Understand the mechanism

Tokenize inputs; if over limit, truncate/summarize/retrieve selectively; the model attends only to what remains; output tokens also consume the budget.

Read the context window path as a pipeline for the contract-review assistant. At each stage, name the representation, the owner, and how information can be lost. Identifiers must mark prompt versions, model versions, indexes, and policies so “randomness” is not the default explanation for every bug in context window.

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. Scoped specifically to context window / contract-review assistant / mechanism.

Keep the unit and the denominator visible when you discuss context window. A percentage without a base, or a latency without a percentile, hides the failure mode this chapter cares about on the contract-review assistant.

Learned stages estimate; deterministic stages enforce. A fluent result from the contract-review assistant does not prove context window used the right evidence. Preserve intermediates when privacy allows—candidate lists, traces, scores, citations—so you can see the first broken stage in the context window path.

Operational correctness for context window includes deadlines on the contract-review assistant. If stages that feed this workload sum past the user’s patience, trim earlier—usually pack less, retrieve less, or parallelize—before blaming the model vendor for context window. Mechanism diagrams that ignore time are incomplete.

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 / mechanism.

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 / mechanism.

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.