CI: push to ipcr-push-85-17-246-67.ip.nsl.sh (the box's cert no longer covers nip.io) [skip ci]

The test job's linker was killed with signal 9 twice (run 705): cargo test links
several test binaries at once, each with full debug info. CARGO_PROFILE_DEV_DEBUG=0
and CARGO_BUILD_JOBS=2 keep the link within the job's memory; the tests are the
same.

.gitea/workflows/publish.yml

@@ -2,10 +2,10 @@
 # folder exists, and GitHub never reads it.
 #
 # Publishes to IPCR on watch.nsl.sh instead of ghcr.io:
-#   push  ipcr-push-85-17-246-67.nip.io/metadec.eth/<repo>:<tag>   docker login as metadec.eth with the
+#   push  ipcr-push-85-17-246-67.ip.nsl.sh/metadec.eth/<repo>:<tag>   docker login as metadec.eth with the
 #         IPCR_PUSH_TOKEN secret (the metadec.eth namespace token from the IPCR dashboard)
 #   pull  ipcr-watch.nsl.sh/ipns/metadec.eth/metadec.eth/<repo>:<tag>   anonymous
-# The push goes to ipcr-push-85-17-246-67.nip.io, the box's own address, not ipcr-push-watch.nsl.sh: Cloudflare fronts
+# The push goes to ipcr-push-85-17-246-67.ip.nsl.sh, the box's own address, not ipcr-push-watch.nsl.sh: Cloudflare fronts
 # *.watch.nsl.sh and refuses any request over 100 MB (413), which a large image layer is.
 # That host's certificate is signed by the box's private Mesh Router CA, given to buildkit from
 # the IPCR_CA_PEM repo variable. docker login would make the job's Docker daemon check the
@@ -27,7 +27,7 @@ on:
     tags: ['v*']
 
 env:
-  REGISTRY: ipcr-push-85-17-246-67.nip.io
+  REGISTRY: ipcr-push-85-17-246-67.ip.nsl.sh
   IMAGE_NAME: metadec.eth/${{ github.event.repository.name }}
 
 jobs:
@@ -71,6 +71,9 @@ jobs:
           tags: |
             type=ref,event=branch
             type=semver,pattern={{version}}
+            # A fix to release 1.2.3 is tagged v1.2.3.1: four numbers are not semver, so
+            # the pattern above makes no tag for it; this one does (image tag 1.2.3.1).
+            type=match,pattern=v(\d+\.\d+\.\d+\.\d+),group=1
             type=semver,pattern={{major}}.{{minor}}
             type=semver,pattern={{major}}
             type=sha,format=short

.gitea/workflows/test.yml

@@ -0,0 +1,31 @@
+# Runs the feeder's tests on every push. The repo had only the image workflow,
+# so nothing checked a change before it shipped.
+
+name: Test
+
+on:
+  push:
+    branches: ["**"]
+
+jobs:
+  rust:
+    name: cargo test
+    runs-on: ubuntu-latest
+    # The job ran out of memory linking the test binaries (ld killed with
+    # signal 9, run 705 twice): no debug info, and at most two links at once.
+    # What is tested does not change.
+    env:
+      CARGO_PROFILE_DEV_DEBUG: "0"
+      CARGO_BUILD_JOBS: "2"
+    steps:
+      - uses: actions/checkout@v4
+      # rust-toolchain.toml pins the channel (with clippy); rustup installs it on
+      # first use, so the action only has to provide rustup itself.
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          components: clippy
+      - uses: Swatinem/rust-cache@v2
+      - name: cargo test
+        run: cargo test --locked
+      - name: cargo clippy
+        run: cargo clippy --all-targets -- -D warnings

.github/workflows/publish.yml

@@ -43,6 +43,9 @@ jobs:
           tags: |
             type=ref,event=branch
             type=semver,pattern={{version}}
+            # A fix to release 1.2.3 is tagged v1.2.3.1: four numbers are not semver, so
+            # the pattern above makes no tag for it; this one does (image tag 1.2.3.1).
+            type=match,pattern=v(\d+\.\d+\.\d+\.\d+),group=1
             type=semver,pattern={{major}}.{{minor}}
             type=semver,pattern={{major}}
             type=sha,format=short

Cargo.lock

@@ -792,7 +792,7 @@ checksum = "f8ca58f447f06ed17d5fc4043ce1b10dd205e060fb3ce5b979b8ed8e59ff3f79"
 [[package]]
 name = "meta-feeder-sdk"
 version = "1.7.0"
-source = "git+https://github.com/worph/meta-feeder-sdk?tag=v1.7.0#027917c8cffe7ca43dadcf3b426cc5f69ecc443a"
+source = "git+https://gitea-watch.nsl.sh/metastar/meta-feeder-sdk?tag=v1.7.0#027917c8cffe7ca43dadcf3b426cc5f69ecc443a"
 dependencies = [
  "anyhow",
  "async-trait",

feeder-plugin/book-feeder/Cargo.toml

@@ -24,7 +24,7 @@ path = "src/main.rs"
 # vendored copy this used to point at had already drifted behind the shared SDK
 # (no `RateBudget`, a capped card cid). The rest of the fleet pins the same
 # way; see meta-feeder-anilist's manifest.
-meta-feeder-sdk = { git = "https://github.com/worph/meta-feeder-sdk", tag = "v1.7.0" }
+meta-feeder-sdk = { git = "https://gitea-watch.nsl.sh/metastar/meta-feeder-sdk", tag = "v1.7.0" }
 
 tokio = { version = "1.40", features = ["full"] }
 axum = { version = "0.7", features = ["json"] }

feeder-plugin/book-feeder/src/gutenberg.rs

@@ -32,7 +32,37 @@ use meta_feeder_sdk::types::{DiscoveryRecord, GatewayError, PluginHealth};
 use serde::Deserialize;
 use tracing::{debug, warn};
 
-use crate::openlibrary_client::{self, OpenLibraryClient};
+use crate::openlibrary::first_filter;
+use crate::openlibrary_client::{self, Etext, OpenLibraryClient};
+
+/// An edition's Open Library work, and how it was found.
+#[derive(Debug, Clone)]
+struct Binding {
+    work: String,
+    /// Bound by title and author ([`OpenLibraryClient::title_work`]) rather than
+    /// by an identifier Open Library records: graded Matched, not Anchored.
+    by_title: bool,
+}
+
+/// One etext's binding: the identifier Open Library records, else — only for an
+/// etext it lists under no work — the title match. Shared by the search path and
+/// the redeem path so the two cannot bind the same book differently.
+async fn bind(
+    client: &OpenLibraryClient,
+    id: u64,
+    title: &str,
+    author: &str,
+    deadline: std::time::Duration,
+) -> Result<Option<Binding>, GatewayError> {
+    Ok(match client.gutenberg_etext(id, deadline).await? {
+        Etext::One(work) => Some(Binding { work, by_title: false }),
+        Etext::Several => None,
+        Etext::Unlisted => client
+            .title_work(title, author, deadline)
+            .await?
+            .map(|work| Binding { work, by_title: true }),
+    })
+}
 
 /// Canonical Gutendex base URL. Overridable via [`GutenbergPlugin::with_base_url`]
 /// for tests + private mirrors.
@@ -390,31 +420,60 @@ impl GutenbergPlugin {
             .await
     }
 
+    /// The Gutendex search a work page's query stands for.
+    ///
+    /// meta-read's book page asks for `openlibraryid:OL…W` with no free text.
+    /// That used to fall through to the catalogue browse — the 32 most
+    /// downloaded books — and the gateway kept those bound to the work, so a
+    /// book outside that list never showed an edition on its own page. Searching
+    /// for the work's main title and its first author's surname finds it
+    /// wherever it ranks; the gateway still keeps only the records bound to the
+    /// work. `None` (no work named, Open Library off or failing) keeps the browse.
+    async fn work_search_text(&self, query: &GatewayQuery) -> Option<String> {
+        let work = first_filter(query, "openlibraryid")?.to_ascii_uppercase();
+        if !openlibrary_client::is_work_id(&work) {
+            return None;
+        }
+        let client = self.openlibrary.as_ref()?;
+        let lookup = client.doc_by_work_id(&work, openlibrary_client::BINDING_DEADLINE);
+        let doc = tokio::time::timeout(BINDING_TIMEOUT, lookup).await.ok()?.ok()??;
+        let title = openlibrary_client::main_title(doc.title.as_deref()?);
+        if title.is_empty() {
+            return None;
+        }
+        let author = doc.author_name.first().map(|a| openlibrary_client::surname(a)).unwrap_or_default();
+        Some(format!("{title} {author}").trim().to_string())
+    }
+
     /// Each result's Open Library work, through the Project Gutenberg id Open
-    /// Library records (reading-model §2.4). Aligned with the input slice;
+    /// Library records (reading-model §2.4), or, for an etext Open Library lists
-    /// `None` means no binding — none recorded, ambiguous, not configured, over
+    /// under no work, by title and author (`by_title`, graded Matched). Aligned
-    /// budget, or the lookup failed.
+    /// with the input slice; `None` means no binding — ambiguous, no title match,
+    /// not configured, over budget, or the lookup failed.
     ///
     /// ⚠ **A failure never blocks or drops a record.** The binding upgrades an
     /// edition that is already readable. Every failure is left uncached, so the
     /// next query that surfaces the book tries again; the budget deadline is
     /// short on purpose (`BINDING_DEADLINE`), so a cold browse row binds the
     /// first handful of books and the cache fills the rest over later queries.
-    async fn fetch_openlibrary_works(&self, books: &[GutendexBook]) -> Vec<Option<String>> {
+    async fn fetch_openlibrary_works(&self, books: &[GutendexBook]) -> Vec<Option<Binding>> {
         use futures::stream::{self, StreamExt};
         const BINDING_CONCURRENCY: usize = 4;
         let Some(client) = self.openlibrary.as_ref() else {
             return vec![None; books.len()];
         };
-        // Owned ids, not `books.iter().map(..)`: a borrowing closure inside the
+        // Owned values, not `books.iter().map(..)`: a borrowing closure inside the
         // stream makes `handle_query`'s future fail the `async_trait` lifetime
         // check ("FnOnce is not general enough").
-        let ids: Vec<u64> = books.iter().map(|b| b.id).collect();
+        let wanted: Vec<(u64, String, String)> = books
-        stream::iter(ids)
+            .iter()
-            .map(|id| async move {
+            .map(|b| (b.id, b.title.clone(), b.authors.first().map(|a| a.name.clone()).unwrap_or_default()))
-                let lookup = client.gutenberg_work(id, openlibrary_client::BINDING_DEADLINE);
+            .collect();
+        stream::iter(wanted)
+            .map(|(id, title, author)| async move {
+                let lookup = bind(client, id, &title, &author, openlibrary_client::BINDING_DEADLINE);
                 match tokio::time::timeout(BINDING_TIMEOUT, lookup).await {
-                    Ok(Ok(work)) => work,
+                    Ok(Ok(binding)) => binding,
                     Ok(Err(e)) => {
                         debug!(
                             target: "meta-share::gateway",
@@ -507,7 +566,16 @@ impl FeederPlugin for GutenbergPlugin {
         // everything, already ordered by download count (identical to
         // `?sort=popular`), so the row leads with Pride and Prejudice rather than
         // with whatever happens to contain punctuation.
-        let q = query.free_text.trim();
+        // A work page's query names its Open Library work and no text: search
+        // for that work's title and author instead of browsing. See
+        // `work_search_text`.
+        let free = query.free_text.trim();
+        let work_text = if free.is_empty() {
+            self.work_search_text(query).await
+        } else {
+            None
+        };
+        let q = work_text.as_deref().unwrap_or(free);
         let params: Vec<(&str, &str)> = if q.is_empty() {
             Vec::new()
         } else {
@@ -568,9 +636,9 @@ impl FeederPlugin for GutenbergPlugin {
         // Not user-facing, so the binding lookup may queue for the full search
         // deadline. A failure still leaves the record unbound, never fails the
         // compute.
+        let author = book.authors.first().map(|a| a.name.clone()).unwrap_or_default();
         let work = match self.openlibrary.as_ref() {
-            Some(ol) => ol
+            Some(ol) => bind(ol, book.id, &book.title, &author, openlibrary_client::SEARCH_DEADLINE)
-                .gutenberg_work(book.id, openlibrary_client::SEARCH_DEADLINE)
                 .await
                 .unwrap_or_else(|e| {
                     debug!(
@@ -644,13 +712,13 @@ impl FeederPlugin for GutenbergPlugin {
 
 /// Convert a `GutendexBook` into our wire-level `DiscoveryRecord`.
 ///
-/// `openlibrary_work` is the binding from [`OpenLibraryClient::gutenberg_work`]:
+/// `openlibrary_work` is the binding from [`bind`]: the one work Open Library
-/// `Some` only when Open Library records exactly one work for this etext.
+/// records for this etext, or, when it records none, the title match.
 fn into_discovery_record(
     book: GutendexBook,
     cover_cid: Option<String>,
     bibrec: BTreeMap<String, String>,
-    openlibrary_work: Option<String>,
+    openlibrary_work: Option<Binding>,
 ) -> DiscoveryRecord {
     let GutendexBook {
         id,
@@ -803,16 +871,25 @@ fn into_discovery_record(
     let _ = download_count;
     fields.insert("fileName".into(), format!("gutenberg-{id}.epub"));
 
-    // The work binding and its trust claim (reading-model §2.4). Written only on
+    // The work binding and its trust claim (reading-model §2.4). On an
-    // an identifier link Open Library itself records, so the claim is the plain
+    // identifier link Open Library itself records, the claim is the plain
-    // Anchored form: `anchored` + `anchorSource`, and deliberately NO
+    // Anchored form: `anchored` + `anchorSource`, and NO `anchorMethod` — that
-    // `anchorMethod` — that marks a title match, which this never is. The
+    // marks a title match. The
     // gateway transcribes the claim into `anchoredBy/openlibrary:OL…W` and
     // strips the transient keys before storing.
-    if let Some(work) = openlibrary_work.filter(|w| openlibrary_client::is_work_id(w)) {
+    //
-        fields.insert("openlibraryid".into(), work);
+    // A title binding (an etext Open Library lists under no work) carries the
+    // same claim plus `anchorMethod=title`: the gateway then records
+    // `matchedBy/openlibrary:OL…W` instead of `anchoredBy/`, and meta-read grades
+    // the edition Matched — listed on its work's page, ranked after any
+    // identifier-bound edition.
+    if let Some(binding) = openlibrary_work.filter(|b| openlibrary_client::is_work_id(&b.work)) {
+        fields.insert("openlibraryid".into(), binding.work);
         fields.insert("anchored".into(), "true".into());
         fields.insert("anchorSource".into(), "openlibrary".into());
+        if binding.by_title {
+            fields.insert("anchorMethod".into(), "title".into());
+        }
     }
 
     for (k, v) in bibrec {
@@ -1678,6 +1755,98 @@ mod tests {
         }
     }
 
+    /// Open Library's answer to `title=…&author=…` for Frankenstein: several
+    /// works by Mary Shelley share the title; the most published is the novel.
+    fn frankenstein_by_title() -> Mock {
+        Mock::given(method("GET"))
+            .and(path("/search.json"))
+            .and(query_param("title", "frankenstein"))
+            .and(query_param("author", "shelley"))
+            .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({ "numFound": 3, "docs": [
+                { "key": "/works/OL26860869W", "title": "Frankenstein", "author_name": ["Mary Shelley"], "edition_count": 84 },
+                { "key": "/works/OL450063W", "title": "Frankenstein; or, The Modern Prometheus", "author_name": ["Mary Shelley"], "edition_count": 2187 },
+                { "key": "/works/OL25595002W", "title": "Mary Shelley's Frankenstein", "author_name": ["Mary Shelley"], "edition_count": 90 }
+            ]})))
+    }
+
+    /// ⚠ Open Library lists etext 84 under no work, so the identifier rule alone
+    /// left the most read novel on Gutenberg unbound and its book page empty.
+    #[tokio::test]
+    async fn an_etext_no_work_lists_is_bound_by_title_and_graded_matched() {
+        let server = MockServer::start().await;
+        Mock::given(method("GET"))
+            .and(path("/books/"))
+            .respond_with(ResponseTemplate::new(200).set_body_json(books_json()))
+            .mount(&server)
+            .await;
+        binding(
+            1342,
+            serde_json::json!({ "numFound": 1, "docs": [
+                { "key": "/works/OL66554W", "id_project_gutenberg": ["1342"] }
+            ]}),
+        )
+        .mount(&server)
+        .await;
+        binding(84, serde_json::json!({ "numFound": 0, "docs": [] }))
+            .mount(&server)
+            .await;
+        frankenstein_by_title().expect(1).mount(&server).await;
+
+        let (plugin, _dir) = configured_plugin_against(&server);
+        let records = plugin
+            .handle_query(&GatewayQuery::from_free_text("alice"), 50)
+            .await
+            .expect("handle_query");
+
+        let pride = &records[0].fields;
+        assert_eq!(pride.get("openlibraryid").map(String::as_str), Some("OL66554W"));
+        assert!(!pride.contains_key("anchorMethod"), "an identifier binding stays Anchored");
+
+        let frankenstein = &records[1].fields;
+        assert_eq!(frankenstein.get("openlibraryid").map(String::as_str), Some("OL450063W"));
+        assert_eq!(frankenstein.get("anchored").map(String::as_str), Some("true"));
+        assert_eq!(frankenstein.get("anchorSource").map(String::as_str), Some("openlibrary"));
+        assert_eq!(frankenstein.get("anchorMethod").map(String::as_str), Some("title"));
+    }
+
+    /// ⚠ A work page asks for its work and no text. That used to browse the 32
+    /// most downloaded books, so a book outside them never reached its own page.
+    #[tokio::test]
+    async fn a_work_page_query_searches_for_the_works_title_and_author() {
+        let server = MockServer::start().await;
+        Mock::given(method("GET"))
+            .and(path("/search.json"))
+            .and(query_param("q", "key:/works/OL450063W"))
+            .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({ "numFound": 1, "docs": [
+                { "key": "/works/OL450063W", "title": "Frankenstein; or, The Modern Prometheus", "author_name": ["Mary Shelley"] }
+            ]})))
+            .mount(&server)
+            .await;
+        Mock::given(method("GET"))
+            .and(path("/books/"))
+            .and(query_param("search", "frankenstein shelley"))
+            .respond_with(ResponseTemplate::new(200).set_body_json(books_json()))
+            .expect(1)
+            .mount(&server)
+            .await;
+        binding(1342, serde_json::json!({ "numFound": 0, "docs": [] }))
+            .mount(&server)
+            .await;
+        binding(84, serde_json::json!({ "numFound": 0, "docs": [] }))
+            .mount(&server)
+            .await;
+        frankenstein_by_title().mount(&server).await;
+
+        let (plugin, _dir) = configured_plugin_against(&server);
+        let mut query = GatewayQuery::from_free_text("");
+        query.filters.insert("openlibraryid".into(), vec!["OL450063W".into()]);
+        query.filters.insert("contentKind".into(), vec!["book".into()]);
+        let records = plugin.handle_query(&query, 50).await.expect("handle_query");
+        let frankenstein = records.iter().find(|r| r.record_id == "84").expect("etext 84 is found");
+        assert_eq!(frankenstein.fields.get("openlibraryid").map(String::as_str), Some("OL450063W"));
+        assert_eq!(frankenstein.fields.get("anchorMethod").map(String::as_str), Some("title"));
+    }
+
     /// ⚠ A binding failure never costs the record.
     #[tokio::test]
     async fn no_work_or_a_failed_lookup_leaves_the_record_unbound_but_present() {

feeder-plugin/book-feeder/src/openlibrary.rs

@@ -71,7 +71,7 @@ fn env_usize(key: &str, default: usize) -> usize {
 }
 
 /// First non-empty value for `key`.
-fn first_filter(q: &GatewayQuery, key: &str) -> Option<String> {
+pub(crate) fn first_filter(q: &GatewayQuery, key: &str) -> Option<String> {
     q.filters
         .get(key)
         .and_then(|v| v.first())
@@ -169,33 +169,14 @@ fn foreign_identity_filter(q: &GatewayQuery) -> Option<&str> {
 /// what passes, permanently. A card that lacks another filter's key does not
 /// match it, which is the truth.
 pub fn to_record(doc: &SearchDoc, description: &str, query: &GatewayQuery) -> Option<DiscoveryRecord> {
-    to_record_with(doc, Some(description), query, false)
-}
-
-/// [`to_record`] with the quality bar as a choice.
-///
-/// `lenient = false` is the bar every search and browse row keeps: a cover **and**
-/// a description, or no card. `lenient = true` is for a lookup that names one work
-/// by `openlibraryid` (a work page): the work is already known to be wanted, so a
-/// title is enough. Its cover may come from the work's ISBNs when the work has none
-/// of its own, and a missing description is simply left out.
-pub fn to_record_with(
-    doc: &SearchDoc,
-    description: Option<&str>,
-    query: &GatewayQuery,
-    lenient: bool,
-) -> Option<DiscoveryRecord> {
     let id = doc.work_id()?;
     let cid = card_cid(id)?;
     let title = doc.title.as_deref().map(str::trim).filter(|t| !t.is_empty())?;
-    let description = description.map(str::trim).filter(|d| !d.is_empty());
+    let poster_url = doc.cover_url()?;
-    let poster_url = if lenient {
+    let description = description.trim();
-        doc.cover_url().or_else(|| doc.isbn_cover_url())
+    if description.is_empty() {
-    } else {
+        return None;
-        // Strict: both are required.
+    }
-        description?;
-        Some(doc.cover_url()?)
-    };
 
     let mut fields: BTreeMap<String, String> = BTreeMap::new();
 
@@ -219,9 +200,7 @@ pub fn to_record_with(
 
     // `description/{lang3}`, never a flat `description` — the flat key is indexed
     // but never written (METADATA_KEYS.md §14.6).
-    if let Some(d) = description {
+    fields.insert("description/eng".into(), description.to_string());
-        fields.insert("description/eng".into(), d.to_string());
-    }
 
     let mut seen = HashSet::new();
     let authors: Vec<&str> = doc
@@ -245,9 +224,7 @@ pub fn to_record_with(
 
     // ⚠ `poster_url`, NOT a pre-hashed `poster`: the gateway's SEEDABLE_FIELDS
     // path fetches it, seeds the bytes and writes the content cid back.
-    if let Some(url) = poster_url {
+    fields.insert("poster_url".into(), poster_url);
-        fields.insert("poster_url".into(), url);
-    }
 
     fields.insert(
         "sourceUrl".into(),
@@ -280,14 +257,13 @@ async fn project(
     query: &GatewayQuery,
     n: usize,
     deadline: Duration,
-    lenient: bool,
 ) -> Vec<DiscoveryRecord> {
     let mut seen = HashSet::new();
     let candidates: Vec<SearchDoc> = docs
         .into_iter()
         .filter(|d| {
             d.work_id().is_some()
-                && (lenient || d.cover_url().is_some())
+                && d.cover_url().is_some()
                 && d.title.as_deref().is_some_and(|t| !t.trim().is_empty())
         })
         .filter(|d| seen.insert(d.key.clone()))
@@ -310,7 +286,7 @@ async fn project(
                         debug!(from = doc.key, to = w.id, "openlibrary: merged work; skipped");
                         return None;
                     }
-                    to_record_with(&doc, w.description.as_deref(), query, lenient)
+                    to_record(&doc, w.description.as_deref()?, query)
                 }
                 Ok(None) => None,
                 Err(e) => {
@@ -461,9 +437,7 @@ impl FeederPlugin for OpenLibraryPlugin {
             let Some(doc) = client.doc_by_work_id(id, SEARCH_DEADLINE).await? else {
                 return Ok(Vec::new());
             };
-            // Lenient: an explicit work lookup keeps the card even without a
+            return Ok(project(client, vec![doc], query, 1, SEARCH_DEADLINE).await);
-            // description, and finds a cover by ISBN when the work has none.
-            return Ok(project(client, vec![doc], query, 1, SEARCH_DEADLINE, true).await);
         }
 
         // A curated row.
@@ -481,7 +455,7 @@ impl FeederPlugin for OpenLibraryPlugin {
                         .await?
                 }
             };
-            return Ok(project(client, docs, query, n, BROWSE_DEADLINE, false).await);
+            return Ok(project(client, docs, query, n, BROWSE_DEADLINE).await);
         }
 
         // Free text.
@@ -496,7 +470,7 @@ impl FeederPlugin for OpenLibraryPlugin {
         let docs = client
             .search(text, (n * SEARCH_OVERFETCH).min(MAX_LIMIT), SEARCH_DEADLINE)
             .await?;
-        Ok(project(client, docs, query, n, SEARCH_DEADLINE, false).await)
+        Ok(project(client, docs, query, n, SEARCH_DEADLINE).await)
     }
 
     /// A card has no bytes: the record IS the payload.
@@ -515,8 +489,7 @@ impl FeederPlugin for OpenLibraryPlugin {
         // different work (a merge) is not this cid's record.
         let record = match self.client.as_ref() {
             Some(c) => match c.doc_by_work_id(id, SEARCH_DEADLINE).await {
-                // Lenient, like the `openlibraryid` lookup that minted this card.
+                Ok(Some(doc)) => project(c, vec![doc], &GatewayQuery::from_free_text(""), 1, SEARCH_DEADLINE)
-                Ok(Some(doc)) => project(c, vec![doc], &GatewayQuery::from_free_text(""), 1, SEARCH_DEADLINE, true)
                     .await
                     .into_iter()
                     .find(|r| r.record_id == id),
@@ -572,7 +545,6 @@ mod tests {
             cover_i: Some(14348537),
             first_publish_year: Some(1813),
             id_project_gutenberg: vec![],
-            isbn: vec![],
         }
     }
 
@@ -679,95 +651,6 @@ mod tests {
         assert!(to_record(&edition, "A novel.", &q("", &[])).is_none());
     }
 
-    #[test]
-    fn a_lenient_card_needs_only_a_title_and_finds_a_cover_by_isbn() {
-        let mut d = pride();
-        d.cover_i = None;
-        d.isbn = vec!["1955252092".into(), "9781955252096".into()];
-        // Strict still refuses it, whatever the description.
-        assert!(to_record_with(&d, Some("A novel."), &q("", &[]), false).is_none());
-        // Lenient mints it with the ISBN-13 cover and no description at all.
-        let r = to_record_with(&d, None, &q("", &[]), true).unwrap();
-        assert_eq!(
-            r.fields["poster_url"],
-            "https://covers.openlibrary.org/b/isbn/9781955252096-L.jpg?default=false"
-        );
-        assert!(!r.fields.keys().any(|k| k.starts_with("description")));
-        assert_eq!(r.fields["openlibraryid"], "OL66554W");
-        assert_eq!(r.fields["fileType"], "card");
-        // Same address as the strict card for the same work.
-        let strict = to_record(&pride(), "A novel.", &q("", &[])).unwrap();
-        let key = |r: &DiscoveryRecord| r.fields.keys().find(|k| k.starts_with("cids/")).cloned();
-        assert_eq!(key(&r), key(&strict));
-    }
-
-    #[test]
-    fn a_lenient_card_with_no_cover_anywhere_has_no_poster_rather_than_no_card() {
-        let mut d = pride();
-        d.cover_i = None;
-        let r = to_record_with(&d, None, &q("", &[]), true).unwrap();
-        assert!(!r.fields.contains_key("poster_url"));
-    }
-
-    #[test]
-    fn isbn_cover_prefers_an_isbn13_and_ignores_junk() {
-        let mut d = SearchDoc::default();
-        d.isbn = vec!["not-an-isbn".into(), "0306406152".into(), "9780306406157".into()];
-        assert_eq!(
-            d.isbn_cover_url().as_deref(),
-            Some("https://covers.openlibrary.org/b/isbn/9780306406157-L.jpg?default=false")
-        );
-        d.isbn = vec!["0306406152".into()];
-        assert!(d.isbn_cover_url().unwrap().contains("0306406152"));
-        d.isbn = vec!["x".into()];
-        assert_eq!(d.isbn_cover_url(), None);
-    }
-
-    #[tokio::test]
-    async fn a_named_work_without_cover_or_description_still_gets_its_card_but_a_search_does_not() {
-        let server = MockServer::start().await;
-        let doc = json!({
-            "key": "/works/OL39284292W",
-            "title": "Mother of Learning",
-            "author_name": ["nobody103"],
-            "first_publish_year": 2021,
-            "language": ["eng"],
-            "isbn": ["9781955252096"]
-        });
-        // The lookup by id and the free-text search both find the same index doc.
-        Mock::given(method("GET"))
-            .and(path("/search.json"))
-            .respond_with(ResponseTemplate::new(200).set_body_json(json!({"numFound": 1, "docs": [doc]})))
-            .mount(&server)
-            .await;
-        Mock::given(method("GET"))
-            .and(path("/works/OL39284292W.json"))
-            .respond_with(ResponseTemplate::new(200).set_body_json(json!({
-                "key": "/works/OL39284292W", "title": "Mother of Learning"
-            })))
-            .mount(&server)
-            .await;
-        let (p, _dir) = plugin_against(&server).await;
-
-        let by_id = p
-            .handle_query(
-                &q("", &[("openlibraryid", "OL39284292W"), ("contentKind", "book")]),
-                10,
-            )
-            .await
-            .unwrap();
-        assert_eq!(by_id.len(), 1, "the work page lookup yields a card");
-        assert_eq!(by_id[0].fields["title"], "Mother of Learning");
-        assert!(by_id[0].fields["poster_url"].contains("/b/isbn/9781955252096-L.jpg"));
-        assert!(!by_id[0].fields.keys().any(|k| k.starts_with("description")));
-
-        let by_search = p
-            .handle_query(&q("mother of learning", &[("fileType", "card"), ("contentKind", "book")]), 10)
-            .await
-            .unwrap();
-        assert!(by_search.is_empty(), "search rows keep the strict bar");
-    }
-
     #[test]
     fn echoes_browse_markers_but_no_other_filter() {
         let r = to_record(

feeder-plugin/book-feeder/src/openlibrary_client.rs

@@ -30,9 +30,6 @@ pub const DEFAULT_API_BASE: &str = "https://openlibrary.org";
 /// Cover image base. `-L` is the large rendition; the gateway seeds whatever
 /// `poster_url` names, so one size is published, not three.
 const COVERS_BASE: &str = "https://covers.openlibrary.org/b/id";
-/// Cover by ISBN. Open Library often holds an *edition's* cover for a work whose
-/// own `covers` list is empty.
-const COVERS_ISBN_BASE: &str = "https://covers.openlibrary.org/b/isbn";
 
 /// Config key (and env seed) for the operator contact sent in the User-Agent.
 pub const CONTACT_KEY: &str = "contact";
@@ -61,9 +58,6 @@ const RETRY_BACKOFF_MS: u64 = 250;
 /// The field selection for every doc-shaped request, written once so a card
 /// resolved by search and the same card resolved by id are byte-identical.
 const SEARCH_FIELDS: &str = "key,title,author_name,cover_i,first_publish_year,language";
-/// A lookup of one named work also asks for its ISBNs, so a cover can be found by
-/// ISBN. Not on the shared list: a search row would carry hundreds of ISBNs per doc.
-const DOC_BY_ID_FIELDS: &str = "key,title,author_name,cover_i,first_publish_year,language,isbn";
 
 /// Cache lifetimes.
 ///
@@ -205,10 +199,6 @@ pub struct SearchDoc {
     pub first_publish_year: Option<i32>,
     #[serde(default)]
     pub id_project_gutenberg: Vec<String>,
-    /// ISBNs of the work's editions. Only requested for an `openlibraryid`
-    /// lookup (see [`DOC_BY_ID_FIELDS`]); empty everywhere else.
-    #[serde(default)]
-    pub isbn: Vec<String>,
 }
 
 impl SearchDoc {
@@ -224,22 +214,6 @@ impl SearchDoc {
             .filter(|c| *c > 0)
             .map(|c| format!("{COVERS_BASE}/{c}-L.jpg"))
     }
-
-    /// A cover looked up by one of the work's ISBNs (an ISBN-13 when there is one),
-    /// for a work with no cover of its own.
-    ///
-    /// `?default=false` makes Open Library answer 404 when it has no image for the
-    /// ISBN, instead of a 1×1 placeholder that would be seeded as the poster.
-    pub fn isbn_cover_url(&self) -> Option<String> {
-        let digits_only = |s: &&String| s.chars().all(|c| c.is_ascii_digit());
-        let isbn = self
-            .isbn
-            .iter()
-            .filter(digits_only)
-            .find(|s| s.len() == 13)
-            .or_else(|| self.isbn.iter().filter(digits_only).find(|s| s.len() == 10))?;
-        Some(format!("{COVERS_ISBN_BASE}/{isbn}-L.jpg?default=false"))
-    }
 }
 
 /// What a work JSON contributes to a card.
@@ -466,13 +440,12 @@ impl OpenLibraryClient {
     ) -> Result<Option<SearchDoc>, GatewayError> {
         let body = self
             .get_cached(
-                // `key2`: entries cached before ISBNs were requested lack them.
+                &format!("openlibrary:key:{id}"),
-                &format!("openlibrary:key2:{id}"),
                 WORK_TTL_SECS,
                 "/search.json",
                 &[
                     ("q", format!("key:/works/{id}")),
-                    ("fields", DOC_BY_ID_FIELDS.into()),
+                    ("fields", SEARCH_FIELDS.into()),
                     ("limit", "1".into()),
                 ],
                 deadline,
@@ -520,7 +493,6 @@ impl OpenLibraryClient {
                 .and_then(|a| a.iter().filter_map(Value::as_i64).find(|c| *c > 0)),
             first_publish_year: None,
             id_project_gutenberg: Vec::new(),
-            isbn: Vec::new(),
         }))
     }
 
@@ -642,6 +614,248 @@ impl OpenLibraryClient {
         self.cache_put(&key, &serde_json::json!({ "work": work }));
         Ok(work)
     }
+
+    /// What Open Library records about an etext: the one work that lists it,
+    /// several works, or none.
+    ///
+    /// Same lookup as [`Self::gutenberg_work`], but it keeps the difference
+    /// between "several" and "none", which that one folds into `None`: only an
+    /// etext no work lists may be bound by title ([`Self::title_work`]). Several
+    /// works listing one etext is an unmerged duplicate, and guessing between
+    /// them is exactly what the identifier rule refuses.
+    pub async fn gutenberg_etext(
+        &self,
+        gutenberg_id: u64,
+        deadline: Duration,
+    ) -> Result<Etext, GatewayError> {
+        let key = format!("openlibrary:etext:{gutenberg_id}");
+        if let Some(c) = self.cache_get(&key) {
+            if let Some(etext) = Etext::from_cached(&c.body) {
+                let ttl = if matches!(etext, Etext::One(_)) {
+                    BINDING_HIT_TTL_SECS
+                } else {
+                    BINDING_MISS_TTL_SECS
+                };
+                if now_secs().saturating_sub(c.at) <= ttl {
+                    return Ok(etext);
+                }
+            }
+        }
+        let body = self
+            .get(
+                "/search.json",
+                &[
+                    ("q", format!("id_project_gutenberg:{gutenberg_id}")),
+                    ("fields", "key,id_project_gutenberg".into()),
+                    ("limit", "2".into()),
+                ],
+                deadline,
+            )
+            .await?;
+        let etext = etext_of(&body, gutenberg_id);
+        self.cache_put(&key, &etext.to_cached());
+        Ok(etext)
+    }
+
+    /// The work a book is bound to by **title and author**, for an etext Open
+    /// Library lists under no work.
+    ///
+    /// Open Library's `id_project_gutenberg` links are patchy even for the most
+    /// read books — none of the editions of *Frankenstein; or, The Modern
+    /// Prometheus* (OL450063W) lists etext 84 — so the identifier rule alone left
+    /// them unbound, and their book pages, which ask for `openlibraryid:OL…W`,
+    /// empty. This is the weaker proof the trust model already has a grade for:
+    /// the caller stamps `anchorMethod=title`, the gateway records
+    /// `matchedBy/openlibrary:OL…W`, and meta-read ranks such an edition after an
+    /// identifier-bound one.
+    ///
+    /// Same main title ([`main_title`]), an author with the same surname, and,
+    /// among those, the work with the most editions — Open Library holds many
+    /// small works titled *Frankenstein* by Mary Shelley (adaptations, annotated
+    /// and school editions), and the canonical one is the one most published.
+    /// `None` when nothing matches or two works tie.
+    pub async fn title_work(
+        &self,
+        title: &str,
+        author: &str,
+        deadline: Duration,
+    ) -> Result<Option<String>, GatewayError> {
+        let (want, surname) = (main_title(title), surname(author));
+        if want.is_empty() || surname.is_empty() {
+            return Ok(None);
+        }
+        let key = format!("openlibrary:titlework:{want}:{surname}");
+        if let Some(c) = self.cache_get(&key) {
+            let work = c.body.get("work").and_then(Value::as_str).map(str::to_string);
+            let ttl = if work.is_some() {
+                BINDING_HIT_TTL_SECS
+            } else {
+                BINDING_MISS_TTL_SECS
+            };
+            if now_secs().saturating_sub(c.at) <= ttl {
+                return Ok(work);
+            }
+        }
+        let body = self
+            .get(
+                "/search.json",
+                &[
+                    ("title", want.clone()),
+                    ("author", surname.clone()),
+                    ("fields", "key,title,author_name,edition_count".into()),
+                    ("limit", "20".into()),
+                ],
+                deadline,
+            )
+            .await?;
+        let work = best_title_work(&body, &want, &surname);
+        self.cache_put(&key, &serde_json::json!({ "work": work }));
+        Ok(work)
+    }
+}
+
+/// What Open Library records about one Gutenberg etext. See
+/// [`OpenLibraryClient::gutenberg_etext`].
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum Etext {
+    /// Exactly one work lists it: an identifier binding, graded Anchored.
+    One(String),
+    /// Several works list it: an unmerged duplicate, so no binding at all.
+    Several,
+    /// No work lists it: the one case a title binding may fill.
+    Unlisted,
+}
+
+impl Etext {
+    fn to_cached(&self) -> Value {
+        match self {
+            Etext::One(work) => serde_json::json!({ "etext": "one", "work": work }),
+            Etext::Several => serde_json::json!({ "etext": "several" }),
+            Etext::Unlisted => serde_json::json!({ "etext": "unlisted" }),
+        }
+    }
+
+    fn from_cached(v: &Value) -> Option<Self> {
+        match v.get("etext")?.as_str()? {
+            "one" => Some(Etext::One(v.get("work")?.as_str()?.to_string())),
+            "several" => Some(Etext::Several),
+            "unlisted" => Some(Etext::Unlisted),
+            _ => None,
+        }
+    }
+}
+
+fn etext_of(body: &Value, gutenberg_id: u64) -> Etext {
+    if let Some(work) = single_gutenberg_work(body, gutenberg_id) {
+        return Etext::One(work);
+    }
+    let docs = parse_docs(body.get("docs"));
+    let total = body
+        .get("numFound")
+        .or_else(|| body.get("num_found"))
+        .and_then(Value::as_u64)
+        .unwrap_or(docs.len() as u64);
+    if total == 0 && docs.is_empty() {
+        Etext::Unlisted
+    } else {
+        Etext::Several
+    }
+}
+
+/// The part of a title two catalogues agree on: before any subtitle (`;`, `:`,
+/// a bracket, or `, or`), lower-cased, letters and digits only. Gutenberg's
+/// *Frankenstein; Or, The Modern Prometheus* and Open Library's *Frankenstein;
+/// or, The Modern Prometheus* are both `frankenstein`.
+pub fn main_title(raw: &str) -> String {
+    let lower = raw.to_lowercase();
+    let head = lower
+        .split([';', ':', '(', '['])
+        .next()
+        .unwrap_or_default();
+    let head = head.find(", or").map_or(head, |i| &head[..i]);
+    head.split(|c: char| !c.is_alphanumeric())
+        .filter(|w| !w.is_empty())
+        .collect::<Vec<_>>()
+        .join(" ")
+}
+
+/// The surname a name ends on, lower-cased, letters and digits only: Gutendex
+/// writes `Shelley, Mary Wollstonecraft`, Open Library `Mary Shelley`; both give
+/// `shelley`.
+pub fn surname(raw: &str) -> String {
+    let raw = raw.trim();
+    let last = match raw.split_once(',') {
+        Some((last, _)) => last,
+        None => raw.rsplit(char::is_whitespace).next().unwrap_or_default(),
+    };
+    last.chars()
+        .filter(|c| c.is_alphanumeric())
+        .collect::<String>()
+        .to_lowercase()
+}
+
+/// Does `name` carry `surname`? Compared with the separators removed, so a
+/// compound surname (`Saint-Exupéry`) matches however it is spaced.
+fn author_has_surname(name: &str, surname: &str) -> bool {
+    let flat = name
+        .chars()
+        .filter(|c| c.is_alphanumeric())
+        .collect::<String>()
+        .to_lowercase();
+    !surname.is_empty() && flat.contains(surname)
+}
+
+/// One document of the title search, with the edition count the choice rests on.
+#[derive(Debug, Deserialize)]
+struct TitleDoc {
+    #[serde(default)]
+    key: String,
+    #[serde(default)]
+    title: Option<String>,
+    #[serde(default)]
+    author_name: Vec<String>,
+    #[serde(default)]
+    edition_count: Option<u64>,
+}
+
+/// Decide a title binding from a `title=…&author=…` response. See
+/// [`OpenLibraryClient::title_work`].
+fn best_title_work(body: &Value, want: &str, surname: &str) -> Option<String> {
+    let docs: Vec<TitleDoc> = body
+        .get("docs")
+        .and_then(Value::as_array)
+        .map(|a| {
+            a.iter()
+                .filter_map(|d| serde_json::from_value::<TitleDoc>(d.clone()).ok())
+                .collect()
+        })
+        .unwrap_or_default();
+    let mut best: Option<(u64, String)> = None;
+    let mut tie = false;
+    for d in docs {
+        let Some(id) = d.key.strip_prefix("/works/").filter(|id| is_work_id(id)) else {
+            continue;
+        };
+        if main_title(d.title.as_deref().unwrap_or_default()) != want {
+            continue;
+        }
+        if !d.author_name.iter().any(|a| author_has_surname(a, surname)) {
+            continue;
+        }
+        let n = d.edition_count.unwrap_or(0);
+        let top = best.as_ref().map(|(m, _)| *m);
+        if top.is_none_or(|m| n > m) {
+            best = Some((n, id.to_string()));
+            tie = false;
+        } else if top == Some(n) {
+            tie = true;
+        }
+    }
+    if tie {
+        None
+    } else {
+        best.map(|(_, id)| id)
+    }
 }
 
 /// Decide a binding from a `q=id_project_gutenberg:<id>` response.
@@ -698,6 +912,70 @@ fn clean_description(raw: &str) -> Option<String> {
 mod tests {
     use super::*;
 
+    #[test]
+    fn a_main_title_drops_the_subtitle_case_and_punctuation() {
+        assert_eq!(main_title("Frankenstein; Or, The Modern Prometheus"), "frankenstein");
+        assert_eq!(main_title("Frankenstein; or, The Modern Prometheus"), "frankenstein");
+        assert_eq!(main_title("Frankenstein, or the Modern Prometheus"), "frankenstein");
+        assert_eq!(main_title("Frankenstein (Oxford Playscripts)"), "frankenstein");
+        assert_eq!(main_title("Alice's Adventures in Wonderland"), "alice s adventures in wonderland");
+        assert_eq!(main_title(""), "");
+    }
+
+    #[test]
+    fn a_surname_reads_both_name_orders() {
+        assert_eq!(surname("Shelley, Mary Wollstonecraft"), "shelley");
+        assert_eq!(surname("Mary Shelley"), "shelley");
+        assert_eq!(surname("  Austen, Jane "), "austen");
+        assert_eq!(surname("Saint-Exupéry, Antoine de"), "saintexupéry");
+        assert!(author_has_surname("Antoine de Saint-Exupéry", "saintexupéry"));
+        assert!(!author_has_surname("Jason Cobley", "shelley"));
+    }
+
+    fn title_doc(key: &str, title: &str, author: &str, editions: u64) -> Value {
+        serde_json::json!({ "key": key, "title": title, "author_name": [author], "edition_count": editions })
+    }
+
+    #[test]
+    fn the_title_binding_takes_the_most_published_work_of_that_title_and_author() {
+        let body = serde_json::json!({ "docs": [
+            title_doc("/works/OL26860869W", "Frankenstein", "Mary Shelley", 84),
+            title_doc("/works/OL450063W", "Frankenstein; or, The Modern Prometheus", "Mary Shelley", 2187),
+            title_doc("/works/OL25595002W", "Mary Shelley's Frankenstein", "Mary Shelley", 9000),
+            title_doc("/works/OL15443537W", "Frankenstein", "Dean Koontz", 5000),
+        ]});
+        // The 9000-edition work has another title, the 5000 one another author.
+        assert_eq!(best_title_work(&body, "frankenstein", "shelley").as_deref(), Some("OL450063W"));
+    }
+
+    #[test]
+    fn a_tie_or_no_match_binds_nothing() {
+        let tie = serde_json::json!({ "docs": [
+            title_doc("/works/OL1W", "Frankenstein", "Mary Shelley", 10),
+            title_doc("/works/OL2W", "Frankenstein", "Mary Shelley", 10),
+        ]});
+        assert_eq!(best_title_work(&tie, "frankenstein", "shelley"), None);
+        let other = serde_json::json!({ "docs": [title_doc("/works/OL3W", "Dracula", "Bram Stoker", 99)] });
+        assert_eq!(best_title_work(&other, "frankenstein", "shelley"), None);
+        let not_a_work = serde_json::json!({ "docs": [title_doc("/books/OL4M", "Frankenstein", "Mary Shelley", 99)] });
+        assert_eq!(best_title_work(&not_a_work, "frankenstein", "shelley"), None);
+    }
+
+    #[test]
+    fn an_etext_is_one_several_or_unlisted() {
+        let one = serde_json::json!({ "numFound": 1, "docs": [{ "key": "/works/OL66554W", "id_project_gutenberg": ["1342"] }] });
+        assert_eq!(etext_of(&one, 1342), Etext::One("OL66554W".into()));
+        let several = serde_json::json!({ "numFound": 2, "docs": [
+            { "key": "/works/OL450063W", "id_project_gutenberg": ["84"] },
+            { "key": "/works/OL18388W", "id_project_gutenberg": ["84"] }
+        ]});
+        assert_eq!(etext_of(&several, 84), Etext::Several);
+        assert_eq!(etext_of(&serde_json::json!({ "numFound": 0, "docs": [] }), 84), Etext::Unlisted);
+        for e in [Etext::One("OL1W".into()), Etext::Several, Etext::Unlisted] {
+            assert_eq!(Etext::from_cached(&e.to_cached()), Some(e));
+        }
+    }
+
     #[test]
     fn work_ids_are_works_only() {
         assert!(is_work_id("OL66554W"));