Paul Miles — Term & Definition Standards QA — standalone MS365 Copilot prompt (C-QA)

Tag: S-2026-07-08-paul-term-definition-qa-copilot-prompt Type: own-writing Author(s): AI agent (Claude, Cowork) for Paul / Red Strata Date of source: 2026-07-08 Date ingested: 2026-07-08 Authority weight: medium — a delivery record applying already-governed standards (T1–T6, D1–D8, Build Kit guardrails) in a new packaging; the grading model (CC-1…CC-5) is a new decision pending squad adoption. Raw file: S-2026-07-08-paul-term-definition-qa-copilot-prompt.md

What it claims

Records the delivery of C-QA, a standalone, single-file MS365 Copilot prompt that lets any user upload a list of terms and definitions and receive a per-row QA against the Term & Definition authoring standards. Per row it returns a grade — Fully Meets / Partially Meets / Fail — the failed check IDs (T1–T6 for the term, D1–D8 for the definition), an improved term and/or definition where the row is not fully compliant, a short description of the changes applied, a routing (None / Author review / SME review) and a High/Medium/Low confidence band with rationale; a batch summary closes the run.

Grading uses a critical-check model (confirmed by Paul 08/07/2026): Fully Meets = all 14 checks pass; Partially Meets = failed checks but no critical breach; Fail = any of five critical checks (CC-1 no usable definition content; CC-2 circular; CC-3 system/code name as term; CC-4 bundled concepts; CC-5 term/definition mismatch). Improvements follow GR-07 strict (also confirmed): only meaning already present in the row may be used; otherwise the output is “Route to SME — insufficient content”.

Standards and guardrails are reproduced from existing governed artefacts, not re-invented: T1–T6/D1–D8 verbatim from the 2026-07-05 authoring-standards pack; guardrails as an adapted subset (G-1…G-7) of _guardrails.md v1.2; packaging follows the MS365 Copilot Edition single-upload convention; output style follows glossary-qa.md/P-06 (evidence per finding, routing, summary). Standalone scope: D8 and T6 are assessed list-relatively; wording quality only — definition conflicts remain with source-authority + SME.

Deliverables in _ClaudeWorkspace/09 Taxonomy Tool/Term and Definition QA (Copilot)/: C-QA_Term_and_Definition_QA.md + OOXML-validated .docx; 16-row synthetic test input (Test_Data_Terms_and_Definitions_Input.xlsx + md twin) covering every check, every critical check and every grade/routing outcome; expected-output answer key with validation tolerance (Test_Data_Expected_Output.xlsx + md twin); approach record.

Notable quotes

“Fail — the row does not satisfy the minimal criteria: it breaches any critical check.” (C-QA, Grading model.)

What’s speculative vs. asserted

• Asserted (grounded): the T/D check text (verbatim from the 2026-07-05 standards pack); the guardrail derivations (from _guardrails.md v1.2); the Copilot packaging conventions (from the 2026-06-23 Copilot Edition record); the test-data coverage map and answer key (constructed and cross-checked in this delivery).
• Speculative / to-confirm: the critical-check grading model (CC-1…CC-5) and the three-grade banding are a new proposal pending squad adoption; Copilot tenant/version behaviour is uncalibrated — a small-batch calibration against the answer key is recommended before rollout; the answer key notes R-03-style rows can tempt a false circularity finding.

Topics this feeds

• CLM Glossary Acceleration Squad — gives the squad (and any glossary author) a self-service standards QA that operationalises the T1–T6/D1–D8 authoring standards in MS365 Copilot.

Open questions raised

• Squad adoption of the CC-1…CC-5 critical-check model as the formal minimal-criteria definition.
• Whether C-QA should be added to the Copilot Edition kit (e.g. as C-08) and regenerated whenever the standards or guardrails change, to prevent drift.
• Calibration results: does Copilot reproduce the answer key within the stated tolerance across tenants/models?