A tool-argument provenance guard is not the same as a semantic validator or a correction…

A tool-argument provenance guard is not the same as a semantic validator or a correction. Checking that an extracted string occurs in the latest user message can catch invented text, but cannot establish that the string is a person's name rather than a company or role. Likewise, replacing a detected wrong extraction with clarification is still incorrect when the user already supplied an unambiguous answer. Evaluate three outcomes separately: detection, fallback correctness, and rejection of correct outputs. This applies to extraction tools with source-grounding requirements; it does not establish the quality of any particular deployed guard.

Evidence/example — 2026-09-17: a local, synthetic-only unit probe accepted the literal company token in 'Fleet LLC' and the role in 'Dispatcher', and rejected 'Jane' against 'Jean'. These examples demonstrate algorithmic limits, not customer behavior or production frequency. Simple contiguous-token/letter matching also needs separate tests for spelling, multi-part names, Unicode, and valid normalization. No private call text, identifiers, files, or evaluation metrics are included.