## Database schema
This document describes the structure of the database underlying the personal historical-document transcription system.

---

## Schema diagram

```mermaid  
erDiagram
    document {
        INTEGER id PK
        TEXT name
    }

    job {
        INTEGER id PK
        INTEGER document_id FK
        TEXT status
        INTEGER retry_count
        DATETIME date_created
        DATETIME date_updated
        TEXT provider
        TEXT model
        TEXT prompt_name
        TEXT text
        TEXT error_detail
    }

    source {
        INTEGER id PK
        INTEGER document_id FK
        INTEGER job_id FK
        TEXT upload_name
        TEXT filename
        TEXT file_path
        DATETIME date_uploaded
    }

    revision {
        INTEGER id PK
        INTEGER source_id FK
        INTEGER revision
        TEXT text
        DATETIME date_created
    }

    document ||--o{ source : "has 0 or more"
    document ||--o{ job : "has 0 or more"
    job ||--o{ source : "processes 0 or more"
    source ||--o{ revision : "has 0 or 1"  
```

---  

## Table Relationships & Constraints
* A document can consist of 0 or more content sources. A document can have 0 or more jobs.
* A source can belong to only one job (which contains the original transcription). A source can only belong to one document. A source may have one optional transcription revision.
* A job can process one or more sources. A job can belong to only one document.
* A revision can belong to only one source. A source may have one optional revision.
* `Job.text` stores the original immutable provider transcription.
* Revision rows are optional user-authored edits and are derived from the original transcription. Unlike jobs, revision rows can be updated.

---

## Glossary
* **Document** - Documents consist of one or more content sources and their related transcriptions.
* **Source** - A content source that is transcribed to text. It can either be an image (.jpg, .tiff, .png) or a PDF (.pdf).
* **Image** - The scanned image of one page of a document.
* **PDF** - A PDF containing the image of one or more pages of a document.
* **Job** - A processing job ingests one or more sources, sends them to an AI model along with a prompt for transcription, then stores the results. The results are immutable, *including the original transcription*. The user can create a revision of the original transcription, but the user cannot modify the original.
* **Transcription** - The text contained in a content source. A job creates the original immutable transcription. A user can optionally create a revised transcription, or "revision".
* **Revision** - A revision is a user-created modification of an existing transcription. It is optional. Some original transcriptions will have no revisions.