During Business Understanding, Meridian's team established that the customer-service virtual assistant should be judged on successful resolution rate and appropriate escalation rate. At the Model Evaluation Go/No-Go, the data science team instead presents only model confidence scores and response latency. What governance gap should the project manager raise?

CPMAI's Model Evaluation task requires that evaluation criteria connect back to the acceptable performance metrics established in Business Understanding — this is the governance thread that keeps a project honest end to end. Here, the data science team is presenting confidence scores and latency, which are legitimate technical measures but not what the business actually agreed to be judged on (resolution rate and escalation rate). The project manager's correct move is to flag this mismatch and insist the Go/No-Go assessment be conducted against the originally agreed metrics before any decision is made — otherwise the project risks passing a technical bar that has quietly drifted from the business bar. Declaring any technical metric an acceptable substitute erases the whole point of setting business-tied criteria up front. Calling confidence scores and latency 'more sophisticated' and therefore fine misses that sophistication isn't the standard — relevance to the agreed business outcome is; a highly technical metric that doesn't answer the business question is the wrong metric for this gate. Retroactively rewriting the original Business Understanding metrics to match whatever happened to get measured is a governance failure in the other direction — it lets whatever was easy to measure define success after the fact, instead of holding the project accountable to what was actually promised. The exam point: evaluation criteria must be traceable to Business Understanding's agreed metrics, and a PM should catch and correct drift between the two before a Go/No-Go decision is finalized.