Module 02 Summary

What this module established

A chunk is the unit that gets retrieved, ranked and cited, so the question is not how many characters it holds but whether a reader could be pointed at it as evidence. Metadata is not decoration: visibility, updated and section each enable a capability later in the course, and two of them cannot be recovered after ingestion.

Carry forward

• Split on the seams the text already has. Fixed-width splitting separates a rule from its exception, and only the rule ranks.
• Permissions belong on the chunk and are enforced before ranking. A prompt instruction reaches a component that already has the document.
• Three string checks at ingestion - pronoun, cross-reference, length - catch most uncitable chunks without a model.

Before moving on

Move on when your chunks all carry the five metadata fields and you can report the stand-alone failure rate as a rate, before and after a re-split.

Mark complete