The numbers we design against
Non-functional requirements written before the engine was built rather than measured afterwards and called targets. If a number here is wrong for your scale, that is a useful thing to discover during an evaluation rather than during a rollout.
Scale envelope
The product is designed against a working set of roughly one hundred thousand parts per tenant, structures up to five thousand lines, assemblies up to fifteen levels deep, and document vaults in the low terabytes. Those are the numbers the indexes, the resolver and the closure tables were sized for.
Latency budgets
Configuration resolution runs at p95 under 400 milliseconds on a five-thousand-line structure. Grid interactions target sixteen milliseconds per frame, which is what makes pasting four hundred rows from a spreadsheet feel like a spreadsheet. Search returns within 300 milliseconds at p95 across all three engines, filtered at query time rather than served from a pre-partitioned index.
Why these are published
Because a PLM that is fast on a demo tenant and slow on yours is the most common disappointment in this category, and the only defence a buyer has is a number they can hold a vendor to. If your structures are larger than the envelope above, say so early and we will tell you honestly whether it is a configuration question or a no.