Update FIELD_NOTES.md

FIELD_NOTES.md

@@ -4,8 +4,7 @@
 
 ## The problem came from a real person
 
-We didn't start with a model - we started with **Jack**, who manages library
-resources in Worcester. His complaint, almost verbatim: *"the library doesn't shout
+We didn't start with a model - we started with **Ben**, who manages works in the library in Worcester. His complaint, almost verbatim: *"the library doesn't shout
 loudly enough about everything it offers."* That's not a marketing gap, it's an
 **information gap** - and it turns out the UK government agrees. DCMS's 2024–25
 research into library non-users found the single biggest fixable barrier is exactly
@@ -25,13 +24,11 @@ data and the graph, and let the model be small and swappable. The payoff: an hon
 with the brief, and a **no-LLM fallback** that still answers from raw live data, so a
 flaky endpoint never breaks the demo.
 
-## What we learned mining the council's data
+## What we learned
 
 **1. The catalogue had a hidden clean API.** The Worcestershire catalogue runs on
-SirsiDynix Enterprise - usually painful to scrape (JS-heavy). But probing it, we found
-an undocumented **Atom feed** (`/client/rss/hitlist/wcc/qu=…`) that returns clean XML
-with title, author, format, year and ISBN per result. One lucky find turned a scraping
-nightmare into a five-line parser.
+SirsiDynix Enterprise. But probing it, we found it returns clean XML
+with title, author, format, year and ISBN per result.
 
 **2. "What you need to sign up" is the part everyone gets wrong.** Eligibility *varies
 wildly* and is buried: digital membership is instant-by-postcode; full membership needs