A vendor tells Meridian's PM, 'our algorithm and our model are the same thing, so once you approve the algorithm, the model never needs re-review.' What is wrong with this statement, and how should the PM correct it?

CPMAI Task 3 requires differentiating between machine-learning algorithms and models, a distinction with real governance consequences: an algorithm is a general, reusable learning procedure (e.g., a specific type of tree-based or neural-network method), while a model is the concrete, trained artifact that results from running that algorithm on a particular dataset — in this case, Meridian's own maintenance or demand data. Approving that an algorithm is a sound general-purpose choice does not certify that every model trained from it on Meridian's specific, evolving data is trustworthy; each trained model (and each retraining) needs its own evaluation, especially given that data drift can change how well a previously-approved algorithm performs once retrained. The 'vendor is correct, interchangeable, no re-review needed' distractor accepts a governance gap that would let an untested retrained model go live without scrutiny — a real operational risk. The 'model is the hardware server' distractor invents an unrelated infrastructure definition; a model is a data-derived artifact, not physical hardware. The 'algorithm only exists in reinforcement learning' distractor is false — algorithms underlie every ML paradigm (supervised, unsupervised, reinforcement), including Meridian's forecasting and maintenance work. Insisting on model-level, not just algorithm-level, review protects Meridian from silently degraded retrained models slipping into production.