A public-records agency compares two classifiers for routing incoming mail. The more accurate model needs a GPU cluster the mailroom cannot host this year. The simpler model runs on the existing scanners and still beats the manual sort used today. Which result should the team recommend for this year's routing?
Select an answer to reveal the explanation.
Short Explanation
A faster engine you cannot bolt onto the truck does not win the route you are running this year. You recommend the scanner model the mailroom can run, not the more accurate model it cannot host. The trap is crowning the accuracy table when the winner cannot be deployed.
Full Explanation
A recommendation has to be something operations can put on the live path inside the decision window. The scanner model runs on hardware the mailroom already has and beats the manual sort, so it changes this year's routing. The memo should name a model the scanners can execute on live mail, not a lab score that would have required the missing cluster. The GPU model is more accurate in the lab and cannot be hosted this year, so its statistical gain is not a recommendation for the mail that must move now. Choosing the GPU model and waiting for a future cluster postpones a gain the scanners can deliver and bets on hardware the stem says is unavailable. Choosing neither model leaves the weaker manual sort in place even though a deployable model already beats it. Choosing the GPU model while clerks keep sorting by hand treats a table as the decision and does not route a single live piece with the better lab score. Exam caveat: if the GPU model could be served this year, the accuracy edge could become the recommendation; this stem says it cannot. Operational check: confirm the artifact named in the decision memo actually runs on the mailroom's current scanners.