"Week 5 - Seeing How the Business Actually Works" - @practicaldatamodeling
Why this is in the vault
Fifth post in the Mixed Model Arts book-club re-read (Chapter 13, "Seeing the Business"): a short discussion-prompt issue (full body rendered, no paywall) whose one transferable idea is that you must trace the real process, workarounds included, before a model or metric is trustworthy.
The core argument
Reis argues a technically correct model can still get the business wrong, because a process crosses teams and a word's meaning shifts at each boundary. Official documentation helps, but the person running the workflow daily (and their spreadsheet) reveals what the architecture diagram omits. The running retailer example adds conflicting goals: Support wants faster resolution, Finance wants lower support cost, Product wants recurring-defect detection. An agent closing a ticket can look like success to one team while the customer still awaits a replacement. His stance: follow one complaint end to end before adding any metric, and do not expect a single status field to settle every disagreement.
The issue is mostly logistics and prompts: three discussion questions (where the process starts and ends, what evidence to demand for "resolved," what to capture when a workaround contradicts documented process), a ten-minute optional sketch exercise, live discussion for paid subscribers only, a Thanksgiving pause, and Week 6 resuming in December.
Mapping against Ray Data Co
Most concrete connection: the "closed vs. actually done" gap is the same failure mode catalogued in feedback_workflow_agent_output_integrity (a sub-agent or workflow writes done/verified and different readers take it to mean different things), and Reis's remedy - trace one item through every handoff and verify with the process owner - is the same discipline as the vault's fresh-eyes gate on RDCO's own multi-stage chains. Concretely, it argues for tracing one newsletter-to-note-to-graph item end to end across process-newsletter, the audit script and self-review, and asking at each boundary what "filed" means.
Second hit: the "follow the workaround, ask the person doing the work" method applies to the phData Org Map / OI work, where documented org structure and actual working relationships diverge; interview-the-operator beats reading the chart. This is a reinforcement, not a new decision. Mapping is medium: valuable framing, no direct action required, and the live discussion is paid-only (sub lapsed), so only the public prompts are usable.
Related
- [[2026-11-08-practical-data-modeling-mma-book-club-week4-chapters-10-12]]
- [[2026-10-29-practical-data-modeling-mma-book-club-week3-chapters-7-9]]
- [[2026-10-23-practical-data-modeling-mma-book-club-week2-chapters-4-6]]
- [[2026-10-16-practical-data-modeling-mma-book-club-week1-chapters-1-3]]
- [[2026-04-12-practical-data-modeling-mma-ch13-seeing-the-business]]
- [[2026-04-03-practical-data-modeling-mma-ch11-semantics]]
- [[feedback_workflow_agent_output_integrity]]