CLM Glossary Build Kit — QA Changes Applied (v1.1)

Tag: S-2026-06-07-glossary-kit-qa-changes-applied Type: own-writing Author(s): Paul Miles (AI-assisted) Date of source: 2026-06-07 Date ingested: 2026-06-07 Authority weight: high — the controlled record of what was changed in the kit and why. Raw file: /_raw_sources/S-2026-06-07-glossary-kit-qa-changes-applied.md

What it claims

Records the v1.1 upgrade of the CLM Business Glossary Build Kit in response to the QA review. All QA recommendations were applied; one was adapted on Paul’s instruction. Applied: a source-authority hierarchy (new source-authority.md, 7 tiers, golden source) carried into T-01/T-03/T-05 and consolidation; a definition-quality standard (new definition-quality.md, DQ-1…DQ-8) referenced by QA; the regulatory taxonomy expanded from 5 to 9 researched instruments (added CRD IV 2013/36/EU, IFRS 9, EBA/GL/2018/06 NPE, Basel III) with a Regulatory Family column; a strengthened bulk-approval control (GR-BA: Unique + High confidence + no regulatory/policy tag + sample pass); execution traceability (GR-12: Run ID, model/prompt/skill version) into T-03/T-06; OCR guidance (P-01); similarity score + Homonym type (P-02); glossary baseline version & change-request ref (P-05); a 15-point QA checklist with a QA score (P-06); T-06 additions incl. a Retired status; extended test data (27 concepts / 40 instances / 5 files) adding homonym, near-synonyms, an obsolete term and an authority-resolved conflict, with a rebuilt Answer Key; and a new Stage 7 (governance & change control: change-governance.md + prompt P-07). Adapted: the numeric confidence auto-approval gate (Findings 1 & 5) was implemented as categorical High/Medium/Low bands with indicative ranges and a calibration caveat (confidence governs routing, never auto-approval), because AI confidence is a self-assessment, not a calibrated probability.

Notable quotes

“Confidence governs routing, never automatic acceptance — a named human decides.” (calibration caveat added to _guardrails.md).

What’s speculative vs. asserted

Asserted: the change list and the verification performed (zero formula errors, V0.1 columns preserved, scripted cross-reference checks pass, docx validated, deck renders clean). Conditional: the expanded regulatory references are real and sourced but remain flagged for Regulatory/Risk SME validation before hard tagging; the confidence bands’ numeric ranges are indicative only.

Topics this feeds

CLM Glossary Acceleration Squad — the v1.1 build kit.

Open questions raised

SME validation of the expanded T-05 regulatory references; pilot calibration of the confidence bands and clustering on the client’s actual tooling.