context_ledger.py

LLM context windows are budgets, not memories, with complex truncation rules

The context window of large language models is often misunderstood as a direct memory capacity. In reality, it functions more like a budget, with tokens consumed by the system prompt, tool schemas, and reserved output s…