Code Execution for Verified Planning Figures for the CCAO-F Exam

From knowing numbers matter to producing them safely

The synthesis-versus-calculation distinction says a plan's numbers must be verified rather than trusted as prose. This knowledge point is the practical answer to how. The Claude Certified Associate - Foundations (CCAO-F) exam treats it as an apply-level skill: given a plan that rests on figures, produce those figures by uploading the data and running code execution on it, rather than accepting an estimate written in the flow of a response.

The move rests on a simple contrast. A figure asked for in prose is generated to sound plausible; a figure produced by code execution is computed from the actual data and can be checked. Those are not two grades of the same thing, they are different in kind. This builds directly on synthesis versus calculation in planning work, which told you numbers need a verifiable method; code execution is that method.

Upload the data and name the mechanism

Two things make this work, and both are deliberate. First, the source data has to be present: you upload the dataset the figures should come from, so the computation runs on the real numbers rather than on the model's impression of them. Second, and this is the part people skip, you direct Claude to use code execution explicitly. Simply asking Claude to "calculate" a figure leaves the door open for it to produce a plausible prose estimate instead of running code, which defeats the purpose.

Naming the mechanism is what closes that door. When you say "use code execution on the attached data to compute the quarterly growth rate and average throughput", Claude runs an actual calculation over the uploaded figures and returns a result you can trace to the data. The instruction does two jobs at once: it points at the right source and it forces the right method. Both are needed, because the right source computed by the wrong method is still a guess, and the right method on absent data has nothing real to compute.

What code execution buys the plan

Code execution can produce the derived figures plans actually depend on: trends over time, per-unit rates like throughput per person, totals, and other calculations run straight from the data. These are exactly the numbers that, if wrong, break a recommendation. Computing them rather than estimating them changes the character of the whole plan.

The payoff is defensibility. A plan built on a guessed utilisation rate is a guess with a recommendation attached; a plan built on a code-executed calculation of the actual data can be defended line by line, because every figure traces back to a computation someone can re-run. That line-by-line defensibility is what lets verified figures become a defensible recommendation that stands up to scrutiny. The synthesis around the numbers stays Claude's strength; code execution is what makes the numbers underneath it trustworthy.

What the CCAO-F exam trips candidates on

The exam tests two traps. The first is asking Claude to "calculate" a figure without directing it to use code execution, which can still yield a plausible-sounding but ungrounded estimate. A scenario may show a planner requesting a number in prose and treating the answer as computed; the credited move is to require code execution on the uploaded data so the figure is actually calculated. The second trap is accepting a headline number without checking whether it came from computation on the real data or from prose estimation. The exam wants you to interrogate the provenance of the figure, not just its value.

Both traps reward the same instinct: a number is only as trustworthy as the method that produced it. Direct the calculation to code execution over the actual dataset, and confirm that is where any decision-driving figure came from.

Common misreadings to avoid

How this shows up on the exam

Domain 4 questions on this knowledge point describe a plan that depends on figures from an uploaded dataset and ask how those figures should be produced, or they show a prose-estimated number and ask what is wrong with relying on it. The reliable reading is to upload the data and direct Claude to use code execution, and to check that any decision-driving figure was computed rather than estimated.

This knowledge point builds on synthesis versus calculation in planning work, which establishes why numbers need a verifiable method, and it feeds into building a defensible recommendation from verified figures and separating synthesis steps from judgment steps in a plan, which together turn computed figures into a recommendation while keeping the final call human.