Capstone: design review · Prasanta Kumar Behera

Assemble every artifact from the track into ParcelPath's design document, trace each NFR to the design and its evidence, and prepare to defend it in front of a review board: the structure, the questions reviewers ask, and a rubric to self-assess.

Assemble the document

Everything you produced in this track becomes one design document for ParcelPath:

Section	Source lesson(s)
Requirements: NFR and SLO sheet, RPO/RTO	01, 10
Platform choice, PoC and stability evidence	02, 03
Sizing sheet and growth triggers	04
Cluster topology, node pools (ADR-001/002)	05
IPAM plan, CNI and routing (ADR-003)	06
Load balancing, ingress, TLS (ADR-004)	07
Edge, WAF, tenancy (ADR-005)	08
Stateful tier (ADR-006)	09
Observability, SLOs, alert catalogue	10
FMEA, chaos results	11
DR design, runbook, drill results (ADR-007)	12

Keep the main document short (15–25 pages), with details in appendices and linked artifacts.

This is the science-fair day. You've done all the experiments; now you put them on one poster that tells the story: the question, what you built, how you tested it, and what you're still unsure about. Then judges ask tough questions, and you answer with your notes, honestly, including "we tried that, and here's what happened".

NFR traceability

NFR	Design answer	ADR	Evidence	Status
NFR-01 99.95%	3 racks, N-1 capacity, HA ingress, active-passive DR	001, 004, 007	Soak + rack-loss chaos (99.97%)	✅
NFR-02 p99 < 300 ms	Native routing, local NVMe DBs, caching	003, 006	Peak test p99 212 ms	✅
NFR-03 peak 3,000 req/s	Sizing sheet (15 workers at 18 months)	—	Peak/burst tests	✅ (re-test at scale)
NFR-04 RPO 5 min / RTO 60 min	Async replica cluster + runbook	006, 007	DR-2: RPO 22 s, RTO 47 min	✅
NFR-06 TLS + encryption	WAF → ingress TLS, encrypted volumes	004	Config audit	⚠️ in-cluster mTLS phase 2
NFR-07 upgrades	Rolling upgrades with PDBs	001	Upgrade under load, 0 failed requests	✅

Gaps are fine if they're visible, with an owner and a date.

Questions reviewers will ask

Failure: what happens if a rack, a site, etcd, DNS, the WAF, GitOps, the identity provider or the monitoring stack fails? (Point at the FMEA row.)
Scale: what breaks first at 2× and 5×? What are the triggers to act?
Operations: who is on call for what? How long does a node replacement, an upgrade, a restore take (measured)?
Security: how are tenants isolated? Where are secrets? How is supply chain verified?
Cost: cost per month now and at 18 months; biggest cost drivers; what you'd cut first.
Alternatives: why not a stretched cluster? Why not managed Kubernetes in the cloud? Why not active-active?

Answer with evidence and trade-offs, and say "we don't know yet; here's how we'll find out" when that's true.

Self-assessment rubric

Area	Weak	Strong
Requirements	Vague goals	Measurable NFRs/SLOs with owners
Decisions	Undocumented	ADRs with context, alternatives, consequences
Evidence	Assertions	Tests, chaos results, drills with numbers
Failure thinking	Happy path only	FMEA, DR runbook, honest residual risks
Operability	Build-focused	Runbooks, alert catalogue, upgrade and restore procedures
Communication	Wall of text	Clear summary, diagrams, traceability, open questions

Try it: your design review

Assemble your artifacts from lessons 01–12 into a document with the structure in the cheat sheet.
Build the NFR traceability table and mark every gap honestly.
Write a one-page summary and one architecture diagram that fits on a slide.
Ask a colleague to play the review board using the question list; note questions you couldn't answer.
Update the risk register and open questions with owners and dates.

Going deeper: after approval

The design document becomes a living reference: update it (and supersede ADRs) as the platform evolves.
Turn the design into an implementation plan with milestones tied to NFR evidence (e.g. "DR drill passed" as a go-live gate).
Run a post-go-live review after the first peak season: which assumptions held, which didn't?

Recap

One design document assembles NFRs, evidence, sizing, topology, network, traffic, storage, security, observability, resilience, DR and cost.
Traceability: every NFR → design → ADR → evidence → status; gaps visible.
Prepare for failure, scale, operations, security, cost and alternatives questions; answer with evidence and trade-offs.
Use the rubric to self-assess; keep the document alive after approval.

This site is a public version of my personal engineering knowledge hub. It intentionally excludes confidential company information and internal operational details.