CLM Taxonomy Dashboard v2 rebuild — session record

Tag: S-2026-07-26-clm-dashboard-v2-rebuild Type: own-writing Author(s): Paul Miles / Claude (Cowork session) Date of source: 2026-07-26 Date ingested: 2026-07-26 Authority weight: high — first-party work record with executed-check evidence for every technical claim. Raw file: /_raw_sources/S-2026-07-26-clm-dashboard-v2-rebuild.md

What it claims

The CLM Taxonomy Dashboard was rebuilt as CLM_Taxonomy_Dashboard_v2.html on the 440-term Second Perspective v2 glossary, replacing the 201-term wireframe dataset (v1 file retained). The data layer — converter, reference extractor, integrity validator and CSV export module — was produced by a four-task Ringer swarm (clm-data-layer-20260726T075830Z-p26961, 4/4 PASS, two tasks rescued on retry), each artifact proved by an executed check against the real workbook and independently re-verified in a second environment. The UI rebuild itself was done directly, not swarmed, per the earlier not-swarm-shaped decision.

Real data now drives definitions, domains, regulatory/process/policy tags, the 10-regulation anchor set (with URLs), the 23-tag governed taxonomy and definition basis. Because the source workbook carries a single status and no owners, workflow screens use a deterministic, banner-flagged demo overlay computed from Term ID at render time (survives refresh; ON/OFF toggle); approve / request-change remain demo stubs. The “Coming next (taxonomy roadmap)” drawer footer was removed; roadmap tabs retained as flagged wireframes.

Refresh is two-path: a governed refresh.sh (converters → validate_glossary.py hard gate → rebuilt HTML; proven end-to-end) and an in-browser session-only workbook load with a light integrity gate. Verification: integrity gate clean (440 terms, 10 regs) and a 33/33-check Node smoke test across all views, personas, drawer, overlay determinism and exports. Ringer’s agent integration (install-agent) is installed; the local model scoreboard shows GPT-5.6 Sol proven on probe tasks and GLM-5.2 on probation with 100% first-try on data-pipeline work.

Notable quotes

“OK: 440 terms, ids unique and well-formed, all required keys present as strings, per-domain counts reconcile with the Coverage Matrix, names and definitions verbatim.” (convert-440 executed check output, run log)

What’s speculative vs. asserted

Asserted with executed evidence: all PASS verdicts, check outputs, smoke-test results, refresh-path proof. Asserted as decision: overlay approach, stubs, v2 file placement. Speculative: none — cost figure “~1c/task” for the OpenCode lane is an estimate from token counts and OpenRouter prices, not a billed amount.

Topics this feeds

Open questions raised

  • When real workflow columns (status/owner) will exist in the workbook so the overlay can be retired.
  • Hosting choice for the Teams Website tab (SharePoint page vs other static host).