Evidence-Grounded Constraint Checking in Construction Documents
A new study published on arXiv proposes an innovative method for enhancing verification in document assessments within the construction industry. The research focuses on standardizing data collection and incorporates four precise rules to manage source citations and unresolved matters. An analysis involving 160 tasks across 29 construction projects demonstrated a 10.6% boost in decision-making accuracy when a method was refined to include specific image layouts. However, a wider evaluation revealed a slight decline in accuracy for a particular methodology. This investigation underscores the critical role of effective evidence management in streamlining document review processes in this sector.
Key facts
- Paper arXiv:2607.29058 presents an evidence-grounded pipeline for constraint checking in construction documents.
- The pipeline normalizes extracted facts, executes four-state rules deterministically, retains source spans, and escalates unresolved cases.
- Evaluation used 160 reference-based tasks from 29 construction projects.
- Reallocating a four-image budget to one overview and three overlapping tiles improved decision accuracy by 10.6 percentage points.
- The improvement was statistically significant (95% CI: 4.3 to 18.0; exact p = 0.031).
- The effect did not persist in a broader block: Region-RAG changed accuracy by -4.1 points (95% CI: -10.2 to 1.9; exact p = 0.209).
- An equal-image sensitivity favored the original approach.
- The study focuses on evidence allocation in automated document review for construction projects.
Entities
Institutions
- arXiv