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

@@ -0,0 +1,88 @@
+# Gitea twin of .github/workflows/publish.yml. Gitea runs only .gitea/workflows/ once this
+# 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.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.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
+# certificate without that CA, so the login is written straight into ~/.docker/config.json.
+# The runner's job image (node:22-bookworm) has no docker CLI, so the static one is
+# installed first; buildx installs itself. No `type=gha` cache: the runner has none.
+
+name: Publish image
+
+# Build and push the feeder image to the IPCR (ipcr-push-watch.nsl.sh):
+#   * every push to main  -> the moving `main` tag
+#   * every `v*` tag       -> semver tags (e.g. v1.0.0 -> 1.0.0, 1.0, 1)
+# The build context is the repo root (the workspace); `meta-feeder-sdk` is a
+# tag-pinned git dependency fetched during the build. The Dockerfile lives under
+# feeder-plugin/book-feeder/.
+on:
+  push:
+    branches: [main]
+    tags: ['v*']
+
+env:
+  REGISTRY: ipcr-push-85-17-246-67.ip.nsl.sh
+  IMAGE_NAME: metadec.eth/${{ github.event.repository.name }}
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install Docker CLI
+        run: curl -fsSL https://download.docker.com/linux/static/stable/x86_64/docker-28.5.1.tgz | tar -xz -C /usr/local/bin --strip-components=1 docker/docker
+
+      - name: Trust the Mesh Router CA
+        run: printf '%s\n' "$IPCR_CA_PEM" > /tmp/ipcr-ca.pem
+        env:
+          IPCR_CA_PEM: ${{ vars.IPCR_CA_PEM }}
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          version: latest
+          buildkitd-config-inline: |
+            [registry."${{ env.REGISTRY }}"]
+              ca = ["/tmp/ipcr-ca.pem"]
+
+      - name: Log in to IPCR
+        env:
+          IPCR_PUSH_TOKEN: ${{ secrets.IPCR_PUSH_TOKEN }}
+        run: |
+          mkdir -p ~/.docker
+          printf '{"auths":{"%s":{"auth":"%s"}}}\n' "$REGISTRY" "$(printf 'metadec.eth:%s' "$IPCR_PUSH_TOKEN" | base64 -w0)" > ~/.docker/config.json
+
+      - name: Extract image metadata
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          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
+
+      - name: Build and push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./feeder-plugin/book-feeder/Dockerfile
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}

.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

@@ -3,8 +3,8 @@ name: Publish image
 # Build and push the feeder image to the GitHub Container Registry (ghcr.io):
 #   * every push to main  -> the moving `main` tag
 #   * every `v*` tag       -> semver tags (e.g. v1.0.0 -> 1.0.0, 1.0, 1)
-# The build context is the repo root (the workspace) so the vendored
+# The build context is the repo root (the workspace); `meta-feeder-sdk` is a
-# `meta-feeder-sdk` path dependency resolves; the Dockerfile lives under
+# tag-pinned git dependency fetched during the build. The Dockerfile lives under
 # feeder-plugin/book-feeder/.
 on:
   push:
@@ -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

@@ -791,7 +791,8 @@ checksum = "f8ca58f447f06ed17d5fc4043ce1b10dd205e060fb3ce5b979b8ed8e59ff3f79"
 
 [[package]]
 name = "meta-feeder-sdk"
-version = "0.1.0"
+version = "1.7.0"
+source = "git+https://gitea-watch.nsl.sh/metastar/meta-feeder-sdk?tag=v1.7.0#027917c8cffe7ca43dadcf3b426cc5f69ecc443a"
 dependencies = [
  "anyhow",
  "async-trait",
@@ -800,13 +801,13 @@ dependencies = [
  "bytes",
  "futures",
  "hyper",
+ "libc",
  "redb",
  "regex",
  "reqwest",
  "serde",
  "serde_json",
  "sha2",
- "tempfile",
  "thiserror 1.0.69",
  "tokio",
  "tracing",

Cargo.toml

@@ -1,6 +1,10 @@
 [workspace]
 resolver = "2"
-members = ["crates/*", "feeder-plugin/*"]
+members = ["feeder-plugin/*"]
+# The vendored SDK copy is no longer built: book-feeder pins meta-feeder-sdk by
+# git tag. Excluded (not deleted) so a stray `cargo` run inside it does not try
+# to join this workspace.
+exclude = ["crates/meta-feeder-sdk"]
 
 [workspace.package]
 edition = "2021"

README.md

@@ -1,12 +1,23 @@
 # meta-feeder-book
 
-Book **feeder sidecar** for [MetaMesh](https://github.com/worph/meta-gateway) —
+Book **feeder sidecar** for [MetaMesh](https://github.com/worph/meta-gateway).
-surfaces public-domain books from **Project Gutenberg** (via the
+One binary, `book-feeder`, hosts two upstreams:
-[Gutendex](https://gutendex.com) API) into a meta-gateway.
 
-It is the proof-of-concept feeder: a single upstream, no preview-seeding, no
+| Upstream | Serves | What it is |
-BitTorrent — just enough to exercise the `meta-feeder-sdk` + feeder HTTP
+|----------|--------|------------|
-contract end to end.
+| `gutenberg` | `document` × `book` | Public-domain **editions** from Project Gutenberg (via the [Gutendex](https://gutendex.com) API), with an epub url-locator per record |
+| `openlibrary` | `card` × `book` | **Work cards** from [Open Library](https://openlibrary.org) — the book identity tier (`TMDB : video :: AniList : comics :: Open Library : books`) |
+
+The two meet through `openlibraryid`. A Gutenberg edition carries it only when
+Open Library itself records the link: exactly one Open Library work lists the
+etext's Project Gutenberg id (`search.json?q=id_project_gutenberg:<id>`). The
+edition then also carries the transient trust claim `anchored=true`,
+`anchorSource=openlibrary`, which the gateway stores as
+`anchoredBy/openlibrary:OL…W`. There is no title or author matching.
+
+An Open Library card is addressed `compute_card_cid("openlibrary", "OL…W")`,
+the bare work id, byte-identical to meta-read's `card_cid::openlibrary`. A work
+with no cover or no description gets no card.
 
 ## Role in MetaMesh
 
@@ -32,9 +43,16 @@ drives the contract:
 |---------|---------|-------|
 | `META_FEEDER_HTTP_LISTEN` | `0.0.0.0:8080` | HTTP listen address |
 | `META_FEEDER_STATE_DIR` | `/data/meta-feeder` | redb cache + state |
+| `OPENLIBRARY_CONTACT` | — | e-mail or URL sent in the Open Library User-Agent. First-boot seed for the `contact` config field; a value saved in the config UI wins |
+| `OPENLIBRARY_TOP_N` | `12` | cards per search |
+| `OPENLIBRARY_DISCOVERY_N` | `20` | cards per browse row (`popular:true` → Open Library trending, `fresh:true` → recently added) |
 | `RUST_LOG` | `info` | tracing filter |
 
-No upstream API key is required — Gutendex is open.
+No upstream API key is required. Open Library asks API clients to identify
+themselves, so set a contact in production: an identified client gets a larger
+request budget. Both upstreams share one Open Library rate budget. The config
+page edits the `gutenberg` upstream, and the contact saved there is read by
+both.
 
 ## Image
 
@@ -47,24 +65,32 @@ tag) and on `v*` tags (semver tags).
 
 ## Build locally
 
-The build context is the **repo root** (the Cargo workspace) so the vendored
+The build context is the **repo root** (the Cargo workspace).
-`meta-feeder-sdk` path dependency resolves:
+`meta-feeder-sdk` is a git dependency pinned by tag, so the builder needs
+network access to github.com:
 
 ```bash
 docker build -f feeder-plugin/book-feeder/Dockerfile -t ghcr.io/worph/meta-feeder-book:dev .
 ```
 
+Tests (`cargo` is not needed on the host; wiremock stands in for every upstream):
+
+```bash
+docker run --rm -v "$PWD":/src -w /src -e CARGO_TARGET_DIR=/target rust:1.89 \
+  cargo test -p book-feeder
+```
+
 ## Repo layout
 
-This repo is a self-contained Cargo workspace vendored out of the
-`meta-gateway` monorepo:
-
 ```
-Cargo.toml                      # workspace: members = crates/*, feeder-plugin/*
+Cargo.toml                      # workspace: members = feeder-plugin/*
-crates/meta-feeder-sdk/         # vendored shared feeder SDK
 feeder-plugin/book-feeder/      # this feeder's crate + Dockerfile
+  src/gutenberg.rs              #   `gutenberg` upstream
+  src/openlibrary.rs            #   `openlibrary` upstream (cards)
+  src/openlibrary_client.rs     #   Open Library client (budget, cache, binding lookup)
+crates/meta-feeder-sdk/         # old vendored SDK copy — no longer built (excluded)
 ```
 
-Upstream source of truth for the SDK and the feeder crate is
+`meta-feeder-sdk` comes from [`worph/meta-feeder-sdk`](https://github.com/worph/meta-feeder-sdk)
-[`worph/meta-gateway`](https://github.com/worph/meta-gateway); changes there are
+at the tag in `feeder-plugin/book-feeder/Cargo.toml`. An SDK change reaches
-vendored back into this repo.
+this feeder only when it is tagged and the pin is bumped.

feeder-plugin/book-feeder/Cargo.toml

@@ -6,9 +6,10 @@ rust-version.workspace = true
 license.workspace = true
 publish = false
 
-# Book feeder sidecar: Project Gutenberg (via Gutendex). The proof-of-concept
+# Book feeder sidecar, two upstreams:
-# feeder — single upstream, no preview-seeding, no BitTorrent — that validates
+# - `gutenberg` — Project Gutenberg editions (via Gutendex), the byte tier.
-# the meta-feeder-sdk + feeder HTTP contract end to end.
+# - `openlibrary` — Open Library work cards, the book identity tier
+#   (`TMDB : video :: AniList : comics :: Open Library : books`).
 
 [lib]
 name = "book_feeder"
@@ -19,7 +20,11 @@ name = "book-feeder"
 path = "src/main.rs"
 
 [dependencies]
-meta-feeder-sdk = { path = "../../crates/meta-feeder-sdk" }
+# Pinned by TAG, never by path. A path dep resolves OUTSIDE the crate, and the
+# 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://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/Dockerfile

@@ -1,12 +1,13 @@
 # syntax=docker/dockerfile:1
 # =============================================================================
-# book-feeder — Project Gutenberg feeder sidecar. Multi-stage build.
+# book-feeder — Project Gutenberg + Open Library feeder sidecar. Multi-stage
+# build; build context is this repo's root (the Cargo workspace).
 #
-# The build context must be `packages/meta-gateway/` (the workspace root) so
+# meta-feeder-sdk arrives as a git dependency (tag-pinned), so the vendored
-# the `meta-feeder-sdk` path dependency (../../crates/meta-feeder-sdk) resolves.
+# `crates/` copy is not copied in — `git` must be present in the builder for
-# Same BuildKit cache-mount strategy as the gateway Dockerfile: the cargo
+# cargo to fetch it. Same BuildKit cache-mount strategy as the gateway
-# registry/git and `target/` persist across builds. The release binary is
+# Dockerfile: the cargo registry/git and `target/` persist across builds. The
-# copied OUT of the target cache mount within the same RUN.
+# release binary is copied OUT of the target cache mount within the same RUN.
 # =============================================================================
 
 FROM rust:1.89-slim-bookworm AS builder
@@ -20,7 +21,6 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     && rm -rf /var/lib/apt/lists/*
 
 COPY Cargo.toml Cargo.lock rust-toolchain.toml ./
-COPY crates/ crates/
 COPY feeder-plugin/ feeder-plugin/
 
 RUN --mount=type=cache,id=cargo-registry,target=/usr/local/cargo/registry \

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

@@ -6,17 +6,24 @@
 //!   don't expose an epub variant fail `compute_outcomes` with a permanent
 //!   error (operator-visible; very rare in practice).
 //!
-//! No auth, no API key, no rate-limit configuration. Moved verbatim from the
+//! - Per-edition metadata: `GET {rdf_base}/cache/epub/{id}/pg{id}.rdf` for the
-//! gateway crate's `plugins/gutenberg.rs`; only the import paths changed
+//!   fields Gutendex does not carry (summary, release date, credits).
-//! (`crate::*`/`super::common` → `meta_feeder_sdk::*`, `GatewayPlugin` →
+//! - Work binding: Open Library's `search.json?q=id_project_gutenberg:{id}` —
-//! `FeederPlugin`).
+//!   see [`OpenLibraryClient::gutenberg_work`] and reading-model §2.4.
+//!
+//! No auth, no API key. Originally moved verbatim from the gateway crate's
+//! `plugins/gutenberg.rs`.
 
 use std::collections::{BTreeMap, HashMap};
 use std::path::Path;
+use std::sync::Arc;
 
 use async_trait::async_trait;
+use meta_feeder_sdk::budget::RateBudget;
 use meta_feeder_sdk::cache::MidhashCache;
 use meta_feeder_sdk::common;
+use meta_feeder_sdk::config::ConfigSchema;
+use meta_feeder_sdk::lang::normalize_lang_code;
 #[cfg(test)]
 use meta_feeder_sdk::plugin::HashKind;
 use meta_feeder_sdk::plugin::{upstream_id_field, ConfigError, FeederPlugin, HashOutcome};
@@ -25,10 +32,52 @@ use meta_feeder_sdk::types::{DiscoveryRecord, GatewayError, PluginHealth};
 use serde::Deserialize;
 use tracing::{debug, warn};
 
+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.
 const DEFAULT_BASE_URL: &str = "https://gutendex.com";
 
+/// Canonical host for the per-edition RDF. Overridable via
+/// [`GutenbergPlugin::with_rdf_base_url`] — until it was, every unit test that
+/// ran a query also fetched live RDF from gutenberg.org.
+const DEFAULT_RDF_BASE_URL: &str = "https://www.gutenberg.org";
+
+/// Upper bound on one binding lookup on the live search path, on top of the
+/// budget deadline. A slow Open Library must not hold a Gutenberg result that
+/// is already in hand.
+const BINDING_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(8);
+
 /// Format we treat as the canonical hashable bytes for a Gutenberg record.
 const EPUB_FORMAT: &str = "application/epub+zip";
 
@@ -64,7 +113,18 @@ const RETRY_BACKOFF_MS: u64 = 250;
 pub struct GutenbergPlugin {
     http: reqwest::Client,
     base_url: String,
+    rdf_base_url: String,
     cache: Option<MidhashCache>,
+    /// The Open Library binding client, built at `configure()`. `None` before
+    /// that — records are then simply unbound.
+    openlibrary: Option<OpenLibraryClient>,
+    /// Test hook for the Open Library API base. Never set in production.
+    openlibrary_base: Option<String>,
+    /// Shared with the `openlibrary` plugin by `main.rs`; built at
+    /// `configure()` when absent.
+    openlibrary_budget: Option<Arc<RateBudget>>,
+    /// The Open Library contact (see `openlibrary_client::load_contact`).
+    contact: Option<String>,
 }
 
 impl GutenbergPlugin {
@@ -80,10 +140,33 @@ impl GutenbergPlugin {
         Self {
             http,
             base_url,
+            rdf_base_url: DEFAULT_RDF_BASE_URL.to_string(),
             cache: None,
+            openlibrary: None,
+            openlibrary_base: None,
+            openlibrary_budget: None,
+            contact: openlibrary_client::env_contact(),
         }
     }
 
+    /// Fetch the per-edition RDF from `rdf_base_url` instead of gutenberg.org.
+    pub fn with_rdf_base_url(mut self, rdf_base_url: String) -> Self {
+        self.rdf_base_url = rdf_base_url;
+        self
+    }
+
+    /// Send the Open Library binding lookup to `base` — tests only.
+    pub fn with_openlibrary_base_url(mut self, base: String) -> Self {
+        self.openlibrary_base = Some(base);
+        self
+    }
+
+    /// Draw binding lookups from a budget shared with the `openlibrary` plugin.
+    pub fn with_openlibrary_budget(mut self, budget: Arc<RateBudget>) -> Self {
+        self.openlibrary_budget = Some(budget);
+        self
+    }
+
     fn cache(&self) -> Result<&MidhashCache, GatewayError> {
         common::require_cache(self.cache.as_ref(), "gutenberg")
     }
@@ -151,11 +234,13 @@ impl GutenbergPlugin {
         const BIBREC_CONCURRENCY: usize = 6;
         let cache = self.cache.clone();
         let http = self.http.clone();
+        let rdf_base = self.rdf_base_url.trim_end_matches('/').to_string();
         let ids: Vec<u64> = books.iter().map(|b| b.id).collect();
         stream::iter(ids.into_iter())
             .map(|id| {
                 let cache = cache.clone();
                 let http = http.clone();
+                let rdf_base = rdf_base.clone();
                 async move {
                     let record_id = id.to_string();
                     if let Some(cached) = cache
@@ -164,7 +249,7 @@ impl GutenbergPlugin {
                     {
                         return cached;
                     }
-                    let url = format!("https://www.gutenberg.org/cache/epub/{id}/pg{id}.rdf");
+                    let url = format!("{rdf_base}/cache/epub/{id}/pg{id}.rdf");
                     let resp = tokio::time::timeout(
                         std::time::Duration::from_secs(8),
                         http.get(&url).send(),
@@ -334,6 +419,86 @@ impl GutenbergPlugin {
             .collect()
             .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), or, for an etext Open Library lists
+    /// under no work, by title and author (`by_title`, graded Matched). Aligned
+    /// 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<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 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 wanted: Vec<(u64, String, String)> = books
+            .iter()
+            .map(|b| (b.id, b.title.clone(), b.authors.first().map(|a| a.name.clone()).unwrap_or_default()))
+            .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(binding)) => binding,
+                    Ok(Err(e)) => {
+                        debug!(
+                            target: "meta-share::gateway",
+                            upstream = "gutenberg",
+                            id,
+                            error = %e,
+                            "open library binding lookup failed; record left unbound"
+                        );
+                        None
+                    }
+                    Err(_) => {
+                        debug!(
+                            target: "meta-share::gateway",
+                            upstream = "gutenberg",
+                            id,
+                            "open library binding lookup timed out; record left unbound"
+                        );
+                        None
+                    }
+                }
+            })
+            .buffered(BINDING_CONCURRENCY)
+            .collect()
+            .await
+    }
 }
 
 impl Default for GutenbergPlugin {
@@ -349,7 +514,22 @@ impl FeederPlugin for GutenbergPlugin {
     }
 
     fn configure(&mut self, cache_dir: &Path) -> Result<(), ConfigError> {
-        self.cache = Some(common::open_midhash_cache(cache_dir, "gutenberg")?);
+        let cache = common::open_midhash_cache(cache_dir, "gutenberg")?;
+        // The binding cache rides this plugin's own redb (the `misc` table), so
+        // a binding survives a restart without touching the card plugin's store.
+        let contact = openlibrary_client::load_contact(cache_dir);
+        let budget = self
+            .openlibrary_budget
+            .clone()
+            .unwrap_or_else(|| openlibrary_client::budget(contact.is_some()));
+        self.openlibrary = Some(OpenLibraryClient::new(
+            budget,
+            Some(cache.clone()),
+            self.openlibrary_base.clone(),
+            contact.as_deref(),
+        ));
+        self.contact = contact;
+        self.cache = Some(cache);
         Ok(())
     }
 
@@ -386,7 +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 {
@@ -405,13 +594,19 @@ impl FeederPlugin for GutenbergPlugin {
             "search returned"
         );
         let raw: Vec<GutendexBook> = body.results.into_iter().take(max_results).collect();
-        let (cover_cids, bibrecs) =
+        let (cover_cids, bibrecs, works) = tokio::join!(
-            tokio::join!(self.fetch_cover_cids(&raw), self.fetch_bibrec_fields(&raw),);
+            self.fetch_cover_cids(&raw),
+            self.fetch_bibrec_fields(&raw),
+            self.fetch_openlibrary_works(&raw),
+        );
         Ok(raw
             .into_iter()
             .zip(cover_cids)
             .zip(bibrecs)
-            .map(|((book, cover_cid), bibrec)| into_discovery_record(book, cover_cid, bibrec))
+            .zip(works)
+            .map(|(((book, cover_cid), bibrec), work)| {
+                into_discovery_record(book, cover_cid, bibrec, work)
+            })
             .collect())
     }
 
@@ -438,7 +633,26 @@ impl FeederPlugin for GutenbergPlugin {
             .ok()
             .flatten()
             .unwrap_or_default();
-        let record = into_discovery_record(book, cover_cid, bibrec);
+        // 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) => bind(ol, book.id, &book.title, &author, openlibrary_client::SEARCH_DEADLINE)
+                .await
+                .unwrap_or_else(|e| {
+                    debug!(
+                        target: "meta-share::gateway",
+                        upstream = "gutenberg",
+                        record_id,
+                        error = %e,
+                        "open library binding lookup failed; record left unbound"
+                    );
+                    None
+                }),
+            None => None,
+        };
+        let record = into_discovery_record(book, cover_cid, bibrec, work);
 
         Ok(common::single_outcome(
             cid,
@@ -481,13 +695,30 @@ impl FeederPlugin for GutenbergPlugin {
     fn served_content_kinds(&self) -> &'static [&'static str] {
         &["book"]
     }
+
+    /// ⚠ Declared HERE, not only on `openlibrary`. The SDK's `/config` page
+    /// edits the lowest hosted upstream id, which is this one — so this is where
+    /// the operator sets the contact both upstreams send to Open Library.
+    fn config_schema(&self) -> ConfigSchema {
+        ConfigSchema {
+            fields: vec![openlibrary_client::contact_field()],
+        }
+    }
+
+    fn config_values(&self) -> serde_json::Value {
+        serde_json::json!({ "contact": self.contact.clone().unwrap_or_default() })
+    }
 }
 
 /// Convert a `GutendexBook` into our wire-level `DiscoveryRecord`.
+///
+/// `openlibrary_work` is the binding from [`bind`]: the one work Open Library
+/// 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<Binding>,
 ) -> DiscoveryRecord {
     let GutendexBook {
         id,
@@ -520,6 +751,8 @@ fn into_discovery_record(
     fields.insert(upstream_id_field("gutenberg"), id.to_string());
 
     if !authors.is_empty() {
+        // Legacy scalar, kept verbatim for existing readers: Gutenberg's own
+        // "Last, First" spelling, comma-joined.
         fields.insert(
             "author".into(),
             authors
@@ -529,9 +762,28 @@ fn into_discovery_record(
                 .join(", "),
         );
     }
+    // `authors/{name}` key-set (reading-model §3.2): one member per person, in
+    // display order, so one author compares equal across sources — Open Library
+    // and AniList both write "Jane Austen", never "Austen, Jane".
+    for a in &authors {
+        if let Some(name) = display_name(&a.name) {
+            fields.insert(format!("authors/{name}"), "true".into());
+        }
+    }
     if let Some(lang) = languages.first() {
         fields.insert("language".into(), iso639_1_to_2(lang).to_string());
     }
+    // `languages/{lang3}` key-set for EVERY language, in the stack's shared
+    // vocabulary (`fre`, where the legacy scalar above says `fra`). Two-letter
+    // codes only the local table knows (`la`, `ca`, `is`) are widened first;
+    // anything still not three letters is dropped rather than written as a
+    // member no `languages:` filter can match.
+    for lang in &languages {
+        let code = normalize_lang_code(iso639_1_to_2(&lang.trim().to_ascii_lowercase()));
+        if code.len() == 3 && code.bytes().all(|b| b.is_ascii_lowercase()) {
+            fields.insert(format!("languages/{code}"), "true".into());
+        }
+    }
     if !subjects.is_empty() {
         fields.insert("subjects".into(), subjects.join(", "));
     }
@@ -619,10 +871,46 @@ 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). On an
+    // identifier link Open Library itself records, the claim is the plain
+    // Anchored form: `anchored` + `anchorSource`, and NO `anchorMethod` — that
+    // marks a title match. The
+    // gateway transcribes the claim into `anchoredBy/openlibrary:OL…W` and
+    // strips the transient keys before storing.
+    //
+    // 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 {
         fields.entry(k).or_insert(v);
     }
 
+    // Registry spellings beside the legacy RDF keys, derived after the merge so
+    // a bibrec cached before these keys existed produces them too.
+    //
+    // `description/eng` from `pgterms:marc520` (kept as `summary`): Gutenberg's
+    // summaries are written in English whatever the book's language, and
+    // `description/{lang3}` is the form every card and client reads.
+    if let Some(summary) = fields.get("summary").cloned() {
+        fields.entry("description/eng".into()).or_insert(summary);
+    }
+    // `releasedate` (iso-date) beside the legacy camelCase `releaseDate`, only
+    // when the RDF value really is one.
+    if let Some(date) = fields.get("releaseDate").and_then(|d| iso_date(d)) {
+        fields.entry("releasedate".into()).or_insert(date);
+    }
+
     DiscoveryRecord {
         upstream_id: "gutenberg".into(),
         record_id,
@@ -630,6 +918,38 @@ fn into_discovery_record(
     }
 }
 
+/// Gutenberg's "Last, First" → "First Last", the display order every other
+/// writer of `authors/{name}` uses.
+///
+/// Only when there is **exactly one** comma. "King, Martin Luther, Jr." or a
+/// corporate name has no safe reordering, and a wrong one is worse than
+/// Gutenberg's own spelling, which is kept as is.
+fn display_name(raw: &str) -> Option<String> {
+    let name = raw.trim();
+    if name.is_empty() {
+        return None;
+    }
+    match name.split(',').map(str::trim).collect::<Vec<_>>().as_slice() {
+        [last, first] if !last.is_empty() && !first.is_empty() => Some(format!("{first} {last}")),
+        _ => Some(name.to_string()),
+    }
+}
+
+/// `raw` when it is an `iso-date` (`YYYY-MM-DD`, or a bare `YYYY`), else `None`.
+///
+/// Never padded: "1998" does not become "1998-01-01", which would claim a
+/// precision the RDF never had.
+fn iso_date(raw: &str) -> Option<String> {
+    let digits = |s: &str, n: usize| s.len() == n && s.bytes().all(|b| b.is_ascii_digit());
+    let t = raw.trim();
+    let ok = match t.split('-').collect::<Vec<_>>().as_slice() {
+        [y] => digits(y, 4),
+        [y, m, d] => digits(y, 4) && digits(m, 2) && digits(d, 2),
+        _ => false,
+    };
+    ok.then(|| t.to_string())
+}
+
 /// Parse a Project Gutenberg `pg{id}.rdf` document for per-edition metadata.
 fn parse_pgterms_rdf(xml: &str) -> BTreeMap<String, String> {
     let mut out = BTreeMap::new();
@@ -762,6 +1082,28 @@ mod rdf_parser_tests {
         let f = parse_pgterms_rdf("<html><body>not rdf</body></html>");
         assert!(f.is_empty());
     }
+
+    /// The legacy RDF keys stay; the registry spellings are derived beside them
+    /// on the record — including from a bibrec cached before those keys
+    /// existed, which holds only the legacy ones.
+    #[test]
+    fn summary_and_issued_date_also_arrive_under_registry_keys() {
+        let xml = r#"<rdf:RDF>
+            <dcterms:issued rdf:datatype="...">1998-06-01</dcterms:issued>
+            <pgterms:marc520>A classic novel of manners.</pgterms:marc520>
+          </rdf:RDF>"#;
+        let book: GutendexBook = serde_json::from_value(serde_json::json!({
+            "id": 1342,
+            "title": "Pride and Prejudice"
+        }))
+        .unwrap();
+        let r = into_discovery_record(book, None, parse_pgterms_rdf(xml), None);
+        let get = |k: &str| r.fields.get(k).map(String::as_str);
+        assert_eq!(get("summary"), Some("A classic novel of manners."));
+        assert_eq!(get("description/eng"), Some("A classic novel of manners."));
+        assert_eq!(get("releaseDate"), Some("1998-06-01"));
+        assert_eq!(get("releasedate"), Some("1998-06-01"));
+    }
 }
 
 /// ISO 639-1 (two-letter) → ISO 639-2/T (three-letter) for the ~20 most common
@@ -877,13 +1219,27 @@ mod tests {
         })
     }
 
+    /// Every upstream — Gutendex, the RDF host and Open Library — on the one mock
+    /// server, so no test reaches the internet. An unmounted path 404s, which
+    /// every optional lookup treats as "nothing there".
     fn configured_plugin_against(server: &MockServer) -> (GutenbergPlugin, tempfile::TempDir) {
         let dir = tempfile::tempdir().expect("tempdir");
-        let mut plugin = GutenbergPlugin::with_base_url(server.uri());
+        let mut plugin = GutenbergPlugin::with_base_url(server.uri())
+            .with_rdf_base_url(server.uri())
+            .with_openlibrary_base_url(server.uri())
+            .with_openlibrary_budget(RateBudget::new(100.0, 100.0));
         plugin.configure(dir.path()).expect("configure");
         (plugin, dir)
     }
 
+    /// Open Library's answer to `q=id_project_gutenberg:<id>`.
+    fn binding(id: u64, body: serde_json::Value) -> Mock {
+        Mock::given(method("GET"))
+            .and(path("/search.json"))
+            .and(query_param("q", format!("id_project_gutenberg:{id}")))
+            .respond_with(ResponseTemplate::new(200).set_body_json(body))
+    }
+
     /// ⚠ The search path must ask for the CANONICAL `/books/`.
     ///
     /// Gutendex 301s `/books` → `/books/`. Following a redirect is free in
@@ -1086,7 +1442,16 @@ mod tests {
             Some("https://www.gutenberg.org/ebooks/1342")
         );
         assert_eq!(r0.fields.get("author").map(String::as_str), Some("Austen, Jane"));
+        assert_eq!(
+            r0.fields.get("authors/Jane Austen").map(String::as_str),
+            Some("true")
+        );
+        assert!(!r0.fields.contains_key("authors/Austen, Jane"));
         assert_eq!(r0.fields.get("language").map(String::as_str), Some("eng"));
+        assert_eq!(r0.fields.get("languages/eng").map(String::as_str), Some("true"));
+        // No Open Library answer mounted: the record is returned, unbound.
+        assert!(!r0.fields.contains_key("openlibraryid"));
+        assert!(!r0.fields.contains_key("anchored"));
         assert_eq!(r0.fields.get("format").map(String::as_str), Some("epub"));
         assert_eq!(
             r0.fields.get("fileName").map(String::as_str),
@@ -1261,6 +1626,303 @@ mod tests {
         }
     }
 
+    #[test]
+    fn display_name_reorders_only_a_single_comma() {
+        assert_eq!(display_name("Austen, Jane").as_deref(), Some("Jane Austen"));
+        assert_eq!(
+            display_name("Shelley, Mary Wollstonecraft").as_deref(),
+            Some("Mary Wollstonecraft Shelley")
+        );
+        assert_eq!(
+            display_name("King, Martin Luther, Jr.").as_deref(),
+            Some("King, Martin Luther, Jr.")
+        );
+        assert_eq!(display_name("Homer").as_deref(), Some("Homer"));
+        assert_eq!(display_name("  "), None);
+    }
+
+    #[test]
+    fn every_language_becomes_a_lang3_key_set_member() {
+        let book: GutendexBook = serde_json::from_value(serde_json::json!({
+            "id": 1,
+            "title": "Polyglot",
+            "languages": ["en", "fr", "la", "xx"]
+        }))
+        .unwrap();
+        let r = into_discovery_record(book, None, BTreeMap::new(), None);
+        let members: Vec<&str> = r
+            .fields
+            .keys()
+            .filter_map(|k| k.strip_prefix("languages/"))
+            .collect();
+        // The SDK vocabulary (`fre`), every language, and nothing that is not
+        // three letters.
+        assert_eq!(members, vec!["eng", "fre", "lat"]);
+        // The legacy scalar is untouched.
+        assert_eq!(r.fields.get("language").map(String::as_str), Some("eng"));
+    }
+
+    #[test]
+    fn releasedate_is_written_only_for_an_iso_date() {
+        assert_eq!(iso_date("2008-06-27").as_deref(), Some("2008-06-27"));
+        assert_eq!(iso_date("1998").as_deref(), Some("1998"));
+        assert_eq!(iso_date("1998-06"), None);
+        assert_eq!(iso_date("June 1998"), None);
+    }
+
+    /// The RDF host is configurable, so the per-edition fields are testable
+    /// without gutenberg.org.
+    #[tokio::test]
+    async fn rdf_summary_and_issue_date_arrive_through_the_configured_rdf_host() {
+        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;
+        Mock::given(method("GET"))
+            .and(path("/cache/epub/1342/pg1342.rdf"))
+            .respond_with(ResponseTemplate::new(200).set_body_string(
+                r#"<rdf:RDF><pgterms:ebook rdf:about="ebooks/1342">
+                     <dcterms:issued rdf:datatype="http://www.w3.org/2001/XMLSchema#date">1998-06-01</dcterms:issued>
+                     <pgterms:marc520>"Pride and Prejudice" is a classic novel.</pgterms:marc520>
+                   </pgterms:ebook></rdf:RDF>"#,
+            ))
+            .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!(pride["description/eng"].contains("classic novel"));
+        assert_eq!(pride.get("releasedate").map(String::as_str), Some("1998-06-01"));
+        // No RDF mounted for 84: neither key, and the record still arrives.
+        assert!(!records[1].fields.contains_key("description/eng"));
+        assert!(!records[1].fields.contains_key("releasedate"));
+        server.verify().await;
+    }
+
+    #[tokio::test]
+    async fn exactly_one_open_library_work_binds_the_edition_with_an_anchored_claim() {
+        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", "42671", "45186"] }
+            ]}),
+        )
+        .mount(&server)
+        .await;
+        // Two works list etext 84: an unmerged duplicate, so no binding at all.
+        binding(
+            84,
+            serde_json::json!({ "numFound": 2, "docs": [
+                { "key": "/works/OL450063W", "id_project_gutenberg": ["84"] },
+                { "key": "/works/OL18388W", "id_project_gutenberg": ["84"] }
+            ]}),
+        )
+        .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");
+        assert_eq!(records.len(), 2);
+
+        let pride = &records[0].fields;
+        assert_eq!(pride.get("openlibraryid").map(String::as_str), Some("OL66554W"));
+        assert_eq!(pride.get("anchored").map(String::as_str), Some("true"));
+        assert_eq!(pride.get("anchorSource").map(String::as_str), Some("openlibrary"));
+        assert!(
+            !pride.contains_key("anchorMethod"),
+            "an identifier link Open Library records is not a title match"
+        );
+
+        let frankenstein = &records[1].fields;
+        for k in ["openlibraryid", "anchored", "anchorSource", "anchorMethod"] {
+            assert!(!frankenstein.contains_key(k), "ambiguous binding must not stamp {k}");
+        }
+    }
+
+    /// 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() {
+        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": 0, "docs": [] }))
+            .mount(&server)
+            .await;
+        Mock::given(method("GET"))
+            .and(path("/search.json"))
+            .and(query_param("q", "id_project_gutenberg:84"))
+            .respond_with(ResponseTemplate::new(500))
+            .mount(&server)
+            .await;
+
+        let (plugin, _dir) = configured_plugin_against(&server);
+        let records = plugin
+            .handle_query(&GatewayQuery::from_free_text("alice"), 50)
+            .await
+            .expect("an Open Library failure must not fail the query");
+        assert_eq!(records.len(), 2);
+        for r in &records {
+            assert!(!r.fields.contains_key("openlibraryid"));
+            assert!(!r.fields.contains_key("anchored"));
+        }
+    }
+
+    /// A book surfaced by every browse must not cost a lookup every browse.
+    #[tokio::test]
+    async fn a_binding_answer_is_looked_up_once_per_book() {
+        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"] }
+            ]}),
+        )
+        .expect(1)
+        .mount(&server)
+        .await;
+        binding(84, serde_json::json!({ "numFound": 0, "docs": [] }))
+            .expect(1)
+            .mount(&server)
+            .await;
+
+        let (plugin, _dir) = configured_plugin_against(&server);
+        for _ in 0..2 {
+            let records = plugin
+                .handle_query(&GatewayQuery::from_free_text("alice"), 50)
+                .await
+                .expect("handle_query");
+            assert_eq!(
+                records[0].fields.get("openlibraryid").map(String::as_str),
+                Some("OL66554W")
+            );
+        }
+        server.verify().await;
+    }
+
+    /// The SDK's config page shows the lowest upstream id — this one — so the
+    /// Open Library contact must be settable here.
+    #[test]
+    fn the_open_library_contact_is_configurable_on_this_upstream() {
+        let schema = GutenbergPlugin::new().config_schema();
+        assert!(schema.fields.iter().any(|f| f.key == "contact"));
+    }
+
     #[test]
     fn health_reflects_configure_state() {
         let plugin = GutenbergPlugin::new();

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

@@ -1,5 +1,13 @@
-//! `book-feeder` library surface — exposes the Gutenberg plugin so integration
+//! `book-feeder` library surface — exposes the plugins so integration tests
-//! tests (and, later, other binaries) can construct it. The binary
+//! (and, later, other binaries) can construct them. The binary (`main.rs`) is a
-//! (`main.rs`) is a thin wrapper over [`meta_feeder_sdk::serve_feeders`].
+//! thin wrapper over [`meta_feeder_sdk::serve_feeders`].
+//!
+//! Two upstreams:
+//! - [`gutenberg`] — Project Gutenberg editions, the byte tier (`document` × `book`);
+//! - [`openlibrary`] — Open Library work cards, the identity tier (`card` × `book`),
+//!   over the client in [`openlibrary_client`], which `gutenberg` also uses to
+//!   bind an edition to its work.
 
 pub mod gutenberg;
+pub mod openlibrary;
+pub mod openlibrary_client;

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

@@ -1,17 +1,27 @@
-//! `book-feeder` — Project Gutenberg feeder sidecar.
+//! `book-feeder` — the book tier's feeder sidecar.
 //!
-//! Hosts the [`gutenberg::GutenbergPlugin`] over the feeder HTTP contract via
+//! Hosts two plugins over the feeder HTTP contract via
-//! [`meta_feeder_sdk::serve_feeders`]. The gateway core reaches it server-to-
+//! [`meta_feeder_sdk::serve_feeders`]:
-//! server on the internal docker network.
+//! - [`GutenbergPlugin`] — Project Gutenberg editions (`document` × `book`);
+//! - [`OpenLibraryPlugin`] — Open Library work cards (`card` × `book`).
+//!
+//! The gateway core reaches it server-to-server on the internal docker network.
 //!
 //! Env:
 //! - `META_FEEDER_HTTP_LISTEN` — listen addr (default `0.0.0.0:8080`)
 //! - `META_FEEDER_STATE_DIR`   — per-plugin cache root (default `/data/meta-feeder`)
+//! - `OPENLIBRARY_CONTACT`     — e-mail/URL sent in the Open Library User-Agent;
+//!   first-boot seed for the `contact` config field (the saved field wins)
+//! - `OPENLIBRARY_TOP_N`, `OPENLIBRARY_DISCOVERY_N` — optional card result sizing
 //! - `RUST_LOG`                — tracing filter (default `info`)
 
 use std::net::SocketAddr;
+use std::path::Path;
 
 use book_feeder::gutenberg::GutenbergPlugin;
+use book_feeder::openlibrary::OpenLibraryPlugin;
+use book_feeder::openlibrary_client;
+use meta_feeder_sdk::plugin::FeederPlugin;
 use meta_feeder_sdk::serve_feeders;
 use tracing_subscriber::EnvFilter;
 
@@ -27,5 +37,16 @@ async fn main() -> anyhow::Result<()> {
     let state_dir =
         std::env::var("META_FEEDER_STATE_DIR").unwrap_or_else(|_| "/data/meta-feeder".to_string());
 
-    serve_feeders(vec![Box::new(GutenbergPlugin::new())], state_dir, listen).await
+    // ONE budget for both plugins: the Gutenberg → Open Library binding and the
+    // card plugin call the same upstream, which rate-limits per client, not per
+    // plugin. Two buckets would each think they had the whole allowance.
+    let contact =
+        openlibrary_client::load_contact(&Path::new(&state_dir).join("gateway").join("openlibrary"));
+    let budget = openlibrary_client::budget(contact.is_some());
+
+    let plugins: Vec<Box<dyn FeederPlugin>> = vec![
+        Box::new(GutenbergPlugin::new().with_openlibrary_budget(budget.clone())),
+        Box::new(OpenLibraryPlugin::new().with_budget(budget)),
+    ];
+    serve_feeders(plugins, state_dir, listen).await
 }

feeder-plugin/book-feeder/tests/contract.rs

@@ -1,24 +1,36 @@
 //! End-to-end contract test for the feeder HTTP surface: boots the real
-//! `serve_feeders` router (book-feeder hosting GutenbergPlugin) on an ephemeral
+//! `serve_feeders` router (book-feeder hosting GutenbergPlugin AND
-//! port, pointed at a wiremock Gutendex, and drives `/manifest`, `/query`, and
+//! OpenLibraryPlugin, as `main.rs` does) on an ephemeral port, pointed at one
-//! `/compute` over HTTP. Deterministic — no live Gutendex.
+//! wiremock server that plays Gutendex, the Gutenberg RDF host and Open Library,
+//! and drives `/manifest`, `/query`, and `/compute` over HTTP. Deterministic —
+//! no live upstream.
 
 use book_feeder::gutenberg::GutenbergPlugin;
+use book_feeder::openlibrary::OpenLibraryPlugin;
+use book_feeder::openlibrary_client;
 use meta_feeder_sdk::{
-    configure_plugins, hash, router, ComputeRequest, ComputeResponse, GatewayQuery, HashKindDto,
+    configure_plugins, hash, router, ComputeRequest, ComputeResponse, FeederPlugin, GatewayQuery,
-    ManifestResponse, QueryRequest, QueryResponse,
+    HashKindDto, ManifestResponse, QueryRequest, QueryResponse,
 };
-use wiremock::matchers::{method, path};
+use wiremock::matchers::{method, path, query_param};
 use wiremock::{Mock, MockServer, ResponseTemplate};
 
-/// Spawn the feeder router (Gutenberg pointed at `upstream`) on an ephemeral
+/// Spawn the feeder router (every upstream pointed at `upstream`) on an
-/// port; return its base URL. The TempDir is returned so the per-plugin cache
+/// ephemeral port; return its base URL. The TempDir is returned so the
-/// outlives the test.
+/// per-plugin cache outlives the test.
 async fn spawn_feeder(upstream: &MockServer) -> (String, tempfile::TempDir) {
     let dir = tempfile::tempdir().expect("tempdir");
-    let plugin = GutenbergPlugin::with_base_url(upstream.uri());
+    let budget = openlibrary_client::budget(false);
-    let plugins =
+    let plugins: Vec<Box<dyn FeederPlugin>> = vec![
-        configure_plugins(vec![Box::new(plugin)], dir.path()).expect("configure plugins");
+        Box::new(
+            GutenbergPlugin::with_base_url(upstream.uri())
+                .with_rdf_base_url(upstream.uri())
+                .with_openlibrary_base_url(upstream.uri())
+                .with_openlibrary_budget(budget.clone()),
+        ),
+        Box::new(OpenLibraryPlugin::with_api_base(upstream.uri()).with_budget(budget)),
+    ];
+    let plugins = configure_plugins(plugins, dir.path()).expect("configure plugins");
     let app = router(plugins, "book-feeder-test".to_string(), dir.path());
 
     let listener = tokio::net::TcpListener::bind("127.0.0.1:0")
@@ -44,7 +56,7 @@ fn books_json() -> serde_json::Value {
 }
 
 #[tokio::test]
-async fn manifest_lists_book_capabilities() {
+async fn manifest_lists_both_upstreams() {
     let upstream = MockServer::start().await;
     let (base, _dir) = spawn_feeder(&upstream).await;
 
@@ -57,21 +69,36 @@ async fn manifest_lists_book_capabilities() {
         .await
         .expect("decode manifest");
 
-    assert_eq!(manifest.plugins.len(), 1);
+    assert_eq!(manifest.plugins.len(), 2);
-    let p = &manifest.plugins[0];
+    let gutenberg = manifest.plugins.iter().find(|p| p.id == "gutenberg").expect("gutenberg");
-    assert_eq!(p.id, "gutenberg");
+    assert_eq!(gutenberg.served_file_types, vec!["document".to_string()]);
-    assert_eq!(p.served_file_types, vec!["document".to_string()]);
+    assert_eq!(gutenberg.served_content_kinds, vec!["book".to_string()]);
-    assert_eq!(p.served_content_kinds, vec!["book".to_string()]);
+    let openlibrary = manifest
+        .plugins
+        .iter()
+        .find(|p| p.id == "openlibrary")
+        .expect("openlibrary");
+    assert_eq!(openlibrary.served_file_types, vec!["card".to_string()]);
+    assert_eq!(openlibrary.served_content_kinds, vec!["book".to_string()]);
 }
 
 #[tokio::test]
-async fn query_returns_gutenberg_records() {
+async fn query_returns_gutenberg_records_with_their_open_library_claim() {
     let upstream = MockServer::start().await;
     Mock::given(method("GET"))
         .and(path("/books/"))
         .respond_with(ResponseTemplate::new(200).set_body_json(books_json()))
         .mount(&upstream)
         .await;
+    Mock::given(method("GET"))
+        .and(path("/search.json"))
+        .and(query_param("q", "id_project_gutenberg:1342"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
+            "numFound": 1,
+            "docs": [{ "key": "/works/OL66554W", "id_project_gutenberg": ["1342", "42671"] }]
+        })))
+        .mount(&upstream)
+        .await;
     let (base, _dir) = spawn_feeder(&upstream).await;
 
     let resp: QueryResponse = reqwest::Client::new()
@@ -90,9 +117,75 @@ async fn query_returns_gutenberg_records() {
 
     assert_eq!(resp.records.len(), 1);
     let r = &resp.records[0];
+    let f = |k: &str| r.fields.get(k).map(String::as_str);
     assert_eq!(r.upstream_id, "gutenberg");
-    assert_eq!(r.fields.get("gutenbergid").map(String::as_str), Some("1342"));
+    assert_eq!(f("gutenbergid"), Some("1342"));
-    assert_eq!(r.fields.get("contentKind").map(String::as_str), Some("book"));
+    assert_eq!(f("contentKind"), Some("book"));
+    assert_eq!(f("authors/Jane Austen"), Some("true"));
+    assert_eq!(f("languages/eng"), Some("true"));
+    // The transient claim the gateway transcribes into
+    // `anchoredBy/openlibrary:OL66554W`.
+    assert_eq!(f("openlibraryid"), Some("OL66554W"));
+    assert_eq!(f("anchored"), Some("true"));
+    assert_eq!(f("anchorSource"), Some("openlibrary"));
+    assert_eq!(f("anchorMethod"), None, "an identifier link is not a title match");
+}
+
+#[tokio::test]
+async fn query_returns_an_open_library_card() {
+    let upstream = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/search.json"))
+        .and(query_param("q", "pride"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
+            "numFound": 1,
+            "docs": [{
+                "key": "/works/OL66554W",
+                "title": "Pride and Prejudice",
+                "author_name": ["Jane Austen"],
+                "cover_i": 14348537,
+                "first_publish_year": 1813
+            }]
+        })))
+        .mount(&upstream)
+        .await;
+    Mock::given(method("GET"))
+        .and(path("/works/OL66554W.json"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
+            "key": "/works/OL66554W",
+            "description": "A novel of manners."
+        })))
+        .mount(&upstream)
+        .await;
+    let (base, _dir) = spawn_feeder(&upstream).await;
+
+    let mut query = GatewayQuery::from_free_text("pride");
+    query.filters.insert("fileType".into(), vec!["card".into()]);
+    let resp: QueryResponse = reqwest::Client::new()
+        .post(format!("{base}/query"))
+        .json(&QueryRequest {
+            upstream_id: "openlibrary".to_string(),
+            query,
+            max_results: 10,
+        })
+        .send()
+        .await
+        .expect("POST /query")
+        .json()
+        .await
+        .expect("decode query response");
+
+    assert_eq!(resp.records.len(), 1);
+    let r = &resp.records[0];
+    assert_eq!(r.upstream_id, "openlibrary");
+    assert_eq!(r.record_id, "OL66554W");
+    assert_eq!(r.fields.get("fileType").map(String::as_str), Some("card"));
+    assert_eq!(
+        r.fields
+            .get("cids/bagdsaaaubnxxazlonruwe4tboj4u6tbwgy2tkncx")
+            .map(String::as_str),
+        Some("true")
+    );
 }
 
 #[tokio::test]
@@ -151,6 +244,57 @@ async fn compute_returns_outcome_with_inline_epub_bytes() {
     assert_eq!(hash::compute_ipfs_cid(&decoded), expected_cid);
 }
 
+#[tokio::test]
+async fn compute_for_an_open_library_card_is_a_byteless_card_locator() {
+    let upstream = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/search.json"))
+        .and(query_param("q", "key:/works/OL45804W"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
+            "numFound": 1,
+            "docs": [{
+                "key": "/works/OL45804W",
+                "title": "Fantastic Mr Fox",
+                "author_name": ["Roald Dahl"],
+                "cover_i": 6498519,
+                "first_publish_year": 1970
+            }]
+        })))
+        .mount(&upstream)
+        .await;
+    Mock::given(method("GET"))
+        .and(path("/works/OL45804W.json"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
+            "key": "/works/OL45804W",
+            "description": "A clever fox outwits three farmers."
+        })))
+        .mount(&upstream)
+        .await;
+    let (base, _dir) = spawn_feeder(&upstream).await;
+
+    let resp: ComputeResponse = reqwest::Client::new()
+        .post(format!("{base}/compute"))
+        .json(&ComputeRequest {
+            upstream_id: "openlibrary".to_string(),
+            record_id: "OL45804W".to_string(),
+        })
+        .send()
+        .await
+        .expect("POST /compute")
+        .json()
+        .await
+        .expect("decode compute response");
+
+    assert_eq!(resp.outcomes.len(), 1);
+    let o = &resp.outcomes[0];
+    assert_eq!(o.hash, "bagdsaaaubnxxazlonruwe4tboj4u6tbugu4dancx");
+    assert_eq!(o.hash_kind, HashKindDto::CardLocator);
+    assert!(o.bytes_b64.is_none());
+    assert!(o.file_extension.is_none());
+    let record = o.record.as_ref().expect("the card record rides the outcome");
+    assert_eq!(record.fields.get("title").map(String::as_str), Some("Fantastic Mr Fox"));
+}
+
 #[tokio::test]
 async fn query_unknown_upstream_is_404() {
     let upstream = MockServer::start().await;