Dimension Detection Overview¶
Status: Fast local dimension clue grouping implemented; confirmed semantics remain later Understand work
Authority: Current contract (2026-08-12)
Last reviewed: 2026-08-14
Question¶
How does Drawing Lab detect dimension evidence without claiming confirmed dimensions?
Short answer¶
Detection gathers mechanical clues. The mechanical clues include text readings, arrowheads, straight-line carriers, witness-line clues, and feature-contact clues. Detection may form possible-dimension bundles. Detection must not claim confirmed dimensions.
Current visible methods¶
| Method | Role |
|---|---|
| Fast Dimension Lines | Current fast Detection path with page-wide straight lines and local clue grouping |
| Dimensions from Line/Arc | Fallback and comparison evidence from fitted primitives |
| Dimensions from Vector Lines | Comparison evidence from Potrace 0.80-derived centrelines |
Detection can produce¶
- text readings that can contain measurements
- arrowhead evidence
- straight-line carriers
- witness-line clues
- feature-contact clues
- possible-dimension bundles
Understand must assign later¶
- dimension type
- parsed value and units
- view ownership
- measured-feature ownership
- ambiguity and confidence
UI boundary¶
The Detection UI lists possible dimensions.
It distinguishes complete, ambiguous, and partial clue sets.
The Detection UI must continue to say 0 confirmed dimensions until Understand assembles membership and ownership.
Raster lengths in Detection are page pixels, not drawing measurements.
Related pages¶
Open Questions¶
- Which sheet set becomes the first complete-dimension recall benchmark?
Sources¶
- Current contract:
table-dimension-detection-architecture.md(2026-08-12) - Status Vocabulary