Document Detail
Deep dive into a single document -- overview, eval queries, chunk inspection, and settings.
What this page does
The Document Detail page provides a comprehensive view of a single ingested document, organized into four tabs: Overview, Eval, Chunks, and Settings.
Navigating here
Click on any document title in the Documents list. A breadcrumb at the top shows "Documents / Document Title" for easy navigation back.
Overview tab
Stats grid
Five metrics for this specific document:
| Metric | Description |
|---|---|
| Pages | Total page count from the PDF |
| Chunks | Number of semantic chunks extracted |
| Vectors | Number of indexed vectors in the search index |
| Images | Total images found, with technical image count in parentheses |
| Quality | Computed score based on chunks-per-page ratio (capped at 100%) |
Document metadata
A panel showing:
- Ingested -- when the document was processed
- Parsers Used -- which AI parsers ran during ingestion (e.g., "primary, validation, structure")
- Equipment Models -- the equipment models this document applies to (e.g., "Model A, Model B" or "_global" for cross-model documentation). This metadata controls which equipment-scoped queries retrieve chunks from this document.
- Chunks per Page -- average ratio (higher is better, indicating more structured content was extracted)
- Parse Errors -- number of pages that had parsing failures
Equipment scope
Same scope selector as the Documents list page. Choose between All Equipment, Specific Model, or Specific Instance. See Equipment Scoping for details.
Eval tab
Shows eval queries that are sourced from this specific document.
Generate for this doc
Click "Generate for this doc" to auto-generate 5 eval queries specifically from this document's content. The AI reads the document's chunks and creates test questions with expected answers.
Eval query table
| Column | Description |
|---|---|
| Status | Badge showing pending, approved, golden, or rejected |
| Query | The test question text |
| Type | Query type (spec, procedure, safety, troubleshooting, parts) |
| Last Run | Date of the most recent eval run that included this query |
| Result | PASS (green) or FAIL (red), or -- if never run |
A summary bar at the bottom shows total queries, passed count, and failed count.
Chunks tab
Chunk summary
Three metrics:
- Total chunks
- Vectors stored
- Chunks per page
Test Retrieval
This is a powerful debugging tool. Enter a query and click "Simulate" to see which chunks from this document would be retrieved, along with their relevance scores.
Each result shows:
- Rank -- position in the result list
- Chunk type -- semantic type badge (e.g., spec_table, procedure, safety_warning)
- Page -- which PDF page this chunk came from
- Score -- relevance percentage, color-coded: green (80%+), yellow (60-79%), red (below 60%)
- Content -- preview of the chunk text (first 3 lines)
Use this when: An eval query fails or a technician reports a bad answer. Simulate the query to see if the right chunks are being retrieved. Low scores or missing chunks indicate a retrieval problem.
Settings tab
Document name
Edit the display name of the document. Note: the doc_id (shown in gray below the input) is immutable -- only the display name can be changed.
Danger zone
A red-bordered section with a "Delete Document" button. This permanently removes:
- The document record
- All extracted chunks
- All indexed vectors
- All associated files
This cannot be undone. You will be redirected to the Documents list after deletion.