Meridian's engine-room crew files maintenance and safety reports in English, Tagalog, and Greek depending on the crew member on duty. The port-operations team wants one unified retrieval index across all three languages. What should most directly drive the embedding-model choice?

An embedding model with demonstrated multilingual coverage is correct because the requirement is a single unified index across three languages, and that only works if semantically equivalent text in English, Tagalog, and Greek is embedded close together in vector space - a model without genuine multilingual training may embed the same underlying meaning far apart across languages, silently breaking cross-language retrieval even though the corpus is fully indexed. Choosing purely on price is wrong because it ignores the one requirement the scenario states as essential - unified retrieval across three languages - and a cheap model that only handles English well would satisfy the budget constraint while failing the actual business goal. Choosing purely on context window is wrong because context window governs how much text a model can accept per embedding call, which has no bearing on whether it represents multiple languages in a shared semantic space; a large-context model can still be functionally monolingual. Choosing purely on smallest output dimensionality is wrong for the same category of reason - dimension count affects storage and compute cost, but says nothing about language coverage, and optimizing for it while ignoring multilingual capability would produce a compact index that still fails to match a Tagalog query against an English report. The exam-relevant point is that the consequence of an embedding-model choice must be evaluated against the specific retrieval requirement in the scenario, not a generic cost or size proxy.