Stop 10 of 10 · Weekly
Fadi's quarter and the capstone
Three scheduled routines carry the quarter, and one agency goes end to end from discovery to proof of concept, graded out of a hundred against a rubric written into this stop.
Three routines, one quarter
Fadi covers six accounts. The work that slips is never the deadline work, it is the reading that would have caught the problem eight weeks earlier. Three routines carry it, and none of them changes anything.
| Routine | When | What it produces |
|---|---|---|
| Pipeline digest | Monday morning | Each account, what moved, what is overdue, which sandbox environments are past their destroy date |
| Compliance-change watch | Weekly | Published changes to the frames on the accounts, each with a citation and the accounts it touches |
| Cost-model refresh | Monthly | Each model re-run against current public rates, flagging any total that moved over five percent |
Every Monday, using Research mode, check the published sources for the compliance frames on my state and local accounts: FedRAMP, StateRAMP, the CJIS Security Policy and the NIST control catalogue they draw on. Report only changes published since your last run: what changed, the document and version, the date, and a link. For each change, list which of my accounts uses that frame and whether it touches a control my mappings already mark as a gap. Change nothing. Do not update a mapping, do not rewrite a document, and do not tell me a change applies to an account until I have confirmed which version that agency is assessed against.
The last paragraph is not decoration. A routine that edits is a routine that edits weekly whether or not anyone read it.
The capstone
One real agency, one quarter, end to end. Six artefacts and the evidence behind them.
The architecture document with its diagram and decision log. The compliance mapping with the responsibility split and an honest gap list. The runbook with its rollback section and a rehearsal transcript. The cost model with its assumption table and sensitivity ranking. The proof of concept: brief, code, plan output and confirmed teardown. The evaluation run: thirty questions, the graded result and the changes it drove.
[upload the six capstone artefacts: architecture document with diagram and decision log, compliance mapping, runbook with rehearsal transcript, cost model, proof-of-concept brief with plan and teardown evidence, and the graded evaluation run] I am the cloud architect on this account. Grade this against the rubric below, out of 100, scoring each criterion separately and quoting the evidence you used. For every point not awarded, say what is missing and what would earn it. Be strict about unsourced claims, about any figure that did not come from the public calculator, and about infrastructure applied without a plan I read first.
The rubric
| Criterion | Points | Full marks means |
|---|---|---|
| Data integrity | Gate | No customer data, credential, unreleased price or real network detail anywhere in any transcript. A breach fails the capstone whatever else scored |
| Sourced claims | 20 | Every capability, control and compliance claim carries the document and version it came from, and nothing reads NOT VERIFIED in a customer-facing file |
| Decisions defensible | 15 | Every architectural choice names the alternative it beat, the consequence accepted and the revisit trigger |
| Mapping honesty | 15 | Every control row states the responsibility split, and every unevidenced control is marked a gap rather than assumed |
| Runbook usable at 02:40 | 15 | Steps name owner, action, expected result and duration, with go criteria as numbers and a rollback section that names the point of no return |
| Model you can defend | 15 | Assumptions sourced and marked, rates typed from the public calculator with the date, sensitivity ranked, levers separated from assumptions |
| Build discipline | 10 | Brief before code, plan read before apply, sandbox only, teardown confirmed from the billing view |
| Measured, not assumed | 10 | Thirty questions with expected answers written first, three axes graded, and fixes landed in the Project or library rather than in a prompt |
Eighty passes. Under eighty, the gap is nearly always the same two: a claim with no document behind it, and a runbook step nobody could follow in the dark.