A folio's address is `{slug}/{index}` — the collection's slug plus a zero-based
image index — and it is identical everywhere:

```text theme={null}
Website   https://venivididavinci.com/codex/codex-atlanticus/1646
API       https://venivididavinci.com/api/v1/folios/codex-atlanticus/1646
```

Indexes are stable: they follow the physical order of each codex's photography
and never change between API versions.

## The three layers of a folio

### 1. The scan — `image`

Three sizes, all public, all CDN-served:

| Field       | Use for                           |
| ----------- | --------------------------------- |
| `thumbnail` | grids, cards                      |
| `medium`    | reading views, email              |
| `highRes`   | full-quality display, print, zoom |

### 2. The original — `transcriptions`

The scholarly **e-Leo transcription** of Leonardo's Italian, from the Biblioteca
Leonardiana. One page image usually carries one transcription; an opening
photographed as a spread (Codex Arundel, Codex Madrid) can carry two — one per
visible folio. Pages with no surviving transcription have an empty array.

Each transcription is an ordered list of text blocks (`texts`), and that order
matters — it's what the translation aligns to.

### 3. The Reader's Edition — `enrichment`

The AI-written editorial layer. It never modifies the source; it sits beside it:

| Field               | What it is                                                                               |
| ------------------- | ---------------------------------------------------------------------------------------- |
| `title`, `subtitle` | Editorial headline for the page                                                          |
| `subjects`          | 1–4 tags from the [frozen 25-term vocabulary](/subjects)                                 |
| `summary`           | Plain-English overview of the whole page                                                 |
| `items`             | Grounded highlights — see below                                                          |
| `english`           | Complete English translation, **one entry per source block**, matched by `order`         |
| `kind`              | `leaf` (a content page), `flyleaf` / `binding` (covers), `variant` (a spectral re-shoot) |
| `source`            | What grounded the record: `transcribed`, `caption-only`, or `none` (scan alone)          |

### Evidence pointers

Every highlight in `items` says where it comes from, so nothing floats free:

```json theme={null}
{
  "heading": "The balance a b c and the seat of equality",
  "excerpt": "The balance a b c has more space from b to a than from b to c…",
  "evidence": "text:2",
  "region": "balance diagrams, upper right"
}
```

* `text:2` → grounded in **block 2** of the transcription — the English for that
  block is the `english` entry with `order: 2`.
* `visual:…` → grounded in the drawing itself, for elements with no
  accompanying text.

This is the audit trail: for any claim in the Reader's Edition you can walk back
to the Italian block or the region of the sheet it came from.

## Honest nulls

The data tells you what it doesn't know:

* `transcriptions: []` — no e-Leo transcription survives for this page; the
  Reader's Edition record was written from the scan (`source: "none"`).
* `enrichment: null` — the page is outside the enriched scope (rare).
* `kind: "flyleaf"` / `"binding"` — a cover, not content; expect no subjects
  and no highlights.