Why it is a question
THEORY-382 holds that OVIS lies between all-materialized and caching alone in one simulator, with uniform costs and single runs. It chooses fragments by estimated speed gain alone, ignoring each one’s cost in freshness, and its estimates ignore that one fragment’s refresh load delays others through the shared queue. Whether it is near the best plan under the same freshness threshold, or far from it, the evidence does not say. Until 2026-10-19 this was THEORY-382’s promote_when.
What would answer it
- An analysis bounding the response time of OVIS(θ), or of any rule of its kind, against the best plan that meets the same freshness threshold: the best fixed plan, or an offline optimum over the same access and update streams.
- A measurement on a real server or a recorded web workload that compares it with that best plan.
What would not answer it: more synthetic simulations in which OVIS lies between all-materialized and caching only. That is what THEORY-382 already holds, and it says nothing about the distance to the best plan.