kb-autofix: #304 — storm-impact-harmonisation: alerts attachment now a styled xlsx; second vendored file (xlsx_style.py)

[chd-ds-kb-steward]

Drafted by the KB steward (headless Claude, Max plan) from #304 and its comment thread. Review before merging — check it against the cited source. Re-comment the correction on #304 and the next run updates this PR.

---

🤖 What the steward changed

All edits are complete and consistent across both pages. The DESIGN.md hit is an unrelated decision-log entry (no vendoring/attachment fact there).

What changed

Both merged PRs verified against the public GitHub PRs before editing.

pipelines/storm-impact-harmonisation.md (facts from PR #11 → merge-cerf-exposure, and the two-file vendoring contract):

• Archive workbook identity columns — recorded that workbook.py's storm key is now atcf_id (NHC ATCF id, e.g. AL132025) on every tab, correcting the collision with the DB's slug-valued storms.nhc_storms.storm_id (melissa_2025); sources joined with | (CHD|GDACS|ADAM), was +; admin_pcode retained (the pcode rename was reverted). Body Steps + new extra.workbook_identity_columns.
• Vendoring contract now covers two files — fm_matching.py and style.py (→ src/xlsx_style.py in ds-storms-alerts, docstrings differ), with the "must stay code-identical" note and the third-file → de-vendor into a shared package threshold. Updated in downstream:, code_ref:, extra.three_source_comparison, the Failure-modes bullet, and Downstream consumers.

pipelines/storms-alerts.md (fact from PR #20 → main):

• Alert email attachment is now a styled per-storm xlsx (was flat CSV): tabs README/adm0_exposure/adm1_exposure/caveats, mirroring the archive workbook and sharing identity columns; by-design differences captured (forecast-based, per-storm per-issued_time, wide by wind threshold with MAX-across-sources), plus the src/xlsx_style.py vendoring pointer. Updated in outputs:, one-liner, Steps 7 & 9, and Outputs.

Why / source

• Align workbook columns with the alerts attachment and the DB ds-storm-impact-harmonisation#11 (into merge-cerf-exposure) — confirmed storm_id→atcf_id on every tab, +→|, admin_pcode retained.
• Attach a styled per-storm Excel workbook aligned with the archive workbook ds-storms-alerts#20 (into main) — confirmed CSV→styled xlsx (README/adm0_exposure/adm1_exposure/caveats) and the vendored xlsx_style.py (code-identical, docstrings differ).

Kept the full vendoring contract on the harmonisation page (its src/source_exposure/ is the source of truth) and left only a pointer + the styling reference on the alerts page, per one-home-per-fact.

Please check

• Branch pinning: storms-alerts.md is pinned to source_branch: adm1-exposure-csv (de38cb5); PR KB drift: hub pages with moved spokes #20 merged to main, which is what the live Databricks job runs (already noted as a [conflict] on the page). I described current runtime behaviour (xlsx) but did not re-pin source_sha/source_branch — a full re-sync of that page is a separate drift-bot job, not this issue's scope.
• I did not run any index generators; the frontmatter downstream edits are minor and the scheduled generators will pick them up.

Closes #304