Reversibility measures how readily an architectural decision can change without disproportionate cost, risk, or disruption. It makes consequence, lifespan, ownership, and the cost of reversal explicit rather than treating lighter choices as automatically better. section 3.2
Use reversibility to calibrate rigor and preserve useful paths for change:
- Accept lighter choices when scope is narrow, ownership is clear, risk is low, and reversal remains inexpensive. section 3.2
- Increase rigor when systems become strategic, shared, audited, long-lived, or difficult to change safely. section 3.2
- Use milestones to deliver value while testing assumptions, validating integrations, and exposing operational concerns. section 3.2
An evidence-producing first slice
Deliver one workflow, one integration, one deployment path, one monitoring approach, and one support model before building the full target platform. Strengthen, generalize, or separate components only when evidence justifies the cost. section 3.2
Table 3.3 Good-enough choices in practice
|
Decision
area
|
A lighter
choice may be enough when...
|
More rigor
is justified when...
|
|
Modular
monolith vs. services
|
One team
owns the system, domain boundaries are still emerging, deployment
independence is not needed, and operational maturity is limited.
|
Different
teams need independent ownership, scaling needs diverge, release coordination
slows delivery, or service boundaries are stable enough to justify
distribution.
|
|
API design
|
The API is
short-lived, internal, used by one known consumer, and easy to change with
direct coordination.
|
The API is
public, strategic, long-lived, consumed by multiple teams or external
parties, or requires versioning, documentation, compatibility guarantees, and
stronger governance.
|
|
Infrastructure
automation
|
The change
is temporary, non-production, one-off, low risk, and unlikely to be repeated
or handed over.
|
The
infrastructure is production-facing, repeated across environments, audited,
security-sensitive, operated by more than one team, or expected to live
beyond the initial delivery phase.
|
Compare when modular monoliths, lightweight APIs, or limited automation are sufficient and when stronger rigor is justified. section 3.2
