Source:
../architecture-review.html· Type: Architecture review · Audience: Staff engineers and architecture board · Mode: Live + Q&A · Time: 25 min + 10 Q&A Fictional case. The deck is the current implementation; this file captures the brief, structure, and acceptance.
Staff engineers and the architecture board.
Intent: Approve phase 1 of the streaming migration.
Decision or action: Approve phase 1; fund two engineers and a game day.
Success: The board can state the trade-off and the one-hour rollback.
Context → requirements → options → trade-offs → recommendation → migration → risks → decision.
| # | Slide | Budget | Purpose / speaker cue |
|---|---|---|---|
| 1 | Streaming event ingestion at Meridian | 1 | State the decision first: this is a review, not a survey. |
| 2 | Context | 2 | The freshness requirement is a product constraint, not an engineering preference. |
| 3 | Requirements | 2 | Four constraints. |
| 4 | Options | 3 | Name the options fairly. |
| 5 | Trade-offs | 3 | The trade-off is not freshness versus cost; it is freshness versus on-call maturity. |
| 6 | Recommendation | 2 | One recommendation. |
| 7 | Migration | 3 | The rollback point is P3. |
| 8 | Risks | 2 | Risks are operational. |
| 9 | Decision | 2 | Ask for phase 1 only. |
| 10 | Appendix | 10 | Q&A. |
2.4B events/day, six-hour freshness, cost ceiling, 90-day replay. Metrics are fictional; validate against current telemetry and cost data.
Fact-check gate: confirm names, numbers, dates, and comparisons before layout. A precise number without a date and source is a liability.
Technical and precise; define exactly-once and replay; no vendor marketing.
Dark blue, cyan accent; one table or diagram per slide; monospace for constraints. Cover uses an engineering review photograph, not a data-center rack.
25 minutes plus 10 Q&A. Must-cut: option C detail and vendor comparison; protect trade-offs, migration, risks, and the decision. (total budget: 30 minutes)
Q&A: prepare the ten most likely questions and one honest answer for each. Do not improvise a fact.
Risks: Straw-man options; ignoring on-call load; no rollback.
Open: Replace fictional metrics; confirm vendor constraints; add the real topology diagram.