- In short
- Mapping business requirements to solution architecture links each business goal to the functional behaviour that serves it and the infrastructure needed to deliver that behaviour. Every architecture decision stays traceable to a requirement, which keeps scope disciplined, and conflicting requirements such as cost versus latency are resolved explicitly against stated business priorities.
Mapping requirements to architecture: joining the two halves
This knowledge point is where the two earlier pieces of task statement 2.1 come together. You have named what the system must do as functional requirements, and you have captured how well it must perform as infrastructure and non-functional requirements. Mapping is the act of turning those into a coherent solution architecture, where each business goal is linked to the behaviour that serves it and the infrastructure that delivers that behaviour. On the Claude Certified Developer - Foundations (CCDV-F) exam this is an apply-level skill: you are not just recognising good requirements, you are building the chain from goal to design and judging whether a proposed architecture holds up.
The reason this deserves its own knowledge point is that a pile of good requirements is not yet an architecture. Requirements describe the target; the mapping produces the design that hits it and, crucially, records why each part of the design exists. That record is what separates a disciplined solution from an accumulation of components someone thought seemed reasonable.
- Requirements-to-architecture mapping
- The linkage from each business goal to the functional behaviour that serves it and the infrastructure needed to deliver that behaviour, such that every architecture decision is traceable to a requirement and every requirement is served by some part of the design.
The mapping is a chain from goal to infrastructure
The core of the mapping is a chain. A business goal, say reduce contract review time, connects to one or more functional behaviours, extract key clauses and flag unusual terms, and each behaviour connects to the infrastructure that delivers it under its non-functional targets, a model sized to the task, an API mode, and a caching strategy that meets the latency and cost budgets. Read forward, the chain shows how a business goal becomes a running system. Read backward, it shows why every component exists.
Building this chain well is the apply skill. It is not enough to know the categories; you have to connect a specific goal to specific behaviour to specific infrastructure, and defend the connections. When a stakeholder asks why the design uses the batch API, the answer is not "it is cheaper" but "the throughput-and-cost target on this overnight job, which serves the goal of a morning compliance report, points to batch." The mapping turns opinions about the design into justifications rooted in requirements.
Traceability keeps scope disciplined
The property that makes a mapping trustworthy is traceability: every architecture decision can be traced back to a requirement, and every requirement can be traced forward to some part of the design. Traceability is not paperwork; it is the mechanism that keeps scope honest in both directions. Trace forward from requirements and you find gaps, a stated need that nothing in the design serves. Trace backward from the design and you find scope creep, a component that answers no requirement and should be cut.
This is exactly why the first exam trap on this knowledge point is designing an architecture that is not traceable to any stated requirement. An untraceable component is a liability: it adds cost, surface area, and maintenance with no justification, and no one can later say whether it is safe to remove. On a Claude project the temptation is often a shiny capability, an extra tool, a bigger model, a caching layer, that no requirement actually asked for. The disciplined move is to add nothing that cannot name the requirement it serves, and to flag any requirement that no component yet delivers.
Resolving conflicting requirements explicitly
Requirements often conflict, and the classic conflict on a Claude project is cost versus latency, or cost versus quality. A tight latency budget wants a faster, possibly larger, model; a firm cost ceiling wants a smaller model or heavier caching. These cannot both be maximised, so the mapping has to resolve the tension, and the second exam trap is resolving it silently. Quietly choosing to spend more for quality, or quietly accepting slower responses to save money, hides a business decision inside an engineering choice.
The correct move is to surface the conflict and resolve it explicitly against the business priority. That means naming the tradeoff, "we can meet the latency target with the larger model but it exceeds the cost ceiling, or we hit the ceiling with a smaller model that adds roughly a second," and confirming which the business values more before committing. The architect does not get to decide unilaterally whether this product prizes speed or spend; that is a business call, and the mapping's job is to make it visible so the right owner makes it. An explicit, confirmed resolution is traceable; a silent one is not.
A worked mapping
Worked example
A business goal states: 'Cut the time analysts spend triaging inbound compliance alerts by half, without increasing our AI spend.' The team must produce a solution architecture.
The goal carries two requirements that pull against each other: a productivity target that wants strong, fast classification, and a hard constraint that spend must not rise. The mapping starts by turning the goal into behaviour. The system must ingest each alert, classify it into a risk tier, and draft a short rationale an analyst can confirm, that is the functional half.
Now the non-functional half meets the constraint. Alerts arrive steadily but no single one blocks a human in real time, so throughput matters more than per-alert latency, which opens the door to processing in batches at lower cost. The shared classification instructions are stable across every alert, so caching that context avoids paying for it repeatedly. Together these let a right-sized model meet the productivity target while holding spend flat, and every one of those choices traces to a requirement: batch to the throughput profile, caching to the cost constraint, model size to the quality target.
Then the conflict surfaces. Pushing classification quality higher would favour a larger model, which threatens the no-increase-in-spend constraint. Rather than silently pick one, the team presents the tradeoff: current design meets the spend constraint with acceptable quality; a larger model would lift quality but breach the ceiling unless volume assumptions change. The business confirms that holding spend flat is the priority, and that confirmation is recorded as the justification for the model choice. The architecture is now both traceable and explicitly reconciled, which is exactly what this knowledge point asks you to produce.
Common misreadings to avoid
Misconception
A good architecture is judged by the capabilities it includes, so adding a useful component is always a plus.
What's actually true
Misconception
When cost and quality conflict, the architect should just pick the better-engineered option.
What's actually true
How this is tested on the CCDV-F exam
Because this is an apply-level knowledge point, questions give you a business goal and a set of requirements and ask you to choose or critique an architecture. The strongest answers show a clean chain from goal to behaviour to infrastructure and can name the requirement behind each decision; the weakest include a component nothing asked for or quietly resolve a cost-versus-quality conflict without confirming the business priority. Watch specifically for a distractor that sounds impressive, an extra model or tool, that no stated requirement justifies, and for one that picks a side of a tradeoff without surfacing it.
This closes task statement 2.1. Once a design is traceable to requirements, it is ready to move through the phases of delivery, which is where the systems life cycle in systems development life cycle phases takes over.
A proposed architecture for a Claude feature includes a vector search tool, a second fallback model, and a caching layer. Reviewing it, you find the vector tool serves no stated requirement, while a stated availability requirement has no component addressing it. What is the correct response?
People also ask
How do you map business requirements to a solution architecture?
What is requirements traceability and why does it matter?
How should conflicting requirements like cost and latency be resolved?
Watch and learn
Official Anthropic Academy lessons first, then hand-picked walkthroughs. Videos load only when you press play.
No videos curated for this concept yet
We are still curating the best official and community videos for this topic.
References & primary sources
Master this concept with Archie
Practice it inside an adaptive study session. Archie, your Socratic AI tutor, tracks your mastery with Bayesian Knowledge Tracing and schedules the perfect next review.