nullspan v1.1.0 — Long-Context (>512 tokens) & Multi-Numeric Contract Disambiguation

Highlights

• Long-Context Sliding Window Retrieval (>512 tokens): Slides an IDF-weighted clause window across multi-section contracts and long documents across all field types (number, list, duration, text, date, boolean, enum).
• Context-Aware Multi-Numeric Disambiguation: Proximity-weighted lexical matching + unit-type alignment ($/USD, %, days/months/years, including hyphenated forms like 10-day and 12-month) + section header filtering + calibrated abstention (p_null = 0.992) on absent numeric fields.
• Clause-Scoped Numeric Lists (items: "number"): Extracts multi-tranche numeric arrays while excluding introductory counts.
• Published Artifacts:
  • Hugging Face Model Hub: https://huggingface.co/gnumanth/nullspan
  • Hugging Face Space: https://huggingface.co/spaces/gnumanth/nullspan
  • PyPI: https://pypi.org/project/nullspan/1.1.0/