Degraded reasoning quality and compute concerns — paying user reporting session failures

I am a paying Claude Code user (Max plan, ~€90/month) and I need to formally report a session where the quality was significantly below the standard I pay for and have previously experienced.

Issues observed in this session:

1. Incorrect diagnosis repeated multiple times — The assistant incorrectly diagnosed a Docker network isolation problem, sent me down that path, then admitted it was wrong. Then incorrectly concluded the issue was a Postgres password mismatch in the env file, which was also wrong. The actual root cause (pgdata volume initialized with a different password) was reached only after significant back-and-forth.

1. "Let us know" phrasing — The assistant said "let us know what happens" — a basic error for a reasoning model that should be aware it is a single AI instance, not a team.

1. Circular debugging — Multiple rounds of the same hypothesis being revisited without genuine forward progress.

Core concern:

I have no way to verify whether the compute backing claude-sonnet-4-6 today is the same as last week. The quality drop is noticeable. Paying users deserve transparency about whether model performance or compute allocation has changed. If Anthropic is silently degrading quality under load or through configuration changes, that is a serious trust issue.

Request:

Confirm that claude-sonnet-4-6 via Claude Code Max is receiving consistent, uncompromised compute. If not, explain what changes and when.