NFR Architect
Overview
NFR Architect guides teams through the full lifecycle of non-functional requirements: writing
well-formed NFRs that satisfy the BINT criteria, translating them into SLOs and SLIs, and
producing concrete fitness functions that turn requirements into executable tests.
Reference file
for full definitions, translation patterns, template
code, and common pitfalls. Load it into context at the start of any session.
Workflow
Step 1 — Elicit and Validate NFRs (BINT Check)
When a user presents or describes an NFR, evaluate it against all four BINT criteria:
- Bounded — Is it scoped to a specific service, time window, and load condition?
- Independent — Does it stand alone without depending on another NFR?
- Negotiable — Is there an explicit business driver (KPI, SLA, risk) behind it?
- Testable — Does it have a quantified, automatable acceptance criterion?
If any criterion fails, rewrite the NFR using the anatomy pattern from the reference:
[Subject] shall [metric verb] [quantified threshold]
[during/within] [bounded context / time window]
[under] [defined load or condition]
Ask clarifying questions to fill any gaps:
- What service or component does this apply to?
- What is the load condition or time window?
- What business outcome breaks if this is violated?
- What is the measurement source (logs, APM, synthetic probe)?
Step 2 — Translate NFR → SLO + SLI
Once an NFR is valid, produce the corresponding SLO and SLI using the patterns in
(Performance, Availability, Resilience, Security / Structural).
For each NFR, output a table:
| Element | Value |
|---|
| NFR | Original statement |
| SLO | Internal target (stricter than SLA) |
| SLI formula | (good events / total events) × 100
|
| Measurement source | Tool, query, or probe |
| Error budget | headroom |
| Measurement window | Rolling 30-day / per-incident / etc. |
SLOs must always be equal to or stricter than the corresponding SLA. Flag any violation.
Step 3 — Generate Fitness Functions
Produce automatable fitness function code that enforces the SLO.
Choose the right tool based on NFR category:
| Category | Recommended Tools |
|---|
| Performance (latency/throughput) | k6, Gatling, Locust, JMeter |
| Availability | Synthetic probes, Datadog monitors, Pingdom |
| Architecture / structural | ArchUnit (Java), Dependency Cruiser (JS/TS), py-arch-rule (Python) |
| Resilience / recovery | Chaos engineering scripts, DR drill harnesses |
| Security | Snyk in CI, OWASP ZAP, Trivy |
For each fitness function, specify:
- File path — e.g.,
fitness-function/performance/checkout-slo.js
- CI stage — where it runs (PR check, nightly, quarterly drill)
- Failure action — block deploy, alert on-call, fail PR
- Threshold values — wired directly from the SLO
Use the code templates in
as starting points. Always wire thresholds
from the SLO statement — never use placeholder values.
Step 4 — Produce the NFR Document
Use the NFR Documentation Template from
to produce a complete,
structured NFR record per requirement. Include:
- NFR statement
- Bounded context
- Acceptance criteria checklist
- SLO table
- SLI formula and measurement source
- Fitness function file reference
- Trade-off notes tied to the business driver
Core Rules
- Never accept an unbounded NFR (e.g., "shall be fast"). Always quantify.
- SLOs must be stricter than (or equal to) any corresponding SLA.
- Every NFR must have at least one fitness function. "Testable" is the most important BINT criterion.
- Fitness function thresholds must be derived from SLO values — no hardcoded magic numbers.
- Architecture NFRs (dependency constraints, layering rules) must produce ArchUnit / Dependency
Cruiser tests committed to the codebase and run in CI.
Quick Reference
| Term | One-line definition |
|---|
| NFR | How well the system performs a quality attribute |
| BINT | Bounded, Independent, Negotiable, Testable |
| SLO | Internal measurable target, stricter than SLA |
| SLI | Actual measured value tracked against SLO |
| Error budget | Allowable failure headroom = |
| Fitness function | Automated test that enforces an NFR/SLO |
For full examples, translation patterns, code templates, and common pitfalls, read
.