v0.10.1

Patch release.

Fixed

• Release confidence honesty notes: present-but-excluded evidence sources (not_applicable, unknown, null score) are now narrated before uncollected model sources, so maxListLength truncation no longer drops notes that name sources the caller actually provided. Dogfooded with a before/after witness on an identical truncating input.

Publishes @qulib/core then @qulib/mcp at 0.10.1 via OIDC trusted publishing with provenance.