Notes
Notes
The depth lives here, not on the pages that sell the work. Each note makes one argument, shows the arithmetic, and names where the number is weakest.
Counts are generated from the notes themselves.
Published
Everything published, start-here first
START
A cost model that ties out exactly is hiding something
Why a zero residual is a red flag, and what a reconciliation harness has to catch.
START
Per-tenant AI cost is a join, not a report
Where to stamp the tenant dimension, and how to price caching, routing and GPU waste.
03
The access appendix, in full
Exactly which roles to create, what is deliberately absent, and what stands in place of a SOC 2 report.
04
When not to hire a cost mechanic
Whether to write the email at all, before you spend a call finding out.
05
What FOCUS normalizes, and the five things it does not
Which export to read per cloud, which column carries amortized cost, and what never backfills.
06
The node is the billing unit, the namespace is the ownership unit
How to split cluster idle into two owned numbers, and how to pick a shared-cost rule.
Topics
Where the depth is, and is not
| Topic | Notes | Weight |
|---|---|---|
| method | 6 | |
| AWS | 5 | |
| Azure | 5 | |
| GCP | 5 | |
| AI and GPU | 1 | |
| Kubernetes | 1 | |
| 6 topics | 6 | notes |
Sourcing
How the sourcing works
Every statistic carries a numbered source link. Figures we judge medium or low confidence say MEDIUM CONFIDENCE or LOW CONFIDENCE next to the number. Every worked example is invented and labelled SYNTHETIC.
The derivations carry the queries themselves.