Glossary¶
Status: One-term glossary for this surface
Authority: Current contract, accepted decision DL-010, and wiki surface plan vocabulary
Last reviewed: 2026-08-14
Question¶
What exact terms does this wiki reuse for one concept each?
Short answer¶
Use one approved term for one concept. Do not invent synonyms for stage names, claim words, or model names.
Stage terms¶
| Term | Meaning |
|---|---|
| Extraction | Capture and retain page evidence and Detection inputs |
| Detection | Produce independently observable mechanical evidence |
| Understand | Assign ownership, relationships, and engineering meaning |
| Validate | Measure results against human-reviewed truth and owner acceptance |
Claim terms¶
| Term | Meaning |
|---|---|
| Implemented | The behavior exists and can run |
| Verified | Software tests or deterministic checks confirm behavior |
| Validated | Representative human-reviewed truth supports the claim |
| Accepted | Owner decision binds the product or architecture boundary |
| Possible dimension | Mechanical clue bundle in Detection |
| Confirmed dimension | Dimension after Understand membership and ownership |
Evidence terms¶
| Term | Meaning |
|---|---|
| Page coordinates | Coordinates on the source page shared across stages |
| Review region | A Detection input area that still uses page coordinates |
| Provider | A named source of table-region proposals |
| Shared verifier | Finite-rule and evidence checks applied to table proposals |
| Outline | Boundary around ink |
| Centreline | Stroke middle line used as an engineering carrier candidate |
| Ridge | Centreline evidence from ridge pixels, paths, and vectors |
| OCR | Optical character recognition readings |
Named methods and models¶
| Term | Meaning |
|---|---|
| PP-DocLayoutV3 | Current preferred full-page table-region provider; threshold 0.50 |
| Geometry detector | Deterministic morphology and grid fallback |
| PicoDet | Optional small comparison table detector |
| Docling Heron | Bare layout checkpoint used for comparison; not the full Docling pipeline |
| Microsoft Table Transformer | Optional independent document-table proposal baseline |
| IBM TableFormer | Crop-only table structure recovery |
| PaddleOCR / PP-OCRv6 | Current OCR engine and model family |
Potrace potrace-0.80 |
Fallback appearance recipe with threshold 0.80 |
| Fast Dimension Lines | Current fast page-wide line and local clue path |
| Detected Sheet | Layered review assembly from compatible caches |
Open Questions¶
- Should product UI copy import this glossary unchanged?
Sources¶
- Current contract:
table-dimension-detection-architecture.md(2026-08-12) - Accepted decision: DL-010
- Status Vocabulary