Vendors are tuning their implementation (e.g. the schema, indexing, and queries used to implement the benchmark) and also their database system, but they cannot

Vendors are tuning their implementation (e.g. the schema, indexing, and queries used to implement the benchmark) and also their database system, but they cannot tune the benchmark workload itself. Apologies if this was ambiguous in my comment. It is quite typical that vendors find performance issues in their systems when they first adopt the benchmark. This then leads to improvements in their operators, optimizer, storage, etc. These improvements will be included in a new release of the system – which is great, as a main point of the benchmark is to drive innovation and raise the general level of performance in the technology space. It's important that the audits can only be conducted on systems that are generally available. We don't allow audits on some one-off version of a system that has special features just to perform well on the benchmark (these are known as "benchmark specials" and are also illegal in TPC benchmarks).