Whole-wiki review, October 2026¶
A review of all 79 topic pages, the 9 section pages and the README, done after sections 08 and 09 were added.
How it was done¶
- Automated checks: frontmatter, reading order, prerequisites, links, banned markup and filler words, prose length.
- Seven section reviews, each by a reviewer that had not written the pages: accuracy, clarity, examples, diagrams, sources.
- Two cross-cutting reviews: consistency of the shared shop example across pages, and whether each learning path can be read in order.
- 21 canonical decisions written down where pages disagreed (see "Shared example world" in GUIDELINES).
- Fixes applied section by section against those decisions, then cross-page fixes, then all checks rerun.
Findings¶
| Review | Must-fix | Should-fix | Nice-to-have |
|---|---|---|---|
| 01 Foundations, 02 Storage | 1 | 10 | 9 |
| 03 Ingestion | 1 | 12 | 8 |
| 04 Modelling | 1 | 10 | 7 |
| 05 Governance | 4 | 14 | 10 |
| 06 Serving | 0 | 11 | 13 |
| 07 Operations | 2 | 10 | 12 |
| 08 and 09 | 1 | 10 | 12 |
| Consistency | 3 | 17 | 11 |
| Learning paths | 4 | 19 | 9 |
About 190 findings were applied. About 20 were skipped, either because they conflicted with a canonical decision or because the page already said the right thing.
Main changes¶
Shared example: - raw_orders is fed only by CDC through the Iceberg sink: four tasks, a commit every 5 minutes, about 1,150 files a day. - clean_orders and clean_order_items are built hourly; fct_daily_sales nightly, ready by 07:00 Europe/Berlin. - customer_key is the surrogate key; customer_pseudo_id is the keyed hash. The golden record lives in dim_customer. - One set of example IDs (o-1001, o-1002, c-88, c-91, p-77), all dates in 2026, one SLO (98% of days over 90 days), one set of retention periods.
Accuracy: - dbt licensing: v1 and dbt-oss are Apache 2.0; the default v2 package is proprietary. - Terraform and OpenTofu licences and dates; Iceberg sink CDC operation codes; Kafka 4.x status. - EDPS v SRB (C-413/23 P) wording on pseudonymised data; transfers to a separate US company. - A fairness example that used precision where recall was meant. - Observability: error budget recalculated for the 98% objective.
Structure: - Event design moved before the optional Ingestion pages (now 3.07). - Core path widened to 27 pages so that every prerequisite is on it; the analyst path now includes BI and dashboards. - Section intros and "Before starting" lines match the pages they introduce. - Glossary and further reading regenerated (489 terms).
Final state¶
- check_wiki.py: 79 pages and 9 section pages, 0 errors, 0 warnings.
- lint_wiki.py: no markup or filler findings; 4 pages between 850 and 950 words of prose, kept on purpose.
- All 79 Mermaid diagrams render.
Suggested spot-checks¶
A human read of these pages covers each section and the most-changed content:
- iceberg-maintenance (storage figures, recently changed)
- kafka-to-iceberg (longest Ingestion page, CDC details)
- deduplication-and-merge (the core MERGE example)
- privacy-lifecycle (erasure example, legal wording)
- encryption-and-key-management (new page, column keys)
- observability-and-slos (SLO and error budget numbers)
- bi-and-dashboards (new page, longest in the wiki)
- responsible-ai-governance (new page, fairness example)