Architecture Fitness Functions Catalogue Template

Catalogue of executable architecture checks per quality attribute: type, metric, threshold, tool, cadence, owner, result

Kiến trúc & Thiết kế hệ thốngChuẩnBuilding Evolutionary Architectures (fitness function concept)fitness functionsCập nhật 20 thg 10, 2026

Mở trong editorTải .md

Cách dùng: Use to protect architectural characteristics as the system evolves. List the quality attributes first, then define one objective, automated check with a threshold for each. The main pitfall is vague checks or thresholds without a recorded rationale, which get disabled later.

Xem trước

Architecture Fitness Functions Catalogue: [System or product]

Catalogue of architecture fitness functions for [System]: the quality attributes that matter, an objective check for each, how it is classified, the metric and threshold, where it is implemented, how often it runs, who owns it and its current result. Based on the fitness function concept from Building Evolutionary Architectures (Neal Ford, Rebecca Parsons, Patrick Kua, and in the second edition Pramod Sadalage). Replace every [bracketed] value and every row marked Example.

Field	Value
Catalogue ID	[FF-SYSTEM]
System	[Name and scope]
Version	[0.1]
Owner	[@architect]
Contributors	[Teams]
Last reviewed	YYYY-MM-DD
Related	[Architecture description], [Quality attribute scenarios], [Tech debt register]

1. Purpose and concept

A fitness function is an objective measure of how well the system meets one architectural characteristic. Together, the functions in this catalogue act as an executable description of the architecture: when a change would erode a property we care about, a check fails early and the team decides consciously whether to fix the change or revise the target. The concept comes from evolutionary computing, where a fitness function scores how close a candidate solution is to the goal.

2. Quality attributes in scope

ID	Quality attribute	Why it matters	Priority
QA-01	Example: Modularity (layering and bounded contexts)	Teams release independently; coupling slows every change	High
QA-02	Example: Performance (API latency)	Checkout abandonment rises above 1 s response time	High
QA-03	Example: Security (dependency hygiene)	Regulator requires patching critical CVEs within 14 days	High
QA-04	Example: Availability	Contractual 99.9% monthly SLA	High
QA-05	Example: Deployability	Target: deploy any service on any working day in < 30 min	Medium
QA-06	Example: Cost efficiency	Unit cost must fall as volume grows	Medium
[QA-nn]	[Attribute]	[Reason]	[H/M/L]

3. Classification dimensions

Dimension	Values	Meaning
Scope	Atomic / Holistic	Atomic checks one characteristic in isolation; holistic checks how several interact (for example performance under a security control)
Trigger	Triggered / Continual	Triggered runs on an event such as a commit or a release; continual runs all the time, typically in production monitoring
Result	Static / Dynamic	Static compares with a fixed threshold; dynamic threshold depends on context such as load or time of day
Execution	Automated / Manual	Manual checks (for example a legal review) are allowed but should be rare and scheduled
Timing	Temporal	Runs on a calendar, for example "upgrade check every quarter"
Intent	Intentional / Emergent	Intentional checks are defined up front; emergent ones are added after the system reveals a new risk

4. Fitness function catalogue

ID	Quality attribute	Fitness function definition	Type	Metric and threshold	Implementation	Cadence	Owner	Current result
FF-01	QA-01 Modularity	Example: Classes in `domain` must not depend on `infrastructure` or `web` packages	Atomic, triggered, static, automated	0 violations	Architecture unit test (ArchUnit) in build	Every commit	[@team-core]	Pass, 0 violations
FF-02	QA-01 Modularity	Example: No cycles between modules	Atomic, triggered, static, automated	0 cycles	dependency-cruiser rule in CI	Every commit	[@team-web]	Pass
FF-03	QA-02 Performance	Example: p95 latency of `/checkout` under 300 RPS load	Atomic, triggered, static, automated	p95 <= 400 ms	k6 load test in staging pipeline	Every release candidate	[@team-checkout]	362 ms, pass
FF-04	QA-03 Security	Example: No dependency with a known critical vulnerability older than 14 days	Atomic, triggered and temporal, static, automated	0 critical findings > 14 days	Software composition analysis in CI plus nightly scan	Every commit and nightly	[@security]	1 finding, 6 days old, pass
FF-05	QA-04 Availability	Example: Monthly availability of public API from synthetic probes	Atomic, continual, static, automated	>= 99.9% per month	Synthetic checks every minute from 3 regions	Continual	[@sre]	99.95% last month, pass
FF-06	QA-02 and QA-03	Example: p95 latency stays within budget with WAF and mTLS enabled	Holistic, triggered, static, automated	p95 <= 400 ms with all security controls on	Load test in production-like environment	Monthly	[@sre]	391 ms, pass (near threshold)
FF-07	QA-05 Deployability	Example: Lead time from merge to production	Atomic, continual, dynamic, automated	Median < 30 min; alert if 7-day median rises 50% above the 90-day median	Pipeline metrics dashboard	Continual	[@platform]	22 min median, pass
FF-08	QA-06 Cost	Example: Infrastructure cost per 1,000 orders	Holistic, temporal, dynamic, automated	Must not rise for two consecutive months	Cost export and query	Monthly	[@finops]	Fell 4%, pass
FF-09	QA-03 Security	Example: Annual review of data flows against records of processing	Atomic, temporal, static, manual	Review completed and signed off	Privacy review checklist	Yearly	[@dpo]	Due YYYY-MM-DD
[FF-nn]	[QA-nn]	[Precise definition]	[Scope, trigger, result, execution]	[Metric and threshold]	[Test, tool, query]	[Cadence]	[@name]	[Result and date]

4.1 Threshold rationale

Record why each threshold has its value, so it can be revisited.

ID	Threshold	Rationale	Source
FF-03	p95 <= 400 ms	Example: Leaves 600 ms of the 1 s page budget for front end and network	Performance budget, [link]
FF-04	14 days	Example: Regulatory patching deadline	[Policy reference]
[FF-nn]	[Threshold]	[Reason]	[Source]

5. Implementation guidelines

Store fitness function code in the same repository as the system it checks, under [path, e.g. tests/architecture/], and review it like any other change.
Each failing check prints the function ID, the threshold, the measured value and a link to this catalogue.
Pipeline position: structural checks block merge; performance and holistic checks block promotion to production; continual checks raise alerts routed to the owner.
New quality attribute scenarios must propose at least one fitness function or explain why one is not possible.
Disabling a check needs an entry in section 6 with an expiry date.

6. Exceptions and waivers

ID	Function	Waiver	Reason	Approved by	Expires
WV-01	Example: FF-02	Allow one known cycle between `billing` and `invoicing`	Being removed by TD-017	[@architect]	YYYY-MM-DD
[WV-nn]	[FF-nn]	[Change]	[Reason]	[@name]	YYYY-MM-DD

7. Review and results history

Review date	Functions	Passing	Near threshold	Failing	Changes made
YYYY-MM-DD	Example: 9	8	1 (FF-06)	0	Added FF-08 cost check; tightened FF-03 from 500 ms to 400 ms
YYYY-MM-DD	[n]	[n]	[IDs]	[IDs]	[Changes]

Actions from the latest review: [list with owners]. Example: profile the WAF rule set to restore headroom on FF-06 ([@sre], YYYY-MM-DD).

Standard and licence note

Based on the architecture fitness function concept described in Building Evolutionary Architectures by Neal Ford, Rebecca Parsons and Patrick Kua (O'Reilly, 2017; second edition with Pramod Sadalage, 2023). The classification dimensions are summarised in original words; no text from the book is reproduced. Tool names are examples only. Template licence: MIT.

Template liên quan

Kiến trúc & Thiết kế hệ thống