diff --git a/web/src/lib/nostr-build.ts b/web/src/lib/nostr-build.ts
index b7d080b..3f0b1ee 100644
--- a/web/src/lib/nostr-build.ts
+++ b/web/src/lib/nostr-build.ts
@@ -55,19 +55,6 @@ export interface BuildReview {
  * characters or matching a common test string do not qualify. They still appear in
  * full on the mint page, nothing is hidden, this strip is just a shop window.
  */
-/** How many candidates to gather per card shown, so the ordering has a real choice. */
-const POOL_FACTOR = 12;
-
-/**
- * How recent a review must still be to be promoted for having a named author.
- *
- * The ordering below prefers reviews whose author has a kind 0, and this is the leash
- * on that preference. Without it the strip would headline a named review from two
- * years ago under a heading that says "Latest reviews"; with it, a named review from
- * within the season can lead and anything older cannot.
- */
-const PROMOTE_MAX_AGE_S = 90 * 24 * 60 * 60;
-
 const TEST_STRINGS = new Set([
   'test', 'testing', 'test test', 'hello', 'hi', 'ok', 'okay', 'good', 'nice',
   'great', 'cool', 'gm', 'a', '.', '..', '...', 'asdf', 'qwerty', '123',
@@ -227,29 +214,30 @@ export async function fetchLatestReviews(
         profile: null,
       });
 
-      if (candidates.length >= limit * POOL_FACTOR) break;
+      // The scan runs newest first, so the first `limit` survivors are the latest
+      // `limit` reviews. Nothing further down the list can outrank them.
+      if (candidates.length >= limit) break;
     }
 
     const profiles = await fetchProfiles(candidates.map((c) => c.pubkey));
     for (const review of candidates) review.profile = profiles.get(review.pubkey) ?? null;
 
     /*
-     * Recent reviews with a named author come first; everything else stays newest
-     * first behind them.
+     * Strict recency, newest first, as the last thing that happens to this list.
      *
-     * Most review keys have published exactly one event in their life: the review
-     * itself. Nothing can be fetched for them, so a strict recency order fills this
-     * strip with anonymous npubs while a named review sits a few rows below the
-     * cut. This is the home page shop window, which already drops junk bodies, so
-     * it prefers a review a reader can attach a person to. Nothing is hidden:
-     * every review is on its mint page, and PROMOTE_MAX_AGE_S keeps "latest"
-     * meaning latest.
+     * The heading says "Latest reviews", so the order under it is the event
+     * `created_at` and nothing else. This used to float reviews whose author had a
+     * kind 0 to the front, which read as broken on the page: a named review from two
+     * months ago sat between two reviews from this week, and the card feet said so,
+     * because the "2mo ago" label is formatted from the very same `created_at` this
+     * sorts on. Author names are still resolved and still shown — they just no longer
+     * decide the order.
+     *
+     * Sorted here rather than left to the scan above, so that the invariant holds
+     * whatever the gathering loop does later, and `id` breaks ties between two events
+     * that share a second so two builds of the same events agree.
      */
-    const cutoff = Math.floor(Date.now() / 1000) - PROMOTE_MAX_AGE_S;
-    const promoted = (review: BuildReview): number =>
-      review.profile?.found && review.created_at >= cutoff ? 1 : 0;
-
-    candidates.sort((a, b) => promoted(b) - promoted(a) || b.created_at - a.created_at);
+    candidates.sort((a, b) => b.created_at - a.created_at || a.id.localeCompare(b.id));
 
     return candidates.slice(0, limit);
   } catch {