SLA Definition and Threshold Traceability for the CCAR-P Exam

What an SLA actually commits to

Once the feedback loop decides that something matters, the SLA defines what happens when it crosses the line. The CCAR-P exam treats SLA design as an understand-level skill built on two ideas: an SLA names three things, and its thresholds have to come from somewhere real. Get either wrong and you have a commitment that either cannot be acted on or cannot be defended.

An SLA is a commitment that makes three things clear: what are we measuring, what counts as a breach, and what happens when a breach occurs. All three are required. A metric with no breach condition is just telemetry; a breach condition with no consequence is a number nobody has to respond to. The SLA is the piece that turns a monitored signal into an obligation.

Thresholds trace to a source, not a round number

The threshold is where SLAs go wrong. A number that "sounds reasonable" is not an SLA threshold; it is a guess wearing an SLA's clothes. Every threshold should trace back to something tangible. Latency should reflect the user-experience expectation identified in discovery, so a two-second budget exists because the stakeholder said the interaction must feel instant, not because two is a tidy number. Availability should reflect how critical the deployment is to the business. Quality should reflect the eval results and acceptance criteria already established during evaluation.

That traceability is what keeps an SLA defensible. When a threshold is challenged, you can point to the discovery constraint, the criticality assessment, or the eval baseline that produced it. When it traces to nothing, you cannot defend it, and you also cannot tell whether a breach is a real problem or an artifact of a number someone picked arbitrarily. A defensible SLA is one whose every threshold has a named source.

Cost is the expectation that breaks after launch

Among all the SLA dimensions, cost is the one most likely to break, and it breaks in a specific way. Production volume routinely runs one to two orders of magnitude above the pilot, so a per-interaction cost that looked trivial in the proof of concept becomes a five-figure monthly line at scale. This is the same dynamic that makes reversal cost the decisive factor: a per-unit number that is fine in isolation becomes a surprise when multiplied by real volume.

The fix is to pre-empt it rather than react to it. Give the stakeholder a consumption forecast at expected production volume, name the spend-control posture, caching, model tiering, budget alerts, and frame the model-tiering narrative before the first invoice arrives rather than after. Leaving cost out of the SLA conversation because the pilot's per-interaction cost looked trivial is exactly how the expectation gets set at a level production immediately blows past.

What the exam trips candidates on

The first trap is setting an SLA threshold that "sounds reasonable" without tracing it to a discovery constraint, an eval result, or a business-criticality source. The exam offers tidy round-number thresholds with no lineage, and the credited answer is the one whose threshold traces to a named source, because that is what makes the SLA defensible.

The second trap is leaving cost out of the SLA conversation because the pilot's per-interaction cost looked trivial. The exam sets up a deployment where cost was never discussed and then scales it, and rewards recognising that cost is the expectation most likely to break and needs a forecast and a spend-control posture set before launch.

Common misreadings to avoid

How this shows up on the exam

Questions ask you to design or critique an SLA, or to spot which threshold is indefensible. The reliable reading is that an SLA names the metric, the breach, and the consequence, that each threshold must trace to a discovery constraint, a criticality assessment, or an eval baseline, and that cost is the dimension most likely to break after launch and the one most often left out.

This builds directly on the feedback loop as a decision layer, since the SLA defines the consequence side of the loop, and it feeds building a governance table, where SLA breaches become rows with triggers, owners, and actions. Its cost theme echoes reversal cost as the decisive factor.