Does keying on repr() make the encoder cache exact?

The metrics pipeline caches the JSON encoding of each tag value it sees. The assistant replaced the cache and says the new one can never return a wrong encoding.

Your task

The assistant stands by the claim in claim.md. Reach a verdict and prove it with a run.

• verdict() returns the string "true" or "false"
• tag_values() returns the values your run encodes, in order
• check() encodes them through the old and the new cache and compares every result with json.dumps
• a verdict on its own does not count; the run has to show it
• leave everything below the dashed line alone; it is the code under test and the graded driver

The suite replays your values at submit and checks the run settles the claim.