diff --git a/.codecov.yml b/.codecov.yml
deleted file mode 100644
index 6adca0c9580..00000000000
--- a/.codecov.yml
+++ /dev/null
@@ -1,7 +0,0 @@
-comment: false
-coverage:
-  status:
-    project: false
-    patch: false
-github_checks:
-    annotations: false
diff --git a/.github/workflows/_backend-tests.yml b/.github/workflows/_backend-tests.yml
new file mode 100644
index 00000000000..b7f49164916
--- /dev/null
+++ b/.github/workflows/_backend-tests.yml
@@ -0,0 +1,120 @@
+name: Backend tests
+on:
+  workflow_call:
+    inputs:
+      julia-version:
+        required: true
+        type: string
+      backend:
+        required: true
+        type: string
+        description: 'Backend name, e.g. CairoMakie, GLMakie, WGLMakie'
+      nbuckets:
+        required: false
+        type: number
+        default: 2
+      system-packages:
+        required: false
+        type: string
+        default: ''
+        description: 'apt packages to install (space-separated)'
+      test-prefix:
+        required: false
+        type: string
+        default: ''
+        description: 'Command prefix for test execution, e.g. "DISPLAY=:0 xvfb-run -s \"-screen 0 1024x768x24\""'
+
+jobs:
+  setup:
+    name: Setup ${{ inputs.backend }} Julia ${{ inputs.julia-version }}
+    runs-on: ubuntu-latest
+    outputs:
+      bucket-list: ${{ steps.buckets.outputs.list }}
+    steps:
+      - id: buckets
+        run: echo "list=$(seq 1 ${{ inputs.nbuckets }} | jq -cs .)" >> $GITHUB_OUTPUT
+      - uses: actions/checkout@v6
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: ${{ inputs.julia-version }}
+      - uses: julia-actions/cache@v3
+        with:
+          cache-name: julia-cache-${{ inputs.backend }}-${{ inputs.julia-version }}
+          include-matrix: false
+      - name: Install system dependencies
+        if: inputs.system-packages != ''
+        run: sudo apt-get update && sudo apt-get install -y ${{ inputs.system-packages }}
+      - name: Install Julia dependencies
+        shell: julia --project=monorepo {0}
+        env:
+          # Don't precompile here; Pkg.test below handles it with the correct
+          # test environment and flags.
+          JULIA_PKG_PRECOMPILE_AUTO: '0'
+        run: |
+          using Pkg;
+          pkg"registry up"
+          Pkg.update()
+          pkg"dev ./Makie ./${{ inputs.backend }} ./ReferenceTests ./ComputePipeline"
+      - name: Precompile via Pkg.test
+        env:
+          PRECOMPILE_ONLY: 'true'
+          # Portable CPU target so pkgimages work across GitHub's heterogeneous
+          # runner fleet (mix of AMD Zen3, Intel Haswell, etc.).
+          JULIA_CPU_TARGET: 'generic;sandybridge,-xsaveopt,clone_all;haswell,-rdrnd,base(1)'
+        # NOTE: coverage is intentionally omitted. On Julia 1.10, coverage=true
+        # disables pkgimages (use_pkgimages=false), causing cache flag mismatches
+        # between setup and bucket jobs. If coverage is re-enabled, this must be
+        # set identically here and in the bucket test step.
+        run: ${{ inputs.test-prefix }} julia --color=yes --project=monorepo -e 'using Pkg; Pkg.test("${{ inputs.backend }}")'
+      - name: Package depot and Manifest
+        run: >
+          cp monorepo/Manifest.toml ~/.julia/monorepo-Manifest.toml &&
+          tar -cf /tmp/depot.tar.zst --use-compress-program zstdmt -C ~
+          .julia/compiled .julia/packages .julia/artifacts .julia/scratchspaces .julia/monorepo-Manifest.toml
+      - name: Upload depot tarball
+        uses: actions/upload-artifact@v7
+        with:
+          name: julia-depot-${{ inputs.backend }}-${{ inputs.julia-version }}
+          path: /tmp/depot.tar.zst
+          retention-days: 1
+
+  test:
+    name: ${{ inputs.backend }} Julia ${{ inputs.julia-version }} bucket ${{ matrix.bucket }}/${{ inputs.nbuckets }}
+    needs: setup
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        bucket: ${{ fromJSON(needs.setup.outputs.bucket-list) }}
+    steps:
+      - uses: actions/checkout@v6
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: ${{ inputs.julia-version }}
+      - name: Download depot tarball
+        uses: actions/download-artifact@v8
+        with:
+          name: julia-depot-${{ inputs.backend }}-${{ inputs.julia-version }}
+      - name: Extract depot and Manifest
+        run: tar -xf depot.tar.zst --use-compress-program unzstd -C ~ && mkdir -p monorepo && cp ~/.julia/monorepo-Manifest.toml monorepo/Manifest.toml
+      - name: Install system dependencies
+        if: inputs.system-packages != ''
+        run: sudo apt-get update && sudo apt-get install -y ${{ inputs.system-packages }}
+      - name: Run the tests
+        continue-on-error: true
+        env:
+          REFTEST_BUCKET: ${{ matrix.bucket }}
+          REFTEST_NBUCKETS: ${{ inputs.nbuckets }}
+          JULIA_CPU_TARGET: 'generic;sandybridge,-xsaveopt,clone_all;haswell,-rdrnd,base(1)'
+        run: >
+          ${{ inputs.test-prefix }}
+          julia --color=yes --project=monorepo -e 'using Pkg; Pkg.test("${{ inputs.backend }}")'
+          && echo "TESTS_SUCCESSFUL=true" >> $GITHUB_ENV
+      - name: Upload test Artifacts
+        uses: actions/upload-artifact@v7
+        with:
+          name: ReferenceImages_${{ inputs.backend }}_${{ inputs.julia-version }}_${{ matrix.bucket }}
+          path: ./${{ inputs.backend }}/test/reference_images/
+      - name: Fail after artifacts if tests failed
+        if: ${{ env.TESTS_SUCCESSFUL != 'true' }}
+        run: exit 1
diff --git a/.github/workflows/_cairomakie.yml b/.github/workflows/_cairomakie.yml
deleted file mode 100644
index afce54308a2..00000000000
--- a/.github/workflows/_cairomakie.yml
+++ /dev/null
@@ -1,53 +0,0 @@
-name: CairoMakie
-on:
-  workflow_call:
-
-jobs:
-  test:
-    name: CairoMakie Julia ${{ matrix.version }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        version:
-          - '1.10'
-          - '1'
-        os:
-          - ubuntu-latest
-        arch:
-          - x64
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: ${{ matrix.version }}
-          arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v3
-      - name: Install Julia dependencies
-        shell: julia --project=monorepo {0}
-        run: |
-          using Pkg;
-          # dev mono repo versions
-          pkg"registry up"
-          Pkg.update()
-          pkg"dev ./Makie ./CairoMakie ./ReferenceTests ./ComputePipeline"
-      - name: Run the tests
-        continue-on-error: true
-        run: >
-          julia --color=yes --project=monorepo -e 'using Pkg; Pkg.test("CairoMakie", coverage=true)'
-          && echo "TESTS_SUCCESSFUL=true" >> $GITHUB_ENV
-      - name: Upload test Artifacts
-        uses: actions/upload-artifact@v7
-        with:
-          name: ReferenceImages_CairoMakie_${{ matrix.version }}
-          path: ./CairoMakie/test/reference_images/
-      - name: Fail after artifacts if tests failed
-        if: ${{ env.TESTS_SUCCESSFUL != 'true' }}
-        run: exit 1
-      - uses: julia-actions/julia-processcoverage@v1
-        with:
-          directories: Makie/src
-      - uses: codecov/codecov-action@v6
-        with:
-          file: lcov.info
diff --git a/.github/workflows/_compute-pipeline.yml b/.github/workflows/_compute-pipeline.yml
index 93d2830665c..d923db80995 100644
--- a/.github/workflows/_compute-pipeline.yml
+++ b/.github/workflows/_compute-pipeline.yml
@@ -26,6 +26,8 @@ jobs:
       - uses: julia-actions/cache@v3
       - name: Develop and test ComputePipeline
         shell: julia --project=monorepo {0}
+        env:
+          JULIA_PKG_PRECOMPILE_AUTO: '0'
         run: |
           using Pkg
           # dev mono repo versions
diff --git a/.github/workflows/_glmakie.yml b/.github/workflows/_glmakie.yml
deleted file mode 100644
index 6572dba7828..00000000000
--- a/.github/workflows/_glmakie.yml
+++ /dev/null
@@ -1,58 +0,0 @@
-name: GLMakie
-on:
-  workflow_call:
-
-jobs:
-  test:
-    name: GLMakie Julia ${{ matrix.version }}
-    env:
-      MODERNGL_DEBUGGING: "true"
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        version:
-          - '1.10'
-          - '1'
-        os:
-          - ubuntu-latest
-        arch:
-          - x64
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: ${{ matrix.version }}
-          arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v3
-      - run: sudo apt-get update && sudo apt-get install -y xorg-dev mesa-utils xvfb libgl1 freeglut3-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libxext-dev xsettingsd x11-xserver-utils
-      - name: Install Julia dependencies
-        shell: julia --project=monorepo {0}
-        run: |
-          using Pkg;
-          # dev mono repo versions
-          pkg"registry up"
-          Pkg.update()
-          pkg"dev ./Makie ./GLMakie ./ReferenceTests ./ComputePipeline"
-      - name: Run the tests
-        id: referencetests
-        continue-on-error: true
-        run: >
-          DISPLAY=:0 xvfb-run -s '-screen 0 1024x768x24' julia --color=yes --project=monorepo -e 'using Pkg; Pkg.test("GLMakie", coverage=true)'
-          && echo "TESTS_SUCCESSFUL=true" >> $GITHUB_ENV
-      - name: Upload test Artifacts
-        uses: actions/upload-artifact@v7
-        with:
-          name: ReferenceImages_GLMakie_${{ matrix.version }}
-          path: |
-            ./GLMakie/test/reference_images/
-      - name: Fail after artifacts if tests failed
-        if: ${{ env.TESTS_SUCCESSFUL != 'true' }}
-        run: exit 1
-      - uses: julia-actions/julia-processcoverage@v1
-        with:
-          directories: Makie/src
-      - uses: codecov/codecov-action@v6
-        with:
-          file: lcov.info
diff --git a/.github/workflows/_makie.yml b/.github/workflows/_makie.yml
index 1a488243aaa..df9d7a8e20c 100644
--- a/.github/workflows/_makie.yml
+++ b/.github/workflows/_makie.yml
@@ -26,6 +26,8 @@ jobs:
       - uses: julia-actions/cache@v3
       - name: Develop and test Makie
         shell: julia --project=monorepo {0}
+        env:
+          JULIA_PKG_PRECOMPILE_AUTO: '0'
         run: |
           using Pkg
           # dev mono repo versions
diff --git a/.github/workflows/_reference-tests.yml b/.github/workflows/_reference-tests.yml
index 6076b788d0b..80ca90ee547 100644
--- a/.github/workflows/_reference-tests.yml
+++ b/.github/workflows/_reference-tests.yml
@@ -4,13 +4,36 @@ on:
 
 jobs:
   cairomakie:
-    uses: ./.github/workflows/_cairomakie.yml
+    strategy:
+      matrix:
+        julia-version: ['1.10', '1']
+    uses: ./.github/workflows/_backend-tests.yml
+    with:
+      julia-version: ${{ matrix.julia-version }}
+      backend: CairoMakie
 
   glmakie:
-    uses: ./.github/workflows/_glmakie.yml
+    strategy:
+      matrix:
+        julia-version: ['1.10', '1']
+    uses: ./.github/workflows/_backend-tests.yml
+    with:
+      julia-version: ${{ matrix.julia-version }}
+      backend: GLMakie
+      system-packages: xorg-dev mesa-utils xvfb libgl1 freeglut3-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libxext-dev xsettingsd x11-xserver-utils
+      test-prefix: 'DISPLAY=:0 xvfb-run -s "-screen 0 1024x768x24"'
 
   wglmakie:
-    uses: ./.github/workflows/_wglmakie.yml
+    strategy:
+      matrix:
+        julia-version: ['1.10', '1']
+    uses: ./.github/workflows/_backend-tests.yml
+    with:
+      julia-version: ${{ matrix.julia-version }}
+      backend: WGLMakie
+      nbuckets: 4
+      system-packages: xorg-dev libosmesa6 libgl1-mesa-dri mesa-utils xvfb libgl1 freeglut3-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libxext-dev
+      test-prefix: 'DISPLAY=:0 xvfb-run -s "-screen 0 2048x2048x24"'
 
   consolidation:
     name: Merge artifacts
@@ -18,42 +41,61 @@ jobs:
     if: always()
     needs: [cairomakie, glmakie, wglmakie]
     steps:
-      - uses: actions/download-artifact@v8
-        with:
-          name: ReferenceImages_WGLMakie_1
-          path: ./ReferenceImages/WGLMakie
-      - uses: actions/download-artifact@v8
+      - name: Download all bucket artifacts for Julia 1
+        uses: actions/download-artifact@v8
         with:
-          name: ReferenceImages_CairoMakie_1
-          path: ./ReferenceImages/CairoMakie
-      - uses: actions/download-artifact@v8
-        with:
-          name: ReferenceImages_GLMakie_1
-          path: ./ReferenceImages/GLMakie
+          pattern: ReferenceImages_*_1_*
+          path: ./BucketArtifacts/
       - name: Consolidate reference image folders
         run: |
-          baseDir="./ReferenceImages"
+          baseDir="./BucketArtifacts"
+          outDir="./ReferenceImagesCombined"
+          mkdir -p "$outDir"
+
+          # Copy reference images from the first available artifact (same across all buckets/backends)
+          first_artifact=$(ls -d ${baseDir}/ReferenceImages_*_1_*/ 2>/dev/null | head -1)
+          if [ -n "$first_artifact" ] && [ -d "${first_artifact}/reference" ]; then
+            cp -r "${first_artifact}/reference/." "$outDir/reference/"
+          fi
+
+          # Initialize output files
+          > "$outDir/scores.tsv"
+          > "$outDir/new_files.txt"
+          > "$outDir/missing_files.txt"
+
+          for backend in CairoMakie GLMakie WGLMakie; do
+              mkdir -p "$outDir/recorded/${backend}/"
+
+              # Merge all buckets for this backend
+              for bucket_dir in ${baseDir}/ReferenceImages_${backend}_1_*/; do
+                  [ -d "$bucket_dir" ] || continue
+
+                  # Copy recorded images (no overlap between buckets)
+                  if [ -d "${bucket_dir}/recorded/${backend}/" ]; then
+                      cp -r "${bucket_dir}/recorded/${backend}/." "$outDir/recorded/${backend}/"
+                  fi
 
-          # Create new top-level directory for combined files
-          mkdir -p "./ReferenceImagesCombined"
+                  # Concatenate scores and new_files from each bucket
+                  [ -f "${bucket_dir}/scores.tsv" ] && cat "${bucket_dir}/scores.tsv" >> "$outDir/scores.tsv"
+                  [ -f "${bucket_dir}/new_files.txt" ] && cat "${bucket_dir}/new_files.txt" >> "$outDir/new_files.txt"
+              done
 
-          # Copy the reference folder from GLMakie, it's the same for all backends
-          cp -r "${baseDir}/GLMakie/reference/." "./ReferenceImagesCombined/reference/"
+              # Compute missing_files: reference files that were neither recorded nor skipped
+              if [ -d "$outDir/reference/${backend}" ]; then
+                  first_bucket=$(ls -d ${baseDir}/ReferenceImages_${backend}_1_*/ 2>/dev/null | head -1)
 
-          # Initialize empty files for concatenation
-          > "./ReferenceImagesCombined/scores.tsv"
-          > "./ReferenceImagesCombined/new_files.txt"
+                  # Build sorted list of files that are accounted for (recorded + skipped)
+                  {
+                      [ -d "$outDir/recorded/${backend}" ] && (cd "$outDir/recorded" && find "${backend}" -type f)
+                      [ -n "$first_bucket" ] && [ -f "${first_bucket}/skipped_names.txt" ] && cat "${first_bucket}/skipped_names.txt"
+                  } | sort -u > /tmp/have_or_skip.txt
 
-          # Loop through the directories and concatenate the files, and copy recorded folders
-          for dir in WGLMakie CairoMakie GLMakie; do
-              # Concatenate scores.tsv, new_files.txt and missing_files.txt
-              cat "${baseDir}/${dir}/scores.tsv" >> "./ReferenceImagesCombined/scores.tsv"
-              cat "${baseDir}/${dir}/new_files.txt" >> "./ReferenceImagesCombined/new_files.txt"
-              cat "${baseDir}/${dir}/missing_files.txt" >> "./ReferenceImagesCombined/missing_files.txt"
+                  # Reference files
+                  (cd "$outDir/reference" && find "${backend}" -type f) | sort > /tmp/ref_files.txt
 
-              # Copy recorded folder
-              mkdir -p "./ReferenceImagesCombined/recorded/${dir}/"
-              cp -r "${baseDir}/${dir}/recorded/${dir}/." "./ReferenceImagesCombined/recorded/${dir}/"
+                  # Missing = reference - (recorded ∪ skipped)
+                  comm -23 /tmp/ref_files.txt /tmp/have_or_skip.txt >> "$outDir/missing_files.txt"
+              fi
           done
 
           echo "Files and folders have been successfully combined into ReferenceImagesCombined."
diff --git a/.github/workflows/_rprmakie.yml b/.github/workflows/_rprmakie.yml
index 2fd252a6720..7370eac3905 100644
--- a/.github/workflows/_rprmakie.yml
+++ b/.github/workflows/_rprmakie.yml
@@ -25,6 +25,8 @@ jobs:
       - uses: julia-actions/cache@v3
       - name: Install Julia dependencies
         shell: julia --project=monorepo {0}
+        env:
+          JULIA_PKG_PRECOMPILE_AUTO: '0'
         run: |
           using Pkg;
           pkg"registry up"
diff --git a/.github/workflows/_wglmakie.yml b/.github/workflows/_wglmakie.yml
deleted file mode 100644
index bc037485a87..00000000000
--- a/.github/workflows/_wglmakie.yml
+++ /dev/null
@@ -1,59 +0,0 @@
-name: WGLMakie
-on:
-  workflow_call:
-
-jobs:
-  test:
-    name: WGLMakie Julia ${{ matrix.version }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        version:
-          - '1.10'
-          - '1'
-        os:
-          - ubuntu-latest
-        arch:
-          - x64
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: ${{ matrix.version }}
-          arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v3
-      - run: sudo apt-get update && sudo apt-get install -y xorg-dev libosmesa6 libgl1-mesa-dri mesa-utils xvfb libgl1 freeglut3-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libxext-dev
-      - name: Install Julia dependencies
-        shell: julia --project=monorepo {0}
-        run: |
-          using Pkg;
-          # dev mono repo versions
-          pkg"registry up"
-          Pkg.update()
-          pkg"dev ./Makie ./WGLMakie ./ReferenceTests ./ComputePipeline"
-      - name: Run the tests
-        continue-on-error: true
-        run: >
-          DISPLAY=:0 xvfb-run -s '-screen 0 2048x2048x24' julia --color=yes --project=monorepo -e 'using Pkg; Pkg.test("WGLMakie", coverage=true)'
-          && echo "TESTS_SUCCESSFUL=true" >> $GITHUB_ENV
-      - name: Upload test Artifacts
-        uses: actions/upload-artifact@v7
-        with:
-          name: ReferenceImages_WGLMakie_${{ matrix.version }}
-          path: ./WGLMakie/test/reference_images/
-      - name: Upload test Electron logs
-        uses: actions/upload-artifact@v7
-        with:
-          name: Electron_Logs_WGLMakie_${{ matrix.version }}
-          path: ./WGLMakie/test/electron.log
-      - name: Fail after artifacts if tests failed
-        if: ${{ env.TESTS_SUCCESSFUL != 'true' }}
-        run: exit 1
-      - uses: julia-actions/julia-processcoverage@v1
-        with:
-          directories: Makie/src
-      - uses: codecov/codecov-action@v6
-        with:
-          file: lcov.info
diff --git a/.github/workflows/compilation-benchmark.yaml b/.github/workflows/compilation-benchmark.yaml
index aa0dde4e33c..86b2375cb9b 100644
--- a/.github/workflows/compilation-benchmark.yaml
+++ b/.github/workflows/compilation-benchmark.yaml
@@ -1,19 +1,21 @@
 name: Benchmark
 on:
-  pull_request:
-    paths-ignore:
-      - 'docs/**'
-      - '*.md'
-    branches:
-      - master
+  issue_comment:
+    types: [created]
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  group: ${{ github.workflow }}-${{ github.event.issue.number }}
   cancel-in-progress: true
 
 jobs:
   benchmark:
     name: ${{ matrix.package }}
+    if: >
+      github.event.issue.pull_request &&
+      contains(github.event.comment.body, '/benchmark') &&
+      (github.event.comment.author_association == 'MEMBER' ||
+       github.event.comment.author_association == 'OWNER' ||
+       github.event.comment.author_association == 'COLLABORATOR')
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
@@ -23,8 +25,19 @@ jobs:
           - GLMakie
           - WGLMakie
     steps:
+      - name: Get PR head ref
+        id: pr
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          pr_data=$(gh api repos/${{ github.repository }}/pulls/${{ github.event.issue.number }})
+          echo "ref=$(echo $pr_data | jq -r .head.ref)" >> $GITHUB_OUTPUT
+          echo "sha=$(echo $pr_data | jq -r .head.sha)" >> $GITHUB_OUTPUT
+          echo "base_ref=$(echo $pr_data | jq -r .base.ref)" >> $GITHUB_OUTPUT
       - name: Checkout
         uses: actions/checkout@v6
+        with:
+          ref: ${{ steps.pr.outputs.ref }}
       - name: Install xvfb
         run: sudo apt-get update && sudo apt-get install -y xorg-dev mesa-utils xvfb libgl1 freeglut3-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libxext-dev
       - uses: julia-actions/setup-julia@v2
@@ -35,9 +48,9 @@ jobs:
       - name: Benchmark
         env:
           GITHUB_TOKEN: ${{ secrets.BENCHMARK_KEY }}
-          PR_NUMBER: ${{ github.event.number }}
+          PR_NUMBER: ${{ github.event.issue.number }}
         run: >
-          DISPLAY=:0 xvfb-run -s '-screen 0 1024x768x24' julia --project=./metrics/ttfp/ ./metrics/ttfp/run-benchmark.jl ${{ matrix.package }} 20 ${{ github.event.pull_request.base.ref }}
+          DISPLAY=:0 xvfb-run -s '-screen 0 1024x768x24' julia --project=./metrics/ttfp/ ./metrics/ttfp/run-benchmark.jl ${{ matrix.package }} 60 ${{ steps.pr.outputs.base_ref }}
       - name: Upload plots as artifact
         uses: actions/upload-artifact@v7
         with:
@@ -45,11 +58,19 @@ jobs:
           path: ./benchmark_results
   post-gist:
     name: Post Benchmark Gist
-    needs: benchmark  # Wait for all benchmark jobs to complete
+    needs: benchmark
     runs-on: ubuntu-latest
     permissions:
-      statuses: write       # Permission to post workflow status
+      statuses: write
     steps:
+      - name: Get PR head SHA
+        id: pr
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          sha=$(gh api repos/${{ github.repository }}/pulls/${{ github.event.issue.number }} --jq .head.sha)
+          echo "sha=$sha" >> $GITHUB_OUTPUT
+
       - name: Check token validity
         id: check_token
         env:
@@ -57,7 +78,6 @@ jobs:
         run: |
           if [ -z "$BENCHMARK_KEY" ]; then
             echo "Benchmark key missing — skipping benchmark upload."
-            echo "NOTE: Benchmark key is not accessible for outside contributors"
             echo "has_token=false" >> $GITHUB_OUTPUT
           else
             echo "has_token=true" >> $GITHUB_OUTPUT
@@ -93,7 +113,7 @@ jobs:
             --method POST \
             -H "Accept: application/vnd.github+json" \
             -H "X-GitHub-Api-Version: 2022-11-28" \
-            /repos/${{ github.repository }}/statuses/${{ github.event.pull_request.head.sha }} \
+            /repos/${{ github.repository }}/statuses/${{ steps.pr.outputs.sha }} \
             -f "state=success" \
             -f "context=Benchmark Results" \
             -f "description=Plots are available under Details" \
@@ -104,11 +124,12 @@ jobs:
         uses: thollander/actions-comment-pull-request@v3
         with:
           github-token: ${{ secrets.BENCHMARK_KEY }}
-          comment-tag: benchmark # this allows to update the same post with new data
+          pr-number: ${{ github.event.issue.number }}
+          comment-tag: benchmark
           message: |
             # Benchmark Results
 
-            SHA: [${{ github.event.pull_request.head.sha }}](https://github.com/${{ github.repository }}/commit/${{ github.event.pull_request.head.sha }})
+            SHA: [${{ steps.pr.outputs.sha }}](https://github.com/${{ github.repository }}/commit/${{ steps.pr.outputs.sha }})
 
             > [!WARNING]
             > These results are subject to substantial noise because GitHub's CI runs on shared machines that are not ideally suited for benchmarking.
@@ -116,4 +137,3 @@ jobs:
             ![GLMakie](${{ env.GIST_URL_USERCONTENT }}/raw/GLMakie.svg)
             ![CairoMakie](${{ env.GIST_URL_USERCONTENT }}/raw/CairoMakie.svg)
             ![WGLMakie](${{ env.GIST_URL_USERCONTENT }}/raw/WGLMakie.svg)
-
diff --git a/CairoMakie/test/runtests.jl b/CairoMakie/test/runtests.jl
index 7d2d2e7d2e4..bf38203cb11 100644
--- a/CairoMakie/test/runtests.jl
+++ b/CairoMakie/test/runtests.jl
@@ -1,5 +1,8 @@
 ENV["ENABLE_COMPUTE_CHECKS"] = "true"
 
+# Allow Pkg.test to be used purely for precompilation (setup job in CI)
+get(ENV, "PRECOMPILE_ONLY", nothing) == "true" && exit(0)
+
 using Test
 using CairoMakie
 using Makie.FileIO
diff --git a/GLMakie/test/runtests.jl b/GLMakie/test/runtests.jl
index 47dfbc3de7b..f47d1edc424 100644
--- a/GLMakie/test/runtests.jl
+++ b/GLMakie/test/runtests.jl
@@ -1,5 +1,8 @@
 ENV["ENABLE_COMPUTE_CHECKS"] = "true"
 
+# Allow Pkg.test to be used purely for precompilation (setup job in CI)
+get(ENV, "PRECOMPILE_ONLY", nothing) == "true" && exit(0)
+
 using Makie
 using GLMakie, Test
 using FileIO
diff --git a/ReferenceTests/src/database.jl b/ReferenceTests/src/database.jl
index 73372ca5baa..45ebd279492 100644
--- a/ReferenceTests/src/database.jl
+++ b/ReferenceTests/src/database.jl
@@ -21,8 +21,59 @@ const RECORDING_DIR = Base.RefValue{String}()
 const SKIP_TITLES = Set{String}()
 const SKIP_FUNCTIONS = Set{Symbol}()
 const COUNTER = Ref(0)
+const TOTAL_COUNTER = Ref(0) # counts all @reference_test invocations, used for bucket assignment
+const TOTAL_TESTS = Ref(0) # total number of tests across all files, set before running for chunked bucketing
 const SKIPPED_NAMES = Set{String}() # names skipped due to title exclusion or function exclusion
 
+"""
+    count_reference_tests(paths)
+
+Count `@reference_test` occurrences by reading files as text, recursively
+following `include("...")` directives relative to each file's directory.
+"""
+function count_reference_tests(paths)
+    n = 0
+    for path in paths
+        n += _count_in_file(path)
+    end
+    return n
+end
+
+function _count_in_file(path)
+    text = read(path, String)
+    n = count(r"@reference_test\b", text)
+    dir = dirname(path)
+    for m in eachmatch(r"include\(\"([^\"]+)\"\)", text)
+        child = joinpath(dir, m.captures[1])
+        isfile(child) && (n += _count_in_file(child))
+    end
+    return n
+end
+
+"""
+    should_run_in_bucket(test_index::Int)
+
+Check whether a test with the given index should run in the current bucket.
+Bucketing is controlled by the `REFTEST_BUCKET` and `REFTEST_NBUCKETS` environment variables.
+If either is unset, all tests run (no bucketing).
+"""
+function should_run_in_bucket(test_index::Int)
+    nbuckets_str = get(ENV, "REFTEST_NBUCKETS", nothing)
+    bucket_str = get(ENV, "REFTEST_BUCKET", nothing)
+    (nbuckets_str === nothing || bucket_str === nothing) && return true
+    nbuckets = parse(Int, nbuckets_str)
+    bucket = parse(Int, bucket_str)
+    # Use contiguous chunks rather than interleaving so that each bucket
+    # touches fewer distinct code paths, reducing JIT/compilation overhead.
+    ntotal = TOTAL_TESTS[]
+    if ntotal <= 0
+        # Total unknown, fall back to interleaved assignment
+        return mod1(test_index, nbuckets) == bucket
+    end
+    chunk_size = cld(ntotal, nbuckets) # ceil division
+    return (bucket - 1) * chunk_size < test_index <= min(bucket * chunk_size, ntotal)
+end
+
 """
     @reference_test(name, code)
 
@@ -34,35 +85,40 @@ macro reference_test(name, code)
     funcs = used_functions(code)
     skip = (title in SKIP_TITLES) || any(x -> x in funcs, SKIP_FUNCTIONS)
     return quote
-        @testset $(title) begin
-            if $skip
-                @test_broken false
-                mark_skipped!($title)
-            else
-                t1 = time()
-                if $title in $REGISTERED_TESTS
-                    error("title must be unique. Duplicate title: $($title)")
-                end
-                println("running $(lpad(COUNTER[] += 1, 3)): $($title)")
-                Makie.set_theme!(;
-                    size = (500, 500),
-                    CairoMakie = (; px_per_unit = 1),
-                    GLMakie = (; scalefactor = 1, px_per_unit = 1),
-                    WGLMakie = (; scalefactor = 1, px_per_unit = 1)
-                )
-                ReferenceTests.RNG.seed_rng!()
-                result = let
-                    $(esc(code))
+        ReferenceTests.TOTAL_COUNTER[] += 1
+        if !ReferenceTests.should_run_in_bucket(ReferenceTests.TOTAL_COUNTER[])
+            # Not in this bucket, skip entirely
+        else
+            @testset $(title) begin
+                if $skip
+                    @test_broken false
+                    mark_skipped!($title)
+                else
+                    t1 = time()
+                    if $title in $REGISTERED_TESTS
+                        error("title must be unique. Duplicate title: $($title)")
+                    end
+                    println("running $(lpad(COUNTER[] += 1, 3)): $($title)")
+                    Makie.set_theme!(;
+                        size = (500, 500),
+                        CairoMakie = (; px_per_unit = 1),
+                        GLMakie = (; scalefactor = 1, px_per_unit = 1),
+                        WGLMakie = (; scalefactor = 1, px_per_unit = 1)
+                    )
+                    ReferenceTests.RNG.seed_rng!()
+                    result = let
+                        $(esc(code))
+                    end
+                    @test save_result(joinpath(RECORDING_DIR[], $title), result)
+                    push!($REGISTERED_TESTS, $title)
+                    elapsed = round(time() - t1; digits = 5)
+                    total = Sys.total_memory()
+                    mem = round((total - Sys.free_memory()) / 10^9; digits = 3)
+                    # TODO, write to file and create an overview in the end, similar to the benchmark results!
+                    println("Used $(mem)gb of $(round(total / 10^9; digits = 3))gb RAM, time: $(elapsed)s")
                 end
-                @test save_result(joinpath(RECORDING_DIR[], $title), result)
-                push!($REGISTERED_TESTS, $title)
-                elapsed = round(time() - t1; digits = 5)
-                total = Sys.total_memory()
-                mem = round((total - Sys.free_memory()) / 10^9; digits = 3)
-                # TODO, write to file and create an overview in the end, similar to the benchmark results!
-                println("Used $(mem)gb of $(round(total / 10^9; digits = 3))gb RAM, time: $(elapsed)s")
+                GC.gc(true) # Run GC, to catch accumulating memory early on (used RAM)
             end
-            GC.gc(true) # Run GC, to catch accumulating memory early on (used RAM)
         end
     end
 end
@@ -115,6 +171,20 @@ macro include_reference_tests(backend::Symbol, path, paths...)
             # prefix the recordings with the backend name so that each backend has its own versions
             ReferenceTests.RECORDING_DIR[] = joinpath(recording_dir, "recorded", $(string(backend)))
             mkpath(ReferenceTests.RECORDING_DIR[])
+
+            # Count total @reference_test occurrences for chunked bucket assignment
+            ReferenceTests.TOTAL_TESTS[] = ReferenceTests.count_reference_tests(include_paths)
+
+            # Log bucket configuration
+            let
+                nb = get(ENV, "REFTEST_NBUCKETS", nothing)
+                b = get(ENV, "REFTEST_BUCKET", nothing)
+                if nb !== nothing && b !== nothing
+                    ntotal = ReferenceTests.TOTAL_TESTS[]
+                    @info "Running reference test bucket $(b)/$(nb) ($(ntotal) total tests)"
+                end
+            end
+
             @testset "Reference tests $($(string(backend)))" begin
                 empty!(ReferenceTests.REGISTERED_TESTS)
                 for include_path in include_paths
@@ -125,6 +195,7 @@ macro include_reference_tests(backend::Symbol, path, paths...)
             recording_dir = recording_dir
             empty!(ReferenceTests.REGISTERED_TESTS)
             ReferenceTests.COUNTER[] = 0
+            ReferenceTests.TOTAL_COUNTER[] = 0
             (recorded_files, recording_dir)
         end
     )
diff --git a/ReferenceTests/src/runtests.jl b/ReferenceTests/src/runtests.jl
index e00effcdb96..16a88636c15 100644
--- a/ReferenceTests/src/runtests.jl
+++ b/ReferenceTests/src/runtests.jl
@@ -37,6 +37,12 @@ function record_comparison(base_folder::String, backend::String; record_folder_n
         end
     end
 
+    open(joinpath(base_folder, "skipped_names.txt"), "w") do file
+        for name in sort!(collect(SKIPPED_NAMES))
+            println(file, joinpath(backend, "$name.png"))
+        end
+    end
+
     return missing_refimages, scores
 end
 
diff --git a/ReferenceTests/src/tests/updating.jl b/ReferenceTests/src/tests/updating.jl
index a8089c411bf..17ff041c6e2 100644
--- a/ReferenceTests/src/tests/updating.jl
+++ b/ReferenceTests/src/tests/updating.jl
@@ -108,7 +108,11 @@ end
 @reference_test "deletion" begin
     f = Figure()
     l = Legend(f[1, 1], [LineElement(color = :red)], ["Line"])
-    s = display(f)
+    # Use colorbuffer to initialize the screen with a properly established session.
+    # The generic `display(f)` inline path doesn't reliably wait for the WGLMakie session.
+    Makie.colorbuffer(f)
+    s = Makie.getscreen(f.scene)
+    @test !isnothing(s)
     @test f.scene.current_screens[1] === s
     @test f.scene.children[1].current_screens[1] === s
     @test f.scene.children[1].children[1].current_screens[1] === s
@@ -123,7 +127,9 @@ end
 @reference_test "deletion and observable args" begin
     obs = Observable(1:5)
     f, ax, pl = scatter(obs; markersize = 150)
-    s = display(f)
+    # Use colorbuffer to initialize the screen with a properly established session.
+    # The generic `display(f)` inline path doesn't reliably wait for the WGLMakie session.
+    Makie.colorbuffer(f)
     # So, for GLMakie it will be 2, since we register an additional listener for
     # State changes for the on demand renderloop
     @test length(obs.listeners) in (1, 2)
diff --git a/WGLMakie/test/runtests.jl b/WGLMakie/test/runtests.jl
index 5f468350376..d24b3e94cc4 100644
--- a/WGLMakie/test/runtests.jl
+++ b/WGLMakie/test/runtests.jl
@@ -1,4 +1,8 @@
 ENV["ENABLE_COMPUTE_CHECKS"] = "true"
+
+# Allow Pkg.test to be used purely for precompilation (setup job in CI)
+get(ENV, "PRECOMPILE_ONLY", nothing) == "true" && exit(0)
+
 ENV["ELECTRON_LOG_FILE"] = joinpath(@__DIR__, "electron.log")
 ENV["ELECTRON_ENABLE_LOGGING"] = "true"