Lacuna Restorer
Mark a gap of known width with ? (one ? per missing character) or Leiden dots ([.], [..]). The tool retrieves similar inscriptions as parallels and proposes candidates for the missing characters, with a calibrated confidence.
Hypothesis tool, not ground truth
- This proposes candidates. It does not recover lost text as fact about what was on the stone.
- Scores are historical agreement with editor restorations in the edition (clean-gold), not a measure of true epigraphic reading.
- v2.0.3 jury: n=66 clean-gold, width-1 dominated (43/66), span-exact ≈0.26–0.29. Model accuracy differences were non-significant; hallucination rate differs by model.
- RAG path (this UI): on the same gold, research reports span-exact 0.258 → 0.379 (paired-bootstrap p=0.025). Treat as an experiment result, not a guarantee for any single call. Prefer high-confidence proposals; treat the rest as candidates.
mi mla?mlakaš
hasti:salia:pucsanas:la?tniθa
θerse:lau?ni|lθ:tetinaś
itan tu?uke