Redstrata Knowledge Base

CLM Glossary Acceleration Squad

35 min read

CLM Glossary Acceleration Squad

Status: active Started: 2026-06-01 Last touched: 2026-07-14 (conceptual-layers diagram of the Business Glossary)

Updated 2026-07-10 based on S-2026-07-10-weekly-briefing — routes two newly-surfaced open loops into Open questions / blockers below: the Change Request process intake decisions (submission channel, ≤2-day triage SLA, whether T-07/T-08 join the master kit templates) and the D9 “Source Anchored” enhancement recommended by the 10 Jul authoring-standards review. No prior content changed [S-2026-07-10-weekly-briefing]. Updated 2026-07-03 based on Weekly Briefing — 3 July 2026 — routes the glossary decisions still pending on Paul (following the 2026-06-28 reconciliation to a 399-term recommendation) into Open questions / blockers below. No prior content changed [S-2026-07-03-weekly-briefing].

Why this project exists

To define the squad structure and delivery method that accelerates creation of the CLM business glossary (and supporting business metadata / tags) across all Day 1 and Day 2 credit processes, so mastered definitions lead the imminent Day 1 system build and the inconsistencies already found between the Matrix, the ODS and the Day 1 design are resolved at source [S-2026-06-01-paul-clm-glossary-squad].

Current state

  • Recommended design captured in a 3-slide leadership deck (black-on-white, editable native diagram) at 01 Taxonomy\CLM_Glossary_Squad_Structure_and_Delivery_Method.pptx [S-2026-06-01-paul-clm-glossary-squad].

  • Two empowered, cross-functional squads on a one-week sprint cadence; Squad A on the Day 1 critical path, Squad B on Day 2 (ODS, monitoring, servicing, collections) in parallel [S-2026-06-01-paul-clm-glossary-squad].

  • Per-squad roles: Squad Lead/PO ×1; Business Data SMEs/Definition Owners ×2–3 (empowered, ~2 days/wk); Data/Business Analysts ×2; Data Steward/Governance Lead ×1; shared Agile Facilitator (~0.5), AI/Data Engineer (~0.5) and on-call Regulatory/Risk SME (~0.3). Core dedicated ≈ 6–7 per squad [S-2026-06-01-paul-clm-glossary-squad].

  • Sizing model: ~500–1,000 terms total; ~120–140 terms/week per squad → ~250/week combined once AI clustering runs; Day 1 critical terms mastered weeks 1–4 (4–6 weeks); full baseline ~5–6 weeks; 3rd surge squad only if scope nears 1,000+ [S-2026-06-01-paul-clm-glossary-squad].

  • Fast-track governance: SMEs approve in-squad (SLA 3–5 working days); AI-unique terms bulk-approved; only unresolved conflicts escalate to a standing fortnightly CLM Design Sub-Committee slot; reuses existing programme/BAU fora (DQ Forum, Enterprise DG Council, BCBS 239 attestation) — no new bodies [S-2026-06-01-paul-clm-glossary-squad].

  • Delivery flow: Harvest → AI cluster & de-duplicate → (unique → fast-track approval | conflicts → AI-proposed resolution → SME workshop) → master definition + Policy/Process/IRB/BCBS 239 tags → publish to governed master (Excel → SharePoint → Confluence/Comala) [S-2026-06-01-paul-clm-glossary-squad].

  • AI usage: harvest/cluster terms; identify unique values for fast baseline approval; detect conflicts and propose consolidated definitions + rationale; suggest regulatory/business-metadata tags; draft & QA via M365 Copilot — humans decide, AI accelerates [S-2026-06-01-paul-clm-glossary-squad].

  • Build kit DELIVERED (2026-06-05, approved by Paul): all 20 files at 09 Taxonomy Tool\Glossary Build Kit\ (prompts/, skills/, templates/, guidelines/ + README index). QA verified programmatically: T-06 preserves all 19 Taxonomy Template V0.1 columns unchanged and adds 20 fields (mastering, hard/soft tags, approval/audit, technical-metadata & DQ-metric link placeholders); every prompt’s column list matches its template’s actual headers; status lifecycle consistent across templates, skills and guidelines; guidelines docx passes OOXML validation. T-05 ships 5 regulatory starter rows (CRR, BCBS 239, EBA/GL/2020/06, EBA/GL/2016/07, AnaCredit) flagged for Regulatory/Risk SME validation before hard tagging [S-2026-06-05-clm-glossary-build-kit-approach].

  • Process diagram deck added (2026-06-05): Glossary Build Kit\guidelines\CLM_Glossary_Process_Diagram_with_Kit.pptx — recreates the squad deck’s slide-3 delivery flow (editable native shapes, AI badges) annotated with the Prompt / skill.md / Template used at each stage, plus a stage-by-stage kit reference table [S-2026-06-05-clm-glossary-build-kit-approach].

  • End-to-end test data added (2026-06-05): Glossary Build Kit\Test Data\ — four synthetic CLM source files (xlsx policy glossary, docx ODS dictionary, pdf Day-1 design defined-terms, xlsx Matrix extract; 34 instances → 22 concepts) plus an Answer Key and README. Deliberately covers all Step-2 scenarios (12 unique, 2 exact duplicate, 2 potential duplicate, 3 synonym, 3 conflict) with discrimination traps (Exposure≠EAD, Past Due≠Days Past Due) and abbreviation matching. Serves both as an agent-validation harness (≥90% match to Answer Key) and squad training set; all synthetic, no real data [S-2026-06-05-clm-glossary-build-kit-approach].

  • QA review applied — build kit now v1.1 (2026-06-07): independent QA (8.8/10, approve for pilot with enhancements) actioned in full. ALL recommendations applied; ONE adapted on Paul’s instruction (numeric confidence auto-approval gate → categorical High/Medium/Low bands + calibration caveat, because AI confidence is not a calibrated probability). Applied: source-authority hierarchy (new source-authority.md, golden source, tiers into T-01/T-03/T-05); definition-quality standard (new definition-quality.md); regulatory taxonomy expanded 5→9 researched instruments with families (T-05); strengthened bulk-approval control (GR-BA); execution traceability (GR-12, into T-03/T-06); OCR guidance (P-01); similarity score + Homonym type (P-02); glossary baseline version & change-request ref (P-05); 15-point QA checklist + QA score (P-06); T-06 Retired status + 29 added fields (all 19 V0.1 columns preserved); extended test data (27 concepts/40 instances/5 files: homonym, near-synonyms, obsolete term, authority-resolved conflict); and a new Stage 7 governance & change control (change-governance.md + P-07). Guidelines, README and process-diagram deck updated to the 7-stage flow. Change report: 09 Taxonomy Tool\CLM_Glossary_Build_Kit_QA_Changes_Applied.md [S-2026-06-07-glossary-kit-qa-review][S-2026-06-07-glossary-kit-qa-changes-applied].

  • Build kit scope (as approved): 6 tool-agnostic prompts (harvest/registration, cluster & de-dup, conflict consolidation, hard/soft policy-process-regulatory tagging, master assembly & status, QA & approval readiness), 7 skill.md files (shared _guardrails.md carrying the Taxonomy Knowledge Tool HLD’s 11 GR controls + six task skills), 6 Excel templates (source register, harvest staging, clustering/disposition, conflict/SME workshop pack, reference lists, master glossary extending the client’s Taxonomy Template V0.1 without deleting fields), and one client-ready Squad Operating Guidelines doc. Components designed for reuse in the Taxonomy Knowledge Tool. Six delivery risks recorded (clustering accuracy, soft-tag overreach, consolidation hallucination, source parse failures, Excel concurrency, prompt portability) [S-2026-06-05-clm-glossary-build-kit-approach].

  • Backlog tooling: an Excel backlog tracker manages definitions across both squads’ sprints from Harvested to Published (Done); its Backlog sheet is structured for CSV import into Jira Data Center as Epics/Stories (29 columns incl. Issue Type, Epic Link, Status, Sprint, Squad/Component, regulatory tags, definition fields), with a Dashboard, Field Guide and dropdown Lists [S-2026-06-01-paul-clm-glossary-backlog-template].

  • Variance + blind testing (2026-06-08): ran the standard test three times (runs 01/02/03) — identical results each time (27/27 concepts, 0 duplicates, 10/10 incremental disposition), i.e. 0 variance but in-session/same-operator (deterministic), not independent; true variance needs separate sessions or a different model (variance_summary.md). Then built Test Data 3 (blind) — a harder independent dataset (16 instances → 13 concepts, 3 sources) with traps (Charge homonym; Cure Period 30d-covenant vs 90d-NPE homonym-vs-conflict; Limit vs Facility Limit non-merge; cross-file exact-dup+synonym for Effective Maturity; preserved typo ‘Colateral Cover Ratio’; non-self-contained & bundled definition-quality fails; obsolete term; reg-tagged-unique bulk-approval control). Cold run scored 13/13 concepts, 0 wrong merges, 16/16 disposition (100%) with all eleven traps handled. Honest caveat recorded: author = solver, so this is a harder cold self-test, not independent blind validation — Test Data 3 + ground truth now exist for an external operator/model to run blind (the recommended real test before go-live) [S-2026-06-08-glossary-incremental-test-data].

  • End-to-end test executed — PASS (2026-06-08, run-2026-06-08-01, human_steps=simulate): ran the harness over Test Data 2. Phase 1 mastered 27 concepts with 0 duplicates (27/27 present); Phase 2 incremental disposition accuracy 10/10 with correct Business Concept Key reuse on every match. All hard gates passed with evidence: Default→Change Request CR-001 (not overwritten), Exposure Class consolidated to one new concept via SME workshop, Security collateral attached to the collateral concept not the instrument, Behavioural Score legacy Retired+superseded, Customer absorbed as alias. Caveats recorded: validates LLM-executed kit logic (not a vendor tool / not blind accuracy — operator authored the dataset), synthetic small data, single in-session run — 3 separate-session runs recommended for true variance before go-live. A scorer header-detection bug (note row mis-read) was found and fixed mid-run; produced data was correct throughout. Report: Glossary Test Harness/run-2026-06-08-01/report/test-report.md [S-2026-06-08-glossary-incremental-test-data].

  • Test harness created (2026-06-08): Glossary Test Harness/ now holds a runnable test kit — P-TEST runner prompt, test-harness.md skill (run-folder layout, hard gates, graded metrics, simulate/gate human steps, 3-run variance), score_results.py deterministic scorer (Answer Key as sole ground truth; syntax-checked, dry-run-verified), and a README. Copies of the prompt + skill also in Test Data 2. Lets Paul execute the end-to-end + incremental functional test on demand; no test run yet [S-2026-06-08-glossary-incremental-test-data].

  • Incremental-load test data + automated-test proposal (2026-06-08): added a self-contained Test Data 2 folder (original Test Data untouched) with Batch 1 (initial load — the 5 sources, 27-concept baseline with BCKs) and Batch 2 (incremental load — two new sources, Source 6/7) exercising every incremental path: match-no-change, match-via-Change-Request (Default), add-alias (Customer→Counterparty), new concepts (LTI, Write-off, Bureau Score), in-batch conflict (Exposure Class), homonym discrimination vs the existing master (Security collateral→BCK-0012 not BCK-0013), and supersession/retirement (Bureau Score supersedes Behavioural Score legacy). 4-sheet Answer Key + README. Also produced Test Approach.md: an honest proposal that an automated functional test is feasible (Claude acts as the AI agent executing P-01..P-07 over the data, scored deterministically against the Answer Key), with caveats (validates LLM-executed kit logic not a vendor tool; report variance over 3 runs; human steps simulated/gated) and the access/folder/thresholds needed — no test run yet, awaiting Paul’s go/no-go [S-2026-06-08-glossary-incremental-test-data].

  • Build kit v1.2 — incremental update support (2026-06-08): added a Business Concept Key (BCK) (durable per-concept identity) to T-03 and T-06, and guardrail GR-14 so additional source-file batches merge into the master deterministically. P-02 now runs first-run vs incremental modes (existing master = comparison scope; match against it first); P-05 reuses an existing BCK to update a concept or mints a new one; a new T-03 disposition “Update existing master concept” routes incremental matches; QA check 9 verifies BCK integrity. Pre-baseline = more rows through Stages 1–5; post-baseline = a Stage 7 delta into the next version. Guidelines §11, the process-diagram deck (incremental callout looping to Step 2) and README updated; change report §6 addendum records it [S-2026-06-08-glossary-kit-incremental-updates].

  • MS365 Copilot edition DELIVERED (2026-06-23): a Copilot-friendly second edition of the build kit at 09 Taxonomy Tool\Glossary Build Kit - MS365 Copilot Edition\. Because Copilot users upload files into the chat window, each of the 7 pipeline steps is merged into a single uploadable file (C-01…C-07, provided as both .md and .docx) that bundles the step’s prompt + its skill method(s) + the shared guardrails + the template column schemas (fallback); the prompt then asks the user to upload the templates that step needs. Master prompts/skills/templates and the original process-diagram deck are unchanged — the edition is an additive sibling folder. A copy of the deck (..._Copilot.pptx) now lists the C-0x file to run at each step. Packaging grounded in Microsoft’s Copilot file-upload guidance; merged content is verbatim from the master kit (only the “read the external skill file” opener was adapted to “method embedded below”). README + work-breakdown record saved in the folder [S-2026-06-23-glossary-copilot-edition].

  • Industry reference Business Glossary DELIVERED (2026-06-27): an exhaustive-but-prioritised industry Business Glossary for the Credit Lifecycle produced as the seed master content the build kit is designed to govern. 140 terms across 18 lifecycle stages (Day 1 = 59, Day 2 = 81), scope Corporate / SME / Capital Markets primary with Consumer aligned to the asset class, each tagged to the 7 named regulatory cycle elements (underwriting, EWS, covenant monitoring, SICR, annual review & re-rating, forbearance, UTP) and carrying relationships + Policy/Process/Regulatory tags. Priority tiers Core 59 / Essential 64 / Extended 17; Hard tags 76 / Soft 64. Uses the T-06 column model (+4 analytical columns); bank source-system metadata left blank for ODS mapping; all rows Drafted. Regulatory anchors web-verified current (CRR3 eff. 1 Jan 2025; EBA/GL/2020/06; EBA/GL/2016/07; IFRS 9; AnaCredit; BCBS 239; ECB RDARR Guide May 2024). Deliverables (Excel + md glossary, B&W PPTX + md exec summary, approach record) saved to 03 Credit Lifecycle\Industry Business Glossary\ [S-2026-06-27-paul-clm-industry-business-glossary].

  • MS365 Copilot edition — governance docs added (2026-06-24): Copilot copies of the three governance documents now sit in Glossary Build Kit - MS365 Copilot Edition\guidelines\ — Squad Operating Guidelines, Squad Operating Procedures (both .md + .docx) and Squad Task Checklist (.xlsx + .md), each suffixed _M365_Copilot. They restate the master governance text verbatim and change only the run model: “execute P-0x with the skill files + _guardrails.md attached” → “execute C-0x in Copilot — upload the one merged file plus the templates the step names”; the delivery-flow table and checklist columns become Copilot file (upload) / Skills (merged in C-file); cross-references point to the Copilot deck and checklist. All rules, roles, statuses, controls, the seven-stage flow, manual fallbacks M-01…M-07 and the 59-task checklist are unchanged. Master governance docs untouched. A governance-docs approach record was saved beside them [S-2026-06-23-glossary-copilot-edition].

  • Programme Tracker (burndown & flow) DELIVERED (2026-06-27): an Excel measurement & forecasting layer over the build kit that reads the counts the kit already emits (T-03 Clustering Summary + T-06 Status Summary 9-state lifecycle) and produces the two reports in the brief — a Cumulative Flow Diagram + current-distribution + flow-totals, and a burndown with rising scope, remaining-to-master, dashed ideal glide path and a velocity-based forecast line to a forecast completion date + RAG vs plan. Per Paul’s confirmed choices: Excel-only (Power BI offered as optional fast-follow, not built); history captured both ways (one-click per-sprint Snapshot Log + a Status Reconstruction sheet backfilling cumulative Approved/Published from T-06 dates); combined programme view (no per-squad split v1.0); parameterised scope that grows with harvest. Workbook CLM_Glossary_Programme_Tracker.xlsx (7 sheets: Exec Summary, Burndown, Flow (CFD), Sprint Snapshot Log, Current Counts, Status Reconstruction, Config & Guide), macro-free, refresh via Power Query or paste then append a snapshot row. A fully-populated demo (8 sprints mock data, ~87% mastered, RAG GREEN) saved to 09 Taxonomy Tool\Programme Tracker Demo\. Verified: zero #REF!/#DIV/0!/#VALUE!/#NAME? errors (only intentional hidden NA() chart-gap markers). Saved to 09 Taxonomy Tool\ with an Approach & Work Breakdown record [S-2026-06-27-paul-clm-glossary-programme-tracker].

  • Term & Definition Authoring Standards DELIVERED (2026-07-05): a 2-slide standards pack (B&W on white, Asperi font; PPTX + md twin) telling squad authors how to create the Term (T1–T6 — a NEW naming standard: business name, singular Title Case noun phrase, regulator’s full official form with acronyms as aliases, one term = one concept with homonym qualification, no definition in the name, consistent compound construction) and the Term Definition (D1–D8 — restates the Build Kit’s DQ-1…8 with ISO/IEC 11179-4 anchors and the ISO 704 intensional-definition pattern). Industry-led (ISO 704, ISO/IEC 11179-4/-5, DAMA-DMBOK — web-verified); scope boundary preserved (wording quality only; conflicts stay with source-authority + SME via P-06). Gap noted: the kit had definition-quality.md but no formal term-naming standard — Slide 1 fills it. Saved to 03 Credit Lifecycle\Glossary Standards\ with an approach record [S-2026-07-05-paul-glossary-term-definition-standards].

  • Standalone standards-QA Copilot prompt (C-QA) DELIVERED (2026-07-08): a single-file MS365 Copilot prompt at 09 Taxonomy Tool\Term and Definition QA (Copilot)\ that any user runs by uploading the prompt plus a term/definition list. Per row it grades Fully Meets / Partially Meets / Fail against the 14 checks (T1–T6, D1–D8 verbatim from the standards pack), lists failed check IDs with evidence, proposes an improved term and/or definition under GR-07-strict (row content only; otherwise “Route to SME — insufficient content”), describes the changes applied, and routes (Author review / SME review) with High/Medium/Low confidence + rationale; a batch summary flags systemic patterns. Fail = any of five critical checks (CC-1 no definition content; CC-2 circular; CC-3 system/code name; CC-4 bundled concepts; CC-5 term/definition mismatch). Ships as .md + OOXML-validated .docx with a 16-row synthetic test input and an expected-output answer key (xlsx + md twins) covering every check, critical check, grade and routing outcome, incl. Fail-with-improvement (system-name term, homonym split) and Fail-route-to-SME cases [S-2026-07-08-paul-term-definition-qa-copilot-prompt].

  • Reference-usage guidance DELIVERED (2026-07-09): a guidance doc (docx + md, 09 Taxonomy Tool\Glossary_Build_Kit_Reference_Usage_Guidance.*) answering Paul’s questions on how T-05 references are applied. Key clarifications, all derived from kit v1.2 artefacts: T-05 rows are tagging metadata (pointers), not content sources — no regulatory source file is needed for P-04 tagging and the prompt never searches online (GR-05/GR-07; the REG-001…009 web research happened once at kit build time, flagged for SME validation); a regulation listed with no source file tags normally but its own definitions never enter the glossary unless the instrument/extract is separately harvested at Tier 1/2 via T-01/P-01; Policy/Process sheets ship empty and are populated at Stage 0 task 0.6 (a policy may need BOTH a T-01 registration and a T-05 row — separate purposes). Also assessed pros/cons of using the 140-term Industry Business Glossary as a co-source (completeness/prioritisation/divergence-detection benefits vs Tier-6 authority discipline, POL/REG-010…013 tag remapping to T-05, +140-term workload, GR-BA review load) with three usage patterns (A seed master — recommended by the file’s design; B co-source; C benchmark); choice pending with Paul/squad [S-2026-07-09-paul-glossary-reference-usage-guidance].

  • Change Request process DELIVERED (2026-07-10): a simple user-facing CR route implementing Stage 7’s missing intake layer, at 09 Taxonomy Tool\Glossary CR Process\ — 3-slide B&W process deck + md record, T-07 Change Request Form (requester Excel: 4 required fields per term — Change Type Add/Change/Retire/Other · Term Name · Proposed Definition/Change · Reason; one row per term so single and bulk share the same Change-List sheet, or attach an existing Excel list with the four minimum columns; yellow-input convention, dropdowns, worked example, Guide sheet) and T-08 Change Request Register (squad-side 22-column log, About rules sheet, formula-driven status summary — recalc-verified 0 errors). Five-step flow: Raise → Log & triage (DS, ≤2 wd, CR-nnn) → Assess & draft [AI, P-07; fallback M-07] → Decide (SME/Definition Owner at the existing 3–5-wd fast-track SLA, +RRS for regulatory-tagged terms, disputes to the fortnightly committee slot) → Apply & close (T-06 Entry Version +1, CR ref linked, history preserved, next baseline + delta report). Triage re-routes large NEW-term sets backed by source documents as incremental Stage-1–5 batches (GR-14); bulk changes run as one Bulk CR under GR-BA discipline. No new roles, forums, T-06 statuses or tooling; guardrails GR-01/11/12/13/BA/14 reused [S-2026-07-10-paul-glossary-cr-process].

  • Terms of Reference DELIVERED (2026-07-11): a one-page squad TOR (03 Credit Lifecycle\CLM_Business_Glossary_Squad_TOR.md + .docx, verified single page) with exactly three sections per Paul’s brief — Squad Objectives; Key scope items (In / Out of scope); Key considerations & dependencies. It consolidates the approved squad design into a governance framing document: objectives (single mastered glossary with Policy/Process/Regulatory metadata, Day 1 terms front-loaded, H2 2026 SREP commitment, 399-term reconciled baseline recommended, publish via the 7-stage Build Kit flow); scope boundaries (out: technical metadata/ODS mapping, DQ metric implementation, new governance bodies, Taxonomy Knowledge Tool build, new vendor tooling, glossary meta terms); and dependencies (SME ~2 d/wk, harvest-count re-baseline, decisions pending on Paul, T-05 validation, guardrail discipline, six delivery risks). No new claims — all content restates existing sourced material [S-2026-07-11-paul-clm-glossary-squad-tor].

  • TOR client-applied update (2026-07-12): Paul manually revised the 2026-07-11 TOR to reflect how it is being applied at the client; the revision supersedes the 2026-07-11 version as the operative framing. Changes: the fortnightly escalation slot moves from the CLM Design Sub-Committee to a CM Business Impact Forum (all three mentions); the SME availability assumption becomes ~5 days/week (vs ~2 days/week in the June sizing model [S-2026-06-01-paul-clm-glossary-squad] — whether the throughput model has been re-run on this assumption is not stated); planning figures (week timings, squad counts, per-week throughput, the 399-term baseline) removed from objectives (scope stays ~500–1,000 terms); decisions pending now sit with the squad, with the 399-term adoption decision replaced by “finalise the scope of terms included based on bottom up assessment and SME input”, the asset-class lead order dropped, and the D9 “Source Anchored” decision removed; publishing simplified to Excel → SharePoint (Confluence/Comala references removed); wording now “AI accelerates, squad members decide” [S-2026-07-12-paul-clm-glossary-squad-tor-client-update].

  • FORUM RENAME (2026-07-12): the fortnightly decision forum formerly named the CLM Design Sub-Committee is now the Credit Monitoring Business Impact Forum (BIF) — declared in Paul’s brief; use BIF in all glossary escalation references going forward. Earlier entries on this page retain the old name as written at the time (never-delete discipline) [S-2026-07-12-paul-glossary-impact-assessment].

  • Step 4 Impact Assessment & BIF routing DELIVERED (2026-07-12): a very simple, indicative SME impact view for terms routed to the BIF, at 09 Taxonomy Tool\Glossary Impact Assessment\ — Glossary_Term_Impact_Assessment_Form.docx/.md (Sections A escalation / B per-term impact — areas, affected assets, indicative scale, pre-effective change activities?, detailed-IA-recommended?, decision sought / C SME sign-off / D BIF use only; 10–20 min per term, “Unknown” valid, single or multiple terms per form) + Glossary_Step4_Impact_Assessment_Sub_Process.pptx/.md (3 B&W slides: 4a.1 Assess → 4a.2 Log & route (DS, ≤2 wd, Impact Backlog) → 4a.3 Decide (BIF; may refer to Enterprise DG Council / ARB / DQ Forum) → 4a.4 Track → 4a.5 Clear & resume at Step 5) + approach record. Impact Backlog statuses: Escalated — awaiting BIF · Referred — detailed IA in progress · Referred — other forum · Deferred — awaiting change delivery · Cleared · Rejected. No new T-06 statuses, roles, forums or tooling; a Deferred term may reach Approved but is not Published/effective until the BIF confirms the pre-effective change backlog is delivered; guardrails GR-01/GR-11/GR-12 reused. Also includes Glossary_Impact_Assessment_Exec_View.pptx — a 1-slide exec-level view of the same sub-process (step name + actor only, decision-outcome strip, single holding rule) [S-2026-07-12-paul-glossary-impact-assessment].

  • Conceptual-layers diagram DELIVERED (2026-07-14): a 1-slide executive view of how the Business Glossary content is layered (03 Credit Lifecycle\Glossary Conceptual Layers\CLM_Glossary_Conceptual_Layers.pptx + md twin + approach record; B&W on white, Aspira font). Read bottom to top under the rule “terms are defined once at the layer where they originate; layers above reuse and aggregate — never redefine”: Layer 1 Day 1 Origination→Drawdown split by the 7 glossary processes with per-process counts (101 terms); Layer 2 Day 2 Monitoring→Recovery (10 operational stages + Regulatory & Capital Reporting data terms; 100 terms); Layer 3 Management Reporting (portfolio credit MI & watchlist, covenant/EWS/forbearance MI, BCBS 239 / ECB RDARR aggregation & CRO attestation); Layer 4 External & Regulatory Consumers (data/loan tapes; IRB rating system PD/LGD/EAD, COREP IRB C 08, FINREP F 18/F 19, Pillar 3 CR6–CR10, AnaCredit — per the Apr-2026 IRB briefing’s ODS-as-certified-upstream model). Prioritisation guidance restates the glossary sequencing (Day 1 Core 59 first → Day 2 → derived reporting terms last); counts flagged as the 201-term baseline with the 399-term reconciled scope alongside. Declared gap: no existing source describes the bank’s loan-tape outputs — that layer came from Paul’s brief and is described generically [S-2026-07-14-paul-glossary-conceptual-layers].

Decisions made

  • 2026-06-01 — Recommend two parallel squads (flex to three) on a one-week sprint cadence, with Day 1 critical terms front-loaded to lead the system build. Why: 4–6 week Day 1 deadline against a 500–1,000 term scope with SMEs at ~2 days/week [S-2026-06-01-paul-clm-glossary-squad].

  • 2026-06-01 — Adopt a fast-track governance route empowering squad SMEs, with escalation only to the existing CLM Design Sub-Committee, to avoid standing up new fora [S-2026-06-01-paul-clm-glossary-squad].

  • 2026-06-01 — Use AI for term harvesting, clustering/de-duplication, unique-value baselining, conflict detection with proposed resolutions, and metadata tagging; tooling stays on Excel + SharePoint + M365 Copilot governed via Confluence/Comala [S-2026-06-01-paul-clm-glossary-squad].

  • 2026-06-05 — Build-kit work breakdown drafted: one prompt + one skill.md per AI-assisted delivery step, templates built first (prompts reference their exact columns), all components client-safe and Taxonomy-Knowledge-Tool-compatible. Why: 1:1 mapping to the proven six-step delivery flow keeps the kit auditable and reusable. Build awaits Paul’s approval [S-2026-06-05-clm-glossary-build-kit-approach].

  • 2026-06-07 — Decline-as-written / adapt: the QA’s numeric confidence auto-approval gate (≥95% auto-eligible) was NOT applied literally. Why: AI confidence is a self-assessment, not a calibrated probability; a numeric auto-approval gate creates false assurance in a regulated setting. Paul chose adapted categorical bands + a calibration caveat with no auto-approval [S-2026-06-07-glossary-kit-qa-review].

  • 2026-06-07 — Apply the remaining QA recommendations in full (source authority, definition quality, expanded regulatory taxonomy, strengthened bulk approval, execution traceability, prompt/template enhancements and a new Stage 7). Why: all are enhancement opportunities consistent with regulated glossary governance; regulatory references were researched and confirmed, not invented [S-2026-06-07-glossary-kit-qa-changes-applied].

  • 2026-06-08 — Adopt a Business Concept Key + GR-14 incremental-matching rule rather than relying on the QA dedup check alone. Why: makes additional-batch merges deterministic (a confirmed match persists by key) and prevents duplicate master rows when sources arrive in waves — answering Paul’s incremental-update question [S-2026-06-08-glossary-kit-incremental-updates].

  • 2026-06-08 — Build incremental test data in a NEW Test Data 2 folder rather than editing the existing folder, and propose (not run) an automated functional test scored against an Answer Key. Why: avoids the OneDrive overwrite friction and keeps the original dataset intact; the test proposal is honest about what an LLM-executed test can and cannot prove, leaving the go/no-go to Paul [S-2026-06-08-glossary-incremental-test-data].

  • 2026-06-23 — Produce the MS365 Copilot edition as ONE merged file per step (prompt + skills + guardrails + schema) in a new sibling folder, rather than modifying the master kit or shipping the files separately. Why: Copilot users upload files into the chat, so a single self-contained upload per step (which then asks for the templates) removes the multi-file juggling and the “can’t access the skill file” failure mode, while the master kit stays the unchanged source of truth. Trade-off accepted: skill/guardrail text is duplicated into each step file, so the edition must be regenerated when the master kit changes [S-2026-06-23-glossary-copilot-edition].

  • 2026-06-24 — Create Copilot editions of the three governance docs by verified literal replacement + a P-0x→C-0x token rename (not a rewrite), changing only the tool-interaction wording and leaving all rules/controls/fallbacks verbatim. Why: keeps the governance text faithful and auditable while making it describe the single-file upload workflow; master docs preserved [S-2026-06-23-glossary-copilot-edition].

  • 2026-06-27 — Build the progress tracker in Excel only (not Power BI) and present a combined programme view with a parameterised scope, capturing history both via a per-sprint Snapshot Log and a date-derived Status Reconstruction. Why: keeps measurement where the data already lives and matches the client’s Teams + Excel stack with no new licensing; Power BI remains an optional exec fast-follow; scope grows with harvest so the burndown re-baselines as backlog is discovered [S-2026-06-27-paul-clm-glossary-programme-tracker].

  • 2026-07-05 — Publish the term/definition authoring standard as a 2-slide pack that restates DQ-1…8 (rather than writing a competing definition standard) and adds a new T1–T6 term-naming standard, each rule carrying its industry anchor (ISO 704 / ISO/IEC 11179-4 & -5 / DAMA-DMBOK). Why: squad authors get one page per artefact while the Build Kit remains the governing standard; naming was the unfilled gap [S-2026-07-05-paul-glossary-term-definition-standards].

  • 2026-07-08 — Grade the standalone standards QA with a critical-check model (Fail = breach of any of CC-1…CC-5; Partially Meets = other failed checks; Fully Meets = all 14 pass) rather than a score threshold, and hold improvements to GR-07 strict (no meaning added; insufficient content routes to SME). Why: a score threshold could pass a row with one fatal flaw or fail one on minor issues, and invented definition content is unacceptable in a regulated glossary — both confirmed by Paul [S-2026-07-08-paul-term-definition-qa-copilot-prompt].

  • 2026-07-10 — Implement glossary change requests as a minimal intake layer over Stage 7 (T-07 form + T-08 register) rather than a new process: single and bulk share one Change-List sheet (or an attached user Excel with four minimum columns), only 4 fields are required per term, and large NEW-term sets backed by source documents are re-routed at triage as incremental batches under GR-14. Why: Stage 7 (P-07 / change-governance.md / Procedures §11) already defines CR handling, approvals, versioning and controls — the gap was user intake and tracking; minimal required fields maximise adoption while the squad completes impact, QA and versioning under P-07 [S-2026-07-10-paul-glossary-cr-process].

  • 2026-07-12 — Keep the Step 4 impact route indicative and inside the squad: the SME form is a 10–20-minute knowledge-based view whose main output is a routing signal (does the BIF need a detailed impact assessment, outside squad scope, before deciding?); pending terms are tracked on a lightweight Impact Backlog rather than new T-06 statuses, and “Approve with conditions” reuses the existing Approved ≠ Published distinction to hold terms until pre-effective change activities are delivered. Why: mirrors the CR-process precedent (minimal intake over existing controls), keeps the master lifecycle untouched, and keeps detailed analysis out of the squad’s scope [S-2026-07-12-paul-glossary-impact-assessment].

  • 2026-07-12 — Adopt the forum rename CLM Design Sub-Committee → Credit Monitoring Business Impact Forum (BIF) in all new glossary artefacts; existing kit documents keep the old name until their next revision. Why: declared in Paul’s brief; wholesale retro-editing would conflict with never-delete document discipline [S-2026-07-12-paul-glossary-impact-assessment].

Open questions / blockers

  • Actual harvested term count across Day 1 + Day 2, and the size of the Day 1 critical subset — both re-baseline the throughput model [S-2026-06-01-paul-clm-glossary-squad].

  • Whether ~2 days/week SME availability holds once squads are live.

  • Whether the CLM Design Sub-Committee will host the standing fortnightly glossary-decision slot or a separate fast-track authority is required.

  • 2026-06-05 (weekly briefing) — throughput and squad-count figures remain planning estimates; re-baseline the sizing model once the actual harvested term count is confirmed [S-2026-06-05-weekly-briefing].

  • 2026-07-03 (weekly briefing) — decisions pending on Paul following the 399-term reconciliation: (a) adopt the single 399-term reconciled final glossary and retire the separate 201- and 440-term versions; (b) endorse the Core / Essential / Extended split and the asset-class lead order; (c) map the indicative Policy domains (POL-01..16) to governed bank policies; (d) confirm the planned completion sprint and RAG thresholds for the Programme Tracker against the real Day-1 deadline; (e) decide when to wire the tracker’s Power Query to live SharePoint, and whether/when to add a per-squad split or a Power BI exec tier [S-2026-07-03-weekly-briefing].

  • 2026-07-09 (reference-usage guidance) — which usage pattern the squad adopts for the Industry Business Glossary (A seed master / B ordinary co-source / C benchmark only), and at what confirmed Source Authority Tier (Tier 6 recommended pending SME review); T-05 must be extended/validated (REG-010…013; POL-01…16 mapping) before any hard tags carried by the file are accepted [S-2026-07-09-paul-glossary-reference-usage-guidance].

  • 2026-07-08 (C-QA) — squad adoption of the CC-1…CC-5 critical-check model as the formal “minimal criteria”; whether C-QA joins the Copilot Edition kit (e.g. as C-08) with regeneration on standards change; Copilot tenant calibration against the answer key before rollout [S-2026-07-08-paul-term-definition-qa-copilot-prompt].

  • 2026-07-10 (CR process) — confirm the requester submission channel (squad mailbox vs SharePoint CR folder); confirm the proposed ≤2-working-day triage SLA; decide whether the T-07 form and T-08 register join the master kit templates folder (and regenerate the MS365 Copilot edition if so) [S-2026-07-10-weekly-briefing].

  • 2026-07-10 (authoring-standards review) — decide whether to adopt the recommended D9 “Source Anchored” control (“where a formal regulatory, policy or process definition exists, the definition shall be traceable to its highest-authority source and the source recorded in glossary metadata”) into the authoring standard / Build Kit definition-quality set [S-2026-07-10-weekly-briefing].

  • 2026-07-12 (impact assessment) — whether the Impact Backlog stays a simple Data Steward log or joins the master kit as a T-09 Excel template (regenerating the MS365 Copilot edition if so); whether to propagate the BIF rename through existing kit documents now or at next revision; who owns detailed impact assessments referred outside the squad (BIF assigns per case in this design) [S-2026-07-12-paul-glossary-impact-assessment].

  • 2026-07-12 (TOR client update) — the client-applied TOR changes the SME availability assumption to ~5 days/week and removes the 399-term-adoption and D9 “Source Anchored” items from the pending list; whether the sizing model is re-run on the new SME assumption, and the status of the removed decisions (adopted / dropped / deferred), are not stated in the document [S-2026-07-12-paul-clm-glossary-squad-tor-client-update].

Connected wiki pages

History

  • 2026-06-01 — Project created; squad structure, sizing model, fast-track governance and delivery-method diagram captured from the leadership deck.

  • 2026-06-01 — Added the Jira-import-ready Excel backlog tracker for managing definitions across both squads’ sprints.

  • 2026-06-05 — Build-kit approach and work breakdown drafted (6 prompts, 7 skill.md, 6 templates, 1 guidelines doc); saved to 09 Taxonomy Tool\CLM_Glossary_Build_Kit_Approach.md; approved by Paul same day.

  • 2026-06-05 — Build kit built, QA-verified and delivered to 09 Taxonomy Tool\Glossary Build Kit\ (20 files incl. README index; guidelines as md + validated docx).

  • 2026-06-05 — Added annotated process-diagram deck (slide-3 flow + Prompt/skill/Template per stage + reference table) next to the guidelines, and a four-file synthetic test dataset (Word/Excel/PDF) with Answer Key covering all clustering scenarios, in a new Test Data folder.

  • 2026-06-07 — Build kit upgraded to v1.1 per independent QA review: all recommendations applied (one adapted on Paul’s instruction). Added two skills + Stage 7 skill/prompt, source-authority & definition-quality standards, expanded T-05 regulatory taxonomy, GR-BA & GR-12 controls, T-03/T-06 fields, extended test data (27 concepts) and a 7-stage diagram. Change report saved to 09 Taxonomy Tool.

  • 2026-06-23 — Created the MS365 Copilot edition: 7 merged single-upload files (C-01…C-07 as .md + .docx) bundling prompt + skills + guardrails + template schemas, a Copilot variant of the process-diagram deck, a README and a work-breakdown record, in a new sibling folder. Master kit and original deck left unchanged.

  • 2026-06-24 — Added MS365 Copilot editions of the three governance docs (Operating Guidelines, Operating Procedures, Task Checklist) in the edition’s guidelines\ subfolder, plus a governance-docs approach record. Master governance docs left unchanged.

  • 2026-06-27 — Delivered the industry reference Business Glossary (Day 1 + Day 2, Corporate/SME/Capital Markets primary + Consumer alignment) in the T-06 column model, tagged to the 7 regulatory cycle elements with relationships and Policy/Process/Regulatory tags; Excel + md + B&W exec PPTX + md + approach record saved to 03 Credit Lifecycle\Industry Business Glossary\. Provides the seed master content + a realistic term-count baseline for the mastering programme.

  • 2026-06-27 — Delivered the Programme Tracker (burndown & flow) — an Excel measurement/forecasting layer reading the build kit’s T-03/T-06 Summary counts into a Cumulative Flow Diagram and a velocity-forecast burndown; Excel-only, combined view, parameterised scope, history via Snapshot Log + Status Reconstruction. Workbook + fully-populated demo (8 sprints mock data) + Approach record saved to 09 Taxonomy Tool\ (demo in Programme Tracker Demo\).

  • 2026-06-27 (same day, extended) — Ran an asset-class coverage assessment vs a European/Irish retail-bank hierarchy (AIB/BOI style, mapped to CRR/CRR3 exposure classes; CLM_Asset_Class_Hierarchy_and_Coverage.xlsx): the 140-term core already covers all asset classes at the lifecycle level, so extended not rebuilt — added a 61-term Asset Class Extension wave (Retail mortgages/consumer/cards ~23, SME ~13, Corporate & CRE ~12, Capital Markets ~8, exposure-class anchors ~5). Master glossary now 201 terms (140 Core + 61 Extension) with a new Glossary Wave column; Excel/md and exec deck/md regenerated; REG-014..016 added (CBI Mortgage Measures, Mortgage/Consumer Credit Directive, specialised-lending slotting).

  • 2026-06-28 — Reconciled with an independent second perspective. Paul supplied a separately-produced 440-term glossary. Normalised matching + triage: 184 of 440 corroborated the 201; of 256 P2-unique, 198 included as net-new, 20 excluded as glossary operating-model “meta” terms, 38 excluded as duplicates (mapped). Recommended a single 399-term reconciled final glossary (provenance Both 151 / P1-only 50 / P2-added 198; Core 124 / Essential 201 / Extended 74; master-first = 325 Core+Essential). P2’s added value = CDE-level granularity (amounts/income/status events), data-governance/DQ controls, CCR/treasury & reporting mechanics. Deliverables CLM_Business_Glossary_Reconciled_Final.xlsx(+md, with a Reconciliation sheet + Source-Perspective & Inclusion-Rationale columns), B&W ..._Reconciled_Summary.pptx(+md), and reconciliation approach record — all in 03 Credit Lifecycle\Industry Business Glossary\.

  • 2026-07-05 — Delivered the Term & Definition Authoring Standards 2-slide pack (T1–T6 naming standard NEW; D1–D8 restating DQ-1…8 with ISO anchors); PPTX + md twin + approach record saved to 03 Credit Lifecycle\Glossary Standards\; recorded as source S-2026-07-05-paul-glossary-term-definition-standards.

  • 2026-07-08 — Delivered C-QA, the standalone MS365 Copilot standards-QA prompt (.md + validated .docx) with 16-row test input and expected-output answer key (xlsx + md twins), in new folder 09 Taxonomy Tool\Term and Definition QA (Copilot)\; grading = critical-check model (CC-1…CC-5), improvements GR-07 strict; recorded as source S-2026-07-08-paul-term-definition-qa-copilot-prompt.

  • 2026-07-09 — Delivered the reference-usage guidance (docx + md + approach record, 09 Taxonomy Tool\): how T-05 Regulatory/Policy/Process references drive P-04 hard/soft tagging without needing source files or online lookup, what a T-05 listing does and does not do, and a pros/cons + A/B/C usage-pattern assessment for feeding the Industry Business Glossary in as a source file; recorded as source S-2026-07-09-paul-glossary-reference-usage-guidance.

  • 2026-07-10 — Delivered the Change Request process (deck + md, T-07 Change Request Form, T-08 Change Request Register, approach record) in new folder 09 Taxonomy Tool\Glossary CR Process\; a minimal intake/tracking layer over Stage 7 covering single and bulk add/change/retire/other requests; recorded as source S-2026-07-10-paul-glossary-cr-process.

  • 2026-07-11 — Delivered the one-page squad Terms of Reference (md + docx) to 03 Credit Lifecycle\; three sections (Objectives / In-out scope / Considerations & dependencies) consolidating the approved squad design; recorded as source S-2026-07-11-paul-clm-glossary-squad-tor.

  • 2026-07-12 — Paul delivered a client-applied revision of the squad TOR (Word upload); recorded as source S-2026-07-12-paul-clm-glossary-squad-tor-client-update, superseding S-2026-07-11-paul-clm-glossary-squad-tor. Key applied changes: CM Business Impact Forum as escalation route, SME availability ~5 days/week, planning figures and the 399-term baseline removed, decisions moved to the squad, publishing Excel → SharePoint.

  • 2026-07-12 — Delivered the Step 4 Impact Assessment form + BIF routing sub-process (Word form + md, 3-slide deck + md, approach record) in new folder 09 Taxonomy Tool\Glossary Impact Assessment\; recorded the CLM Design Sub-Committee → Credit Monitoring Business Impact Forum (BIF) rename; recorded as source S-2026-07-12-paul-glossary-impact-assessment.

  • 2026-07-12 (same day, extended) — Added Glossary_Impact_Assessment_Exec_View.pptx, a 1-slide exec-level view of the same sub-process (flow with step name + actor only, BIF decision-outcome strip, single holding rule), on Paul’s request; detailed deck and form unchanged.

  • 2026-07-14 — Delivered the Business Glossary conceptual-layers diagram (1-slide B&W PPTX + md twin + approach record) to new folder 03 Credit Lifecycle\Glossary Conceptual Layers\; four bottom-to-top layers (Day 1 by process → Day 2 → Management Reporting → External data/loan tapes & IRB/supervisory regulatory consumers) with per-process term counts and prioritisation guidance; recorded as source S-2026-07-14-paul-glossary-conceptual-layers.

  • 2026-07-14 (same day, extended) — Added a client-entry variant CLM_Glossary_Conceptual_Layers_Client_Entry.pptx on Paul’s request: same layout, but the 201-term baseline counts and the “Scope baseline” panel removed and replaced with [ ] placeholders so client term counts can be entered per process box; the original baseline slide kept unchanged as the reference.

Sources / notes

Graph View

Backlinks

Recently updated

See 113 more →