Paying customer: repeated incomplete/buggy results, churn, and constant clarifying questions instead of using available references

I am a paying Claude / Claude Code customer (100 €/month, several months). This is a formal complaint and a request for refund + cancellation handling by a human — not a generic "AI sometimes makes mistakes" reply.

What happened

• I set one clear goal for a project in my own GitHub repo (with README, committed config, a committed OBS scene file, and a GitHub Wiki — i.e. all the references were already present).
• Instead of reading those canonical references and doing the job, I got:
• Constant clarifying / approval questions for things that were already documented or discoverable.
• Large amounts of plumbing / refactor work with no visible product value.
• New bugs and regressions that I, the customer, had to find by live-testing.
• Output that was incomplete and did not match my actual setup (e.g. a UI listing ~76 "overlays" with duplicates when my real setup has 2; a telemetry value shown as a fake constant; stale docs missing one of my actual sim scenes).
• The net result over months: significant time lost, repeated frustration, and the core goal still not delivered.

Why "AI makes mistakes" is not an acceptable answer here

The issue is not occasional errors. It is a repeated failure to read the provided references, retain/document the information correctly, and execute the stated goal — while repeatedly asking the user instead. As a paying customer I am not a free trainer.

What I want

1. Refund of the affected billing periods.
2. Confirmation of cancellation.
3. Handling by a human, not an automated reply.

Account: kopp.patrick88@gmail.com

This issue was filed at my explicit request to document the experience publicly.