Make the OpenLibrary fallback reachable

[krejko]

Summary

SearchService merges OpenLibrary-derived candidates when no enriched metadata was found, so books with no Audible edition still produce results:

// Merge OpenLibrary-derived results … so OpenLibrary-only augmentation produces
// visible, scoreable items without calling Amazon.
if ((openLibraryDerivedResults != null && openLibraryDerivedResults.Any()) && !enrichedList.Any())

That branch can never be taken. Three independent defects upstream of it leave OpenLibraryDerivedResults empty, so the fallback silently does nothing and such books return zero results.

The three defects

1. The composed query is used as the OpenLibrary title.

var books = await _openLibraryService.SearchBooksAsync(query, null, 5);
//                                    ↑ "AUTHOR:Abelson TITLE:Structure and Interpretation…"

NormalizeForOpenLibrary tokenises on [a-z0-9\-]+, dropping the colons and leaving AUTHOR/TITLE as literal search terms. The author is also discarded as null, despite the caller having parsed it.

2. Exact title matches are discarded.

if (!string.IsNullOrEmpty(book.Title) && !string.Equals(book.Title, query, StringComparison.OrdinalIgnoreCase))

An exact title match is the strongest candidate available. Searching Radicalized by Cory Doctorow returns six documents, the first two titled exactly Radicalized — all dropped. The Garden of Rama kept only a four-book omnibus, precisely because the omnibus title differs from the query while the actual book's does not.

3. Query terms are joined with +, then URL-escaped.

var q = string.Join("+", qParts);
searchParams.Add($"q={Uri.EscapeDataString(q)}");   // "+" → %2B

OpenLibrary receives literal plus characters between words. Measured against the live API:

Query	numFound
radicalized%2Bcory%2Bdoctorow	0
radicalized cory doctorow	6
the%2Bgarden%2Bof%2Brama%2Barthur%2Bc%2Bclarke	1 (omnibus only)
the garden of rama arthur c clarke	5

This is the dominant one — it corrupts every OpenLibrary query the app makes.

Changes

Fixed

• Pass the caller's already-parsed title and author into SearchBooksAsync (SearchService logs both one line before the call); strip FIELD: prefixes for callers that only have a composed string.
• Keep documents whose title equals the query and let existing scoring/dedupe rank them.
• Join query terms with spaces so escaping produces separators rather than literal +.

Added

• AsinCandidateCollector_OpenLibraryQueryTests — parsed fields preferred, prefixes stripped when absent, plain queries untouched, and no call when OpenLibrary is disabled.

Testing

34/34 pass, including the existing parser suite.

Verified end-to-end against a running instance. Log lines that never appeared before now do:

Collecting candidates from OpenLibrary (title='Radicalized', author='Cory Doctorow')
OpenLibrary suggested title: Radicalized
Merging 1 OpenLibrary-derived candidate(s) into enriched results

Real searches, before → after:

Query	Before	After
Radicalized / Cory Doctorow	0 results	2 — 'Radicalized' first
The Garden of Rama / Arthur C. Clarke	1 — 'Rama Series, Collection Set of 4 Books…'	1 — 'The Garden of Rama'
Structure and Interpretation… / Abelson	0 results	1

A control case still correctly returns nothing: Chronicles of Narnia Intro is an intro track rather than a book, and OpenLibrary has no such title (numFound=0); dropping "Intro" returns 234.

Notes

Each defect masked the next, so fixing any one alone leaves the symptom unchanged — I confirmed each independently against the live OpenLibrary API before and after.

Behaviour is unchanged when EnableOpenLibrarySearch is off, and no Audible/Audnexus path is touched; this only affects queries that would otherwise have returned nothing.

🤖 Generated with Claude Code