Your local model's usable context is smaller than its context window and Memory isn't why

I've been building an open-source tool that tests whether self-hosted models can actually drive agents. Running it against my own setup produced a result I didn't expect, and I think it generalizes. Disclosure up front: I build QuantaMind (Apache 2.0, runs offline, no telemetry). The data below came out of it. This post is about the finding, not the tool. Setup Qwen3.5-9B, Q4_K_M llama.cpp, native function calling 16GB M-series Mac 5 multi-step agentic tasks, k=4 runs each k=4 matters. Scoring…