Calibrating Analysis Prompts (CCAO-F)

Analysis lives at the control end

With the control-latitude dial in mind, analysis is the type that sits furthest toward control. The Claude Certified Associate - Foundations (CCAO-F) exam treats calibrating an analysis prompt as an apply-level skill because the quality of an analysis depends almost entirely on how tightly you specify the criteria, the standards, and the scope. Get those explicit and Claude produces a ranked, committed judgement; leave them implicit and it produces a hedged summary that decides nothing.

The defining move is to tell Claude exactly what to measure and against what standard. An analysis is a judgement made against criteria, so the criteria are not optional detail - they are the substance of the task. The one thing you leave loose is phrasing; everything about what is measured and how it is weighed should be nailed down in the prompt.

State what to measure and against what standard

The core of an analysis prompt is the criteria. If you ask Claude to "compare these two contracts," you have named the objects but not the yardstick, so Claude picks its own criteria and its own sense of what "better" means - which may not be yours. If you instead ask it to compare them on payment terms, termination rights, and liability caps, and to state for each which is more favourable to you and why, you have supplied the yardstick, and the analysis becomes a committed, standard-based judgement.

The scope matters too: which items are in, over what period, under what assumptions. Tightening scope keeps the analysis bounded and comparable rather than sprawling. Together, explicit criteria and defined scope are what turn "analyse this" into a task Claude can execute to a standard you recognise. The phrasing of the result - the exact wording of each finding - is the part you can safely leave to Claude, because it is not where an analysis succeeds or fails.

Resolve the ambiguity yourself

A distinctive discipline of analysis prompts is that ambiguity should be resolved by the prompt, not by Claude. In a brainstorm you might welcome Claude interpreting an open prompt freely; in an analysis, an unresolved ambiguity means Claude guesses at the standard, and a guessed standard is unlikely to be the one you needed. So if there is a question of how to weigh two criteria, how to handle a tie, or what counts as a threshold, the prompt should say - not leave it open.

This is the low-latitude nature of analysis made concrete. High specification is not fussiness here; it is the mechanism by which the analysis becomes reliable and reproducible. The more of the analytical judgement you encode in the criteria and standards, the less Claude has to invent, and the closer the result lands to what you would have concluded yourself with the time to do it. Leaving the ambiguity for Claude to resolve is the characteristic way analysis prompts underperform.

What the CCAO-F exam trips candidates on

Two traps recur, and both leave the analysis under-specified.

The first is leaving evaluation criteria implicit and expecting Claude to infer the same standard the requester has in mind. A scenario asks for a comparison without naming the criteria and a distractor treats the vague result as Claude's shortcoming. The credited reading is that the criteria were never supplied, so Claude guessed - the fix is to state exactly what to measure and against what standard.

The second is applying brainstorming-style loose constraints to an analytical comparison, producing an unranked, uncommitted answer. A scenario loosens an analysis prompt in the name of not over-constraining, and a distractor praises the openness. The reliable reading is that analysis wants the dial toward control; loose constraints here yield a hedged, non-committal answer that fails the task's purpose.

Common misreadings to avoid

How this shows up on the exam

Domain 1 questions on this knowledge point show an analysis that came back vague or unranked and ask for the fix, or contrast a tight and a loose analysis prompt. The reliable reading is that analysis wants explicit criteria, a named standard, and defined scope, with ambiguity resolved in the prompt - a hedged result almost always traces to criteria left implicit.

This knowledge point applies the four task types and the control-latitude dial to the control end of the range, and it stands in deliberate contrast to calibrating brainstorming prompts at the latitude end. It is one of the four calibrations that feed into selecting strategy for a given scenario.