diff --git a/src-tauri/src/commands.rs b/src-tauri/src/commands.rs
index 5b200ab..0403b54 100644
--- a/src-tauri/src/commands.rs
+++ b/src-tauri/src/commands.rs
@@ -1545,7 +1545,16 @@ pub async fn catalog_artist_socials(
 // v4: the alpha ramp now scales to the measured ink distance, so every cached
 // mask predates the anti-aliasing this pipeline relies on and must be redrawn.
 const COVER_LOGO_RECIPE: &str = "cover-logo-v4-ink-scaled-alpha";
-const COVER_LOGO_MAX_COVERS: usize = 14;
+// A prolific artist's wordmark is often nowhere near the top of this list. The
+// rank puts albums first and penalises `remix|mix|edit|live` titles — but for
+// Poolside (74 releases) every one of the SEVEN covers that OCR reads as
+// "POOLSIDE" is a remix single, so all seven sat outside a 14-cover window and
+// the artist silently lost its logo. Raising the cap costs nothing for artists
+// that hit early: the locate loop stops as soon as COVER_LOGO_MAX_HITS is
+// reached, so the extra covers are only ever fetched when the early ones miss.
+// The price is paid by artists with no wordmark anywhere, who now OCR more
+// covers before giving up — once, then it is cached.
+const COVER_LOGO_MAX_COVERS: usize = 40;
 const COVER_LOGO_MAX_HITS: usize = 5;
 const COVER_LOGO_LOCATE_WINDOW: usize = 4;
 const COVER_LOGO_UPGRADE_WINDOW: usize = 4;
diff --git a/src-tauri/src/cover_logo.rs b/src-tauri/src/cover_logo.rs
index 8b63fc2..2c06727 100644
--- a/src-tauri/src/cover_logo.rs
+++ b/src-tauri/src/cover_logo.rs
@@ -45,6 +45,10 @@ fn normalize(s: &str) -> String {
 // Segmentation tuning (RGB distance space, 0..~441).
 const BG_NEAR_DIST: f32 = 46.0; // a pixel this close to the dominant colour counts as background
 const BG_MIN_DOMINANCE: f32 = 0.5; // background must fill at least this share of the crop, else it's a photo → reject
+/// Cover long edge the background histogram and dominance guard are calibrated
+/// at. Larger sources are normalised down for that test only — see the note at
+/// the guard for why its verdict is otherwise resolution-dependent.
+const DOMINANCE_REF_COVER: f32 = 1000.0;
 const ALPHA_LO: f32 = 46.0; // distance-from-bg below which a pixel is background (transparent)
 const ALPHA_HI: f32 = 120.0; // floor for the opaque end of the ramp (low-contrast marks)
 /// Share of the measured ink-vs-background distance the alpha ramp spans. The
@@ -123,11 +127,36 @@ pub(crate) fn extract_wordmark(
     let crop = image.crop_imm(x0, y0, x1 - x0, y1 - y0).to_rgba8();
     let (w, h) = crop.dimensions();
 
+    // Both the histogram and the dominance guard read a copy normalised to the
+    // cover resolution they were tuned at.
+    //
+    // They are scale-sensitive in opposite directions to intuition: a coarse 16³
+    // histogram over a BIGGER source resolves more film grain and JPEG texture,
+    // which spreads pixels across bins and shrinks the dominant colour's share.
+    // Alvvays' "Blue Rev" measures dominance 0.746 at 1000px and 0.226 at 3000px
+    // for the identical crop — so once the extractor started receiving hi-res
+    // artwork, covers that had segmented cleanly for months began failing the
+    // 0.5 guard as "text over a photo". The guard did not change; its input grew.
+    //
+    // Normalising here keeps the verdict identical whatever resolution the
+    // upgrade path supplies, while the alpha ramp below still runs at FULL
+    // resolution, which is the whole point of fetching the larger cover.
+    let guard_scale = DOMINANCE_REF_COVER / (iw.max(ih) as f32);
+    let guard_copy = (guard_scale < 1.0).then(|| {
+        image::imageops::resize(
+            &crop,
+            ((w as f32 * guard_scale).round() as u32).max(1),
+            ((h as f32 * guard_scale).round() as u32).max(1),
+            image::imageops::FilterType::Triangle,
+        )
+    });
+    let guard = guard_copy.as_ref().unwrap_or(&crop);
+
     // Background = the dominant colour, via a coarse 16³ RGB histogram over the
     // whole crop. Robust to sparse texture (faint line-art on black) in a way a
     // ring mean is not — the text is a minority, so the mode is the background.
     let mut hist = [0u32; 16 * 16 * 16];
-    for p in crop.pixels() {
+    for p in guard.pixels() {
         let (r, g, b) = (p.0[0] >> 4, p.0[1] >> 4, p.0[2] >> 4);
         hist[(r as usize) * 256 + (g as usize) * 16 + b as usize] += 1;
     }
@@ -145,8 +174,8 @@ pub(crate) fn extract_wordmark(
     // Dominance guard: what share of the crop is near that background colour?
     // A clean wordmark has a dominant background (text is the minority); a photo
     // has no single dominant colour, so this share is low → reject.
-    let total = (w * h) as f32;
-    let near = crop
+    let total = (guard.width() * guard.height()) as f32;
+    let near = guard
         .pixels()
         .filter(|p| dist([p.0[0] as f32, p.0[1] as f32, p.0[2] as f32], bg) < BG_NEAR_DIST)
         .count() as f32;
diff --git a/src/lib/state.svelte.js b/src/lib/state.svelte.js
index 6621b74..950fb9b 100644
--- a/src/lib/state.svelte.js
+++ b/src/lib/state.svelte.js
@@ -1285,12 +1285,17 @@ function albumLogoRank(a) {
     return typeRank + noisy;
 }
 
+// Must not undercut the backend's own cap (COVER_LOGO_MAX_COVERS): whatever is
+// trimmed here the extractor can never see. A prolific artist's wordmark is
+// frequently on a low-ranked remix single — all seven of Poolside's readable
+// covers sat past position 14 — and the backend stops early once it has enough
+// hits, so a longer list costs nothing for artists whose top covers work.
 function coverLogoCandidates(page) {
     return [...(page.albums ?? [])]
         .sort((a, b) => albumLogoRank(a) - albumLogoRank(b))
         .map((a) => ({ title: a.title ?? "", url: a.cover }))
         .filter((c) => c.url)
-        .slice(0, 14);
+        .slice(0, 40);
 }
 
 function coverLogoInputKey(artistId, page, covers) {