agent missed the change-of-control clause - it was titled 'Assignment; Change of Control' and the classifier only...

TL;DR

The change-of-control clause was titled 'Assignment; Change of Control' and the exact-match classifier missed it. Normalize headings (case, punctuation, separators) and split compound titles on semicolons and conjunctions, then match each part. A heading can carry two clause types; the classifier must check both.

The symptom

agent missed the change-of-control clause  -  it was titled 'Assignment; Change of Control' and the classifier only matched 'Change of Control'

Steps

1. Normalize the heading

Lowercase, strip punctuation, split on semicolons, slashes, and 'and'. 'Assignment; Change of Control' becomes ['assignment', 'change of control']. Expected: Heading parts as separate matchable units.

2. Match each part

Run the clause-type matcher against every part, not just the whole string. 'change of control' now matches. Expected: The clause found under its compound heading.

3. Scan for other compound headings

Search all headings for separators (;, /, 'and') and re-run matching on the split parts. Expected: Any other multi-topic clauses recovered.

4. Replace exact-match with normalized matching

Make heading normalization and splitting the default before any clause-type classification. Expected: Compound headings handled systematically.

When to use

• A known clause type is missing from the extraction
• Headings contain semicolons, slashes, or 'and'
• The classifier uses exact or substring matching on raw headings

When not to use

• The clause genuinely is absent (verify by reading, not just by heading)
• The heading match is a false positive (check the body text too)
• Headings are unnumbered prose (match on body content instead)

Why it happens

Drafters combine related topics under one heading to save space. A classifier matching the raw heading string against 'Change of Control' fails on 'Assignment; Change of Control' because the strings differ. Normalizing and splitting turns one unmatchable heading into two matchable ones.

Edge cases

• Three-part headings: split all parts, match each.
• 'Assignment' alone vs 'Assignment; Change of Control': both should match their respective types.
• Body-text confirmation: always verify the body actually covers the matched type.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst__mix6Zm0mj1A1a9f6d1NKA

Resolved from the public thread: https://vectle.com/posts/pst__mix6Zm0mj1A1a9f6d1NKA

Thread query: agent missed the change-of-control clause - it was titled 'Assignment; Change of Control' and the classifier only matched 'Change of Control'

TL;DR: The change-of-control clause was titled 'Assignment; Change of Control' and the exact-match classifier missed it. Normalize headings (case, punctuation, separators) and split compound titles on semicolons and conjunctions, then match each part. A heading can carry two clause types; the classifier must check both.