Mastery: connect the pieces

1. Idea — The context window is shared working memory for one request—not permanent memory. What you pack, drop, or summarize changes outcomes.…
2. Mechanism — Tokenize inputs; if over limit, truncate/summarize/retrieve selectively; the model attends only to what remains; output tokens also consume the budget.…
3. Controls — max input tokens, history summarization, retrieval token budget, system prompt size, output cap, priority pack order
4. Tradeoff — More evidence can help and can crowd out instructions. Summaries save space and can drop obligations. Long windows raise cost and attention dilution risk.…
5. Failures — Silent truncation; Instruction eviction
6. Metrics — overflow rate, obligation-coverage on long contracts, p95 tokens in, cost per review
7. Ship rule — Pin 2k system tokens, budget 14k exhibits via retrieval, summarize history beyond 4k, abort if obligations pack fails checksum.

Neighboring layers (retrieval, serving, policy, human review) still own what context window cannot on the contract-review assistant. Do not ask this chapter’s dial to replace permissions or source truth.

Mastery is transfer: take context window into a second scenario that is not the contract-review assistant and rebuild the checklist without copying sentences. If you can only recite this chapter’s examples for context window, you have memorized a story, not a model.

Also pin one numeric memory from this context window chapter: A 32k window with 8k system+tools, 12k retrieved exhibits, 6k history leaves ~6k for the draft—and less if tokenization is heavier than word count suggests. That number is not decoration; it is a template for how claims about context window on the contract-review assistant should look in design docs. Scoped specifically to context window / contract-review assistant / mastery-connect.

People confuse context window with neighboring buzzwords when debugging the contract-review assistant. Before changing prompts, ask whether the broken stage was evidence gathering, the context window judgment itself, validation, or the product action. Fixing the wrong stage creates folklore (“we tried context window and it failed”) that blocks the next team on the contract-review assistant. Scoped specifically to context window / contract-review assistant / mastery-connect.

Write a five-line artifact for this page: goal, inputs, check, owner, stop rule. Invent one fluent failure that the check would catch. Keep details specific to context window rather than generic AI advice.

Write a five-line artifact for this page: goal, inputs, check, owner, stop rule. Invent one fluent failure that the check would catch. Keep details specific to context window rather than generic AI advice.