Lesson 2: From ranked chunks to the model's context
36 min read·13 Oct 2026
Module 8 · Lesson 2 · About 50 minutes · Needs: Lesson 1 (the cross-encoder reranker) and Module 6 Lesson 3 (score thresholds, near-duplicate collapse)
The situation
On Friday 9 October 2026 Ellie Shaw asks whether her repeated E7 fault is covered. Lesson 1's reranker returns ten documents she may see, each scored. The answer step (Module 9) will read all of whatever you hand it.
Hannah spots the archived 2024 terms and the injected forum post among them. Owen asks what ten whole documents cost. Grace asks what happens when none is any good. This lesson builds hollin/context.py, the step in between. The Hollin data is simulated.