diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 154121e670..312fe701a0 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -387,6 +387,36 @@ jobs:
 
           echo "✅ Production assets built successfully"
 
+      - name: Prepare and seed benchmark database
+        # The Pro Rails suite serves DB-backed routes (e.g. /posts_page). Without
+        # a migrated + seeded database those routes 500 on every request (the
+        # `posts` table does not exist), so the benchmark must create and seed it
+        # before the server starts. On a fresh CI runner `db:prepare` creates the
+        # database, loads the schema, and — because the database is newly created —
+        # runs db/seeds.rb (deterministic and faker-free so it works under
+        # RAILS_ENV=production). Note: `db:prepare` only seeds on first creation, so
+        # the row-count check below fails the job loudly if seeding was skipped (an
+        # empty table would otherwise serve a 200 "No posts found" payload that the
+        # benchmark scores as healthy).
+        if: matrix.server_kind == 'rails' && matrix.pro_env == 'true'
+        working-directory: ${{ matrix.app_directory }}
+        # Pass the matrix value via env (treated as data, not interpolated into
+        # the shell), matching the "Execute benchmark suite" step below.
+        env:
+          RAILS_ENV: production
+          NODE_ENV: production
+          SUITE_NAME: ${{ matrix.suite_name }}
+        run: |
+          set -euo pipefail
+          echo "🌱 Preparing and seeding $SUITE_NAME benchmark database..."
+          bundle exec rails db:prepare
+          # Guard every table `/posts_page` reads (posts, plus the users and
+          # comments it joins), not just posts: a partial seed would otherwise
+          # let the route 500 (or render an empty page) while the job stayed
+          # green. `set -o pipefail` keeps this abort fatal even if the command
+          # is ever piped.
+          bundle exec rails runner 'counts = { posts: Post.count, users: User.count, comments: Comment.count }; empty = counts.select { |_, c| c.zero? }.keys; abort("Benchmark DB seeding incomplete after db:prepare (empty: #{empty.join(", ")}); seeds did not run") if empty.any?; puts "✅ Database seeded (#{counts.map { |k, v| "#{v} #{k}" }.join(", ")})"'
+
       - name: Start Pro node renderer
         if: matrix.server_kind == 'node-renderer'
         working-directory: ${{ matrix.app_directory }}
diff --git a/react_on_rails_pro/spec/dummy/db/seeds.rb b/react_on_rails_pro/spec/dummy/db/seeds.rb
index 980f5288df..017660c43b 100644
--- a/react_on_rails_pro/spec/dummy/db/seeds.rb
+++ b/react_on_rails_pro/spec/dummy/db/seeds.rb
@@ -1,48 +1,70 @@
 # frozen_string_literal: true
 
-# This file should contain all the record creation needed to seed the database with its default values.
-# The data can then be loaded with the rake db:seed (or created alongside the db with db:setup).
+# Seed data for the dummy app. The benchmark suite's `/posts_page` route renders
+# these records, and they are handy for local development too.
 #
-# Examples:
-#
-#   cities = City.create([{ name: 'Chicago' }, { name: 'Copenhagen' }])
-#   Mayor.create(name: 'Emanuel', city: cities.first)
+# This file is intentionally deterministic and free of the `faker` gem: it must
+# run under `RAILS_ENV=production` in CI (via `rails db:prepare`), where the
+# development/test-only `faker` gem is not loaded. Deterministic data also keeps
+# benchmark runs reproducible from one CI run to the next.
+
+# Local variables (not top-level constants) so re-running `rails db:seed` in an
+# already-loaded process — seeds.rb is `load`ed, not `require`d — does not emit
+# `warning: already initialized constant`. Per-record counts index into these
+# arrays, so cycling stays correct regardless of each range's bounds or step.
+user_count = 10
+post_counts = (3..7).to_a
+comment_counts = (2..5).to_a
+
+lorem = %w[
+  lorem ipsum dolor sit amet consectetur adipiscing elit sed do eiusmod tempor
+  incididunt ut labore et dolore magna aliqua enim ad minim veniam quis nostrud
+  exercitation ullamco laboris nisi aliquip ex ea commodo consequat duis aute
+  irure reprehenderit voluptate velit esse cillum fugiat nulla pariatur
+].freeze
+
+lorem_words = lambda do |count, offset|
+  Array.new(count) { |i| lorem[(offset + i) % lorem.size] }
+end
+lorem_sentence = ->(word_count, offset) { "#{lorem_words.call(word_count, offset).join(' ').capitalize}." }
+lorem_paragraph = lambda do |sentence_count, offset|
+  Array.new(sentence_count) { |i| lorem_sentence.call(6 + ((offset + i) % 6), offset + (i * 7)) }.join(" ")
+end
 
-# Clear existing data
 puts "Clearing existing data..."
 Comment.delete_all
 Post.delete_all
 User.delete_all
 
-# Create Users
 puts "Creating users..."
-10.times do
-  User.create!(
-    name: Faker::Name.name,
-    email: Faker::Internet.unique.email
-  )
+users = Array.new(user_count) do |i|
+  User.create!(name: "User #{i + 1}", email: "user-#{i + 1}@example.com")
 end
 
-# Create Posts
 puts "Creating posts..."
-User.all.each do |user|
-  rand(3..7).times do
-    user.posts.create!(
-      title: Faker::Lorem.sentence(word_count: 3),
-      body: Faker::Lorem.paragraphs(number: 3).join("\n\n")
+posts = []
+users.each_with_index do |user, user_index|
+  post_count = post_counts[user_index % post_counts.size]
+  post_count.times do |post_index|
+    seed = (user_index * 11) + post_index
+    posts << user.posts.create!(
+      title: lorem_sentence.call(3, seed),
+      body: Array.new(3) { |p| lorem_paragraph.call(4, seed + p) }.join("\n\n")
     )
   end
 end
 
-# Create Comments
 puts "Creating comments..."
-Post.all.each do |post|
-  rand(2..5).times do
+posts.each_with_index do |post, post_index|
+  comment_count = comment_counts[post_index % comment_counts.size]
+  comment_count.times do |comment_index|
+    seed = (post_index * 7) + comment_index
     post.comments.create!(
-      user: User.all.sample,
-      body: Faker::Lorem.paragraph
+      user: users[seed % users.size],
+      body: lorem_paragraph.call(2, seed)
     )
   end
 end
 
-puts "Seed data created successfully!"
+puts "Seed data created successfully! " \
+     "(#{User.count} users, #{Post.count} posts, #{Comment.count} comments)"
diff --git a/react_on_rails_pro/spec/dummy/spec/db/seeds_spec.rb b/react_on_rails_pro/spec/dummy/spec/db/seeds_spec.rb
new file mode 100644
index 0000000000..2cbc419110
--- /dev/null
+++ b/react_on_rails_pro/spec/dummy/spec/db/seeds_spec.rb
@@ -0,0 +1,66 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+
+# Unit coverage for `db/seeds.rb`, the layer that actually broke in issue #3602.
+#
+# The Pro benchmark workflow runs `rails db:prepare` under RAILS_ENV=production,
+# where the development/test-only `faker` gem is NOT in the bundle. If seeds.rb
+# reintroduces a `Faker::...` call (or otherwise stops populating the tables),
+# the production CI run leaves `/posts_page` with no data — exactly the #3602
+# failure. These examples reproduce that environment by hiding the `Faker`
+# constant while loading the seed file, so such a regression fails here (on every
+# Pro dummy spec run) instead of silently in the benchmark suite.
+#
+# Unlike posts_page_spec.rb this needs no node renderer — it only touches the DB.
+RSpec.describe "db/seeds.rb" do
+  before do
+    # maintain_test_schema! is disabled in this suite (see posts_page_spec.rb), so
+    # load the schema if the tables the seeds touch are missing.
+    unless %i[users posts comments].all? { |t| ActiveRecord::Base.connection.table_exists?(t) }
+      ActiveRecord::Schema.verbose = false
+      load Rails.root.join("db/schema.rb")
+    end
+  end
+
+  after do
+    Comment.delete_all
+    Post.delete_all
+    User.delete_all
+  end
+
+  # seeds.rb is `load`ed (not `require`d) in production via db:prepare; do the
+  # same here. It clears the tables itself, so each load starts from a clean slate.
+  # Its progress `puts` output is captured to keep spec output quiet.
+  def load_seeds
+    original_stdout = $stdout
+    $stdout = StringIO.new
+    load Rails.root.join("db/seeds.rb")
+  ensure
+    $stdout = original_stdout
+  end
+
+  it "populates users, posts, and comments without the faker gem" do
+    hide_const("Faker")
+
+    expect { load_seeds }.not_to raise_error
+
+    expect(User.count).to be_positive
+    expect(Post.count).to be_positive
+    expect(Comment.count).to be_positive
+  end
+
+  it "produces identical data on every load (deterministic, reproducible benchmarks)" do
+    hide_const("Faker")
+
+    load_seeds
+    first = { users: User.count, posts: Post.count, comments: Comment.count,
+              titles: Post.order(:id).pluck(:title) }
+
+    load_seeds
+    second = { users: User.count, posts: Post.count, comments: Comment.count,
+               titles: Post.order(:id).pluck(:title) }
+
+    expect(second).to eq(first)
+  end
+end
diff --git a/react_on_rails_pro/spec/dummy/spec/requests/posts_page_spec.rb b/react_on_rails_pro/spec/dummy/spec/requests/posts_page_spec.rb
new file mode 100644
index 0000000000..29d0cd9a34
--- /dev/null
+++ b/react_on_rails_pro/spec/dummy/spec/requests/posts_page_spec.rb
@@ -0,0 +1,74 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+
+# Integration coverage for the `/posts_page` benchmark route (issue #3602).
+#
+# `/posts_page` renders DB records synchronously (non-streaming), so when the
+# `posts` table was missing in the Pro benchmark suite it surfaced as a hard 500
+# on every request. (The streaming RSC posts routes mask that class of failure
+# because their HTTP 200 status line is flushed before the body errors.)
+#
+# These examples do not reproduce the missing-table case itself — the `before`
+# block loads the schema (see below), so the table is always present here. The
+# missing-table 500 is prevented upstream by seeding the benchmark DB before the
+# server starts (.github/workflows/benchmark.yml). What these examples guard is
+# the layer above that: with the table in place the route must server-render
+# seeded posts and return 200, and an empty table must return 200 "No posts
+# found" rather than 500.
+#
+# Requires the Pro node renderer to be running, like the other server-rendering
+# request specs: the page is rendered with `prerender: true`.
+RSpec.describe "Posts page", :server_rendering do
+  before do
+    # The suite intentionally leaves `maintain_test_schema!` disabled, so make
+    # this DB-backed spec self-sufficient: load the schema if it is not present
+    # yet. Check every table the examples touch (not just `posts`) so a partial
+    # schema can't slip past the guard.
+    unless %i[users posts comments].all? { |t| ActiveRecord::Base.connection.table_exists?(t) }
+      ActiveRecord::Schema.verbose = false
+      load Rails.root.join("db/schema.rb")
+    end
+
+    Comment.delete_all
+    Post.delete_all
+    User.delete_all
+
+    2.times do |i|
+      user = User.create!(name: "User #{i + 1}", email: "user-#{i + 1}@example.com")
+      post = user.posts.create!(title: "Sentinel Post #{i + 1}", body: "Body of sentinel post #{i + 1}.")
+      post.comments.create!(user: user, body: "Comment on sentinel post #{i + 1}.")
+    end
+  end
+
+  after do
+    Comment.delete_all
+    Post.delete_all
+    User.delete_all
+  end
+
+  it "server-renders the seeded posts and returns 200" do
+    get "/posts_page"
+
+    expect(response).to have_http_status(:ok)
+
+    # Parse for the heading so we assert "Posts Page" renders specifically in an
+    # <h1> (the page chrome), not just anywhere in the body. The seeded titles
+    # below are plain-text content, so a raw substring match is sufficient there.
+    html = Nokogiri::HTML(response.body)
+    expect(html.css("h1").map(&:text)).to include("Posts Page")
+    expect(response.body).to include("Sentinel Post 1")
+    expect(response.body).to include("Sentinel Post 2")
+  end
+
+  it "returns 200 (not 500) when there are no posts to render" do
+    Comment.delete_all
+    Post.delete_all
+    User.delete_all
+
+    get "/posts_page"
+
+    expect(response).to have_http_status(:ok)
+    expect(response.body).to include("No posts found")
+  end
+end