diff --git a/.github/workflows/Tests.yml b/.github/workflows/Tests.yml
index 25a0b6c82..1b143b01f 100644
--- a/.github/workflows/Tests.yml
+++ b/.github/workflows/Tests.yml
@@ -9,33 +9,76 @@ on:
     paths-ignore:
       - 'docs/**'
   pull_request:
+    types: [opened, synchronize, reopened, ready_for_review, converted_to_draft]
   workflow_dispatch:
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   # Cancel intermediate builds: only if it is a pull request build.
   cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
-
 jobs:
-  tests:
-    name: "Tests"
+  setup-matrix:
+    runs-on: ubuntu-latest
+    outputs:
+      groups: ${{ steps.mk.outputs.groups }}
+      version: ${{ steps.mk.outputs.version }}
+      os: ${{ steps.mk.outputs.os }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: mk
+        shell: bash
+        run: |
+          # Auto-discover test groups from all subdirectory names.
+          groups=$(find test -mindepth 1 -maxdepth 1 -type d \
+            | xargs -I{} basename {} | sort \
+            | jq -R -s -c '[split("\n")[] | select(length > 0) | select(.!= "testsuite")]')
+
+          echo "groups=${groups}" >> "$GITHUB_OUTPUT"
+
+          # Draft PR: only run ubuntu-latest + version=1
+          if [[ "${{ github.event_name }}" == "pull_request" && "${{ github.event.pull_request.draft }}" == "true" ]]; then
+            echo 'version=["1"]' >> "$GITHUB_OUTPUT"
+            echo 'os=["ubuntu-latest"]' >> "$GITHUB_OUTPUT"
+          else
+            echo 'version=["min","1"]' >> "$GITHUB_OUTPUT"
+            echo 'os=["ubuntu-latest","macOS-latest","windows-latest"]' >> "$GITHUB_OUTPUT"
+          fi
+
+  test:
+    name: "Tests (${{ matrix.group }}, ${{ matrix.os }}, Julia ${{ matrix.version }})"
+    needs: setup-matrix
     strategy:
       fail-fast: false
       matrix:
-        version:
-          - 'lts' # minimal supported version
-          - '1' # latest released Julia version
-        # group:
-        os:
-          - ubuntu-latest
-          - macOS-latest
-          - windows-latest
-    uses: "QuantumKitHub/QuantumKitHubActions/.github/workflows/Tests.yml@main"
-    with:
-      group: "${{ matrix.group }}"
-      nthreads: 4
-      julia-version: "${{ matrix.version }}"
-      os: "${{ matrix.os }}"
-      timeout-minutes: 120
-    secrets:
-      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+        version: ${{ fromJSON(needs.setup-matrix.outputs.version) }}
+        os: ${{ fromJSON(needs.setup-matrix.outputs.os) }}
+        group: ${{ fromJSON(needs.setup-matrix.outputs.groups) }}
+        exclude:
+          - os: macos-latest
+            version: 1
+            group: enzyme 
+          - os: macos-latest
+            version: "min" 
+            group: enzyme 
+    runs-on: ${{ matrix.os }}
+    timeout-minutes: 120
+    steps:
+      - uses: actions/checkout@v6
+      - uses: julia-actions/setup-julia@v3
+        with:
+          version: ${{ matrix.version }}
+      - uses: julia-actions/cache@v3
+      - uses: julia-actions/julia-buildpkg@v1
+      - uses: julia-actions/julia-runtest@v1
+        with:
+          test_args: '${{ matrix.group }}${{ github.event.pull_request.draft == true && '' --fast'' || '''' }}'
+        env:
+          JULIA_NUM_THREADS: "4"
+      - uses: julia-actions/julia-processcoverage@v1
+        with:
+          directories: 'src,ext'
+      - uses: codecov/codecov-action@v6
+        with:
+          files: lcov.info
+          token: ${{ secrets.CODECOV_TOKEN }}
+          fail_ci_if_error: false
diff --git a/test/chainrules.jl b/test/chainrules/chainrules.jl
similarity index 88%
rename from test/chainrules.jl
rename to test/chainrules/chainrules.jl
index c0ab618ae..f3b123688 100644
--- a/test/chainrules.jl
+++ b/test/chainrules/chainrules.jl
@@ -4,7 +4,7 @@ using Test
 #BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 BLASFloats = (Float32, ComplexF64) # full suite is too expensive on CI
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/algorithms.jl b/test/common/algorithms.jl
similarity index 100%
rename from test/algorithms.jl
rename to test/common/algorithms.jl
diff --git a/test/codequality.jl b/test/common/codequality.jl
similarity index 100%
rename from test/codequality.jl
rename to test/common/codequality.jl
diff --git a/test/projections.jl b/test/common/projections.jl
similarity index 94%
rename from test/projections.jl
rename to test/common/projections.jl
index 16044f639..74f4a6429 100644
--- a/test/projections.jl
+++ b/test/common/projections.jl
@@ -8,7 +8,7 @@ using CUDA, AMDGPU
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/truncate.jl b/test/common/truncate.jl
similarity index 100%
rename from test/truncate.jl
rename to test/common/truncate.jl
diff --git a/test/eig.jl b/test/decompositions/eig.jl
similarity index 96%
rename from test/eig.jl
rename to test/decompositions/eig.jl
index 12aecfbc4..6c6636e6b 100644
--- a/test/eig.jl
+++ b/test/decompositions/eig.jl
@@ -9,7 +9,7 @@ using CUDA, AMDGPU
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/eigh.jl b/test/decompositions/eigh.jl
similarity index 97%
rename from test/eigh.jl
rename to test/decompositions/eigh.jl
index d9016eca9..0181e481f 100644
--- a/test/eigh.jl
+++ b/test/decompositions/eigh.jl
@@ -8,7 +8,7 @@ using CUDA, AMDGPU
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/gen_eig.jl b/test/decompositions/gen_eig.jl
similarity index 100%
rename from test/gen_eig.jl
rename to test/decompositions/gen_eig.jl
diff --git a/test/lq.jl b/test/decompositions/lq.jl
similarity index 98%
rename from test/lq.jl
rename to test/decompositions/lq.jl
index c4abe15f7..ce38f42f9 100644
--- a/test/lq.jl
+++ b/test/decompositions/lq.jl
@@ -9,7 +9,7 @@ using CUDA, AMDGPU, GenericLinearAlgebra
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (Float16, BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/orthnull.jl b/test/decompositions/orthnull.jl
similarity index 94%
rename from test/orthnull.jl
rename to test/decompositions/orthnull.jl
index a8aedcb9a..6d37dba15 100644
--- a/test/orthnull.jl
+++ b/test/decompositions/orthnull.jl
@@ -8,7 +8,7 @@ using CUDA, AMDGPU
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/polar.jl b/test/decompositions/polar.jl
similarity index 96%
rename from test/polar.jl
rename to test/decompositions/polar.jl
index 6fa0e8d97..a259386e3 100644
--- a/test/polar.jl
+++ b/test/decompositions/polar.jl
@@ -7,7 +7,7 @@ using CUDA, AMDGPU
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/qr.jl b/test/decompositions/qr.jl
similarity index 97%
rename from test/qr.jl
rename to test/decompositions/qr.jl
index b25b4aa74..3a7c2f861 100644
--- a/test/qr.jl
+++ b/test/decompositions/qr.jl
@@ -8,7 +8,7 @@ using CUDA, AMDGPU, GenericLinearAlgebra
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (Float16, BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/schur.jl b/test/decompositions/schur.jl
similarity index 95%
rename from test/schur.jl
rename to test/decompositions/schur.jl
index cbc02d343..ec879314b 100644
--- a/test/schur.jl
+++ b/test/decompositions/schur.jl
@@ -7,7 +7,7 @@ using LinearAlgebra: I, Diagonal
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
diff --git a/test/svd.jl b/test/decompositions/svd.jl
similarity index 97%
rename from test/svd.jl
rename to test/decompositions/svd.jl
index 662346a78..8cf34d834 100644
--- a/test/svd.jl
+++ b/test/decompositions/svd.jl
@@ -5,7 +5,7 @@ using CUDA, AMDGPU
 BLASFloats = (Float32, Float64, ComplexF32, ComplexF64)
 GenericFloats = (BigFloat, Complex{BigFloat})
 
-@isdefined(TestSuite) || include("testsuite/TestSuite.jl")
+@isdefined(TestSuite) || include("../testsuite/TestSuite.jl")
 using .TestSuite
 
 is_buildkite = get(ENV, "BUILDKITE", "false") == "true"