diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index d1acec8ff..92b406b3d 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -23,7 +23,7 @@ env:
 
 concurrency:
   # For a given workflow, if we push to the same branch, cancel all previous builds on that branch except on main.
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: "${{ github.workflow }}-${{ github.ref }}"
   cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
 
 permissions:
@@ -57,8 +57,9 @@ jobs:
         uses: actions/setup-python@82c7e631bb3cdc910f68e0081d67478d79c6982d # v5.1.0
         with:
           python-version: ${{ matrix.python-version }}
-      - name: Install pylint and tox
-        run: pip install pylint tox~=4.0
+      - name: Install pip, pylint, and tox
+        run: |
+          python -m pip install flit pip~=24.0 pylint tox~=4.0
       - name: Run pylint
         run: |
           python -m pylint --rcfile=.pylintrc.toml --disable=import-error --exit-zero xclim
@@ -97,7 +98,7 @@ jobs:
           python-version: ${{ matrix.python-version }}
       - name: Install tox
         run: |
-          python -m pip install tox~=4.0
+          python -m pip install flit pip~=24.0 tox~=4.0
       - name: Test with tox
         run: |
           python -m tox -e ${{ matrix.tox-env }}
@@ -118,17 +119,17 @@ jobs:
       matrix:
         include:
           # Windows builds
-          - tox-env: py39-prefetch-coverage
+          - tox-env: py39-coverage-prefetch
             python-version: "3.9"
             markers: -m 'not slow'
             os: windows-latest
           # macOS builds
-          - tox-env: py310-coverage
+          - tox-env: py310-coverage-lmoments-doctest
             python-version: "3.10"
             markers: -m 'not slow'
             os: macos-latest
           # Linux builds
-          - tox-env: py39-coverage-sbck
+          - tox-env: py39-coverage-sbck-doctest
             python-version: "3.9"
             markers: -m 'not slow'
             os: ubuntu-latest
@@ -139,11 +140,11 @@ jobs:
             python-version: "3.11"
             markers: -m 'not slow'
             os: ubuntu-latest
-          - tox-env: py312-coverage-numba
+          - tox-env: py312-coverage-lmoments-doctest
             python-version: "3.12"
             markers: -m 'not slow'
             os: ubuntu-latest
-          - tox-env: notebooks_doctests
+          - tox-env: notebooks
             python-version: "3.10"
             os: ubuntu-latest
           - tox-env: offline-prefetch
@@ -180,7 +181,7 @@ jobs:
           python-version: ${{ matrix.python-version }}
       - name: Install tox
         run: |
-          python -m pip install tox~=4.0
+          python -m pip install flit pip~=24.0 tox~=4.0 tox-gh
       - name: Test with tox
         run: |
           python -m tox -e ${{ matrix.tox-env }} -- ${{ matrix.markers }}
@@ -192,17 +193,16 @@ jobs:
 
   test-conda:
     needs: lint
-    name: test-conda-Python${{ matrix.python-version }}
+    name: Python${{ matrix.python-version }} (Conda, ${{ matrix.os }})
     if: |
       contains(github.event.pull_request.labels.*.name, 'approved') ||
       (github.event.review.state == 'approved') ||
       (github.event_name == 'push')
-    runs-on: ubuntu-latest
+    runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        include:
-          - python-version: "3.9"
-          - python-version: "3.12"
+        os: [ubuntu-latest]
+        python-version: ["3.9", "3.12"]
     defaults:
       run:
         shell: bash -l {0}
diff --git a/CHANGES.rst b/CHANGES.rst
index 05989cbbe..65ac93e40 100644
--- a/CHANGES.rst
+++ b/CHANGES.rst
@@ -19,12 +19,15 @@ Bug fixes
 * Fixed an bug in sdba's ``map_groups`` that prevented passing DataArrays with cftime coordinates if the ``sdba_encode_cf`` option was True. (:issue:`1673`, :pull:`1674`).
 * Fixed bug (:issue:`1678`, :pull:`1679`) in sdba where a loaded training dataset could not be used for adjustment
 * Fixed bug with loess smoothing for an array full of NaNs. (:pull:`1699`).
-* Fixed and adapted ``time_bnds`` to the newest xarray. (:pull:`1700`)
+* Fixed and adapted ``time_bnds`` to the newest xarray. (:pull:`1700`).
+* Fixed "agreement fraction" in ``robustness_fractions`` to distinguish between negative change and no change. Added "negative" and "changed negative" fractions (:issue:`1690`, :pull:`1711`).
 
 Internal changes
 ^^^^^^^^^^^^^^^^
 * Added "doymin" and "doymax" to the possible operations of ``generic.stats``. Fixed a warning issue when ``op`` was "integral". (:pull:`1672`).
-
+* Reorganized GitHub CI build matrices to run the doctests more consistently. (:pull:`1709`).
+* Removed the experimental `numba` and `llvm` dependency installation steps in the `tox.ini` file. Added `numba@main` to the upstream dependencies. (:pull:`1709`).
+* Added the `tox-gh` dependency to the development installation recipe. This will soon be required for running the `tox` test ensemble on GitHub Workflows. (:pull:`1709`).
 
 v0.48.2 (2024-02-26)
 --------------------
diff --git a/pyproject.toml b/pyproject.toml
index 20ea95e52..9f8ec848d 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -83,6 +83,7 @@ dev = [
   "tokenize-rt",
   "tox >=4.0",
   # "tox-conda",  # Will be added when a tox@v4.0+ compatible plugin is released.
+  "tox-gh >=1.3.1",
   "xdoctest",
   "yamllint",
   # Documentation and examples
@@ -123,7 +124,7 @@ target-version = [
 ]
 
 [tool.bumpversion]
-current_version = "0.48.3-dev.5"
+current_version = "0.48.3-dev.6"
 commit = true
 commit_args = "--no-verify"
 tag = false
diff --git a/requirements_upstream.txt b/requirements_upstream.txt
index 4758fded4..71145c0ce 100644
--- a/requirements_upstream.txt
+++ b/requirements_upstream.txt
@@ -1,4 +1,5 @@
 bottleneck @ git+https://github.com/pydata/bottleneck.git@master
 cftime @ git+https://github.com/Unidata/cftime.git@master
 flox @ git+https://github.com/xarray-contrib/flox.git@main
+numba @ git+https://github.com/numba/numba.git@main
 xarray @ git+https://github.com/pydata/xarray.git@main
diff --git a/tests/test_ensembles.py b/tests/test_ensembles.py
index 67aac38aa..63d62e6d4 100644
--- a/tests/test_ensembles.py
+++ b/tests/test_ensembles.py
@@ -735,13 +735,14 @@ def test_robustness_fractions_weighted(robust_data):
 
 
 def test_robustness_fractions_delta(robust_data):
-    delta = xr.DataArray([-2, 1, -2, -1], dims=("realization",))
+    delta = xr.DataArray([-2, 1, -2, -1, 0, 0], dims=("realization",))
     fracs = ensembles.robustness_fractions(delta, test="threshold", abs_thresh=1.5)
-    np.testing.assert_array_equal(fracs.changed, [0.5])
+    np.testing.assert_array_equal(fracs.changed, [2 / 6])
     np.testing.assert_array_equal(fracs.changed_positive, [0.0])
-    np.testing.assert_array_equal(fracs.positive, [0.25])
-    np.testing.assert_array_equal(fracs.agree, [0.75])
+    np.testing.assert_array_equal(fracs.positive, [1 / 6])
+    np.testing.assert_array_equal(fracs.agree, [3 / 6])
 
+    delta = xr.DataArray([-2, 1, -2, -1], dims=("realization",))
     weights = xr.DataArray([4, 3, 2, 1], dims=("realization",))
     fracs = ensembles.robustness_fractions(
         delta, test="threshold", abs_thresh=1.5, weights=weights
diff --git a/tox.ini b/tox.ini
index 264aee431..afbc68cd5 100644
--- a/tox.ini
+++ b/tox.ini
@@ -3,18 +3,26 @@ min_version = 4.0
 env_list =
     lint
     docs
-    notebooks_doctests
+    notebooks
     offline-prefetch
     py39-upstream-doctest
-    py310
+    py310-doctest
     py311-lmoments
-    py312-numba
+    py312-lmoments-doctest
 labels =
     test = py39, py310-upstream-doctest, py311, notebooks_doctests, offline-prefetch
 requires =
-    pip >= 23.0
+    pip >= 24.0
+    flit
 opts = -vv
 
+[gh]
+python =
+    3.12 = py312-coverage-lmoments-doctest
+    3.11 = py311-coverage-lmoments-sbck-doctest, offline-coverage-prefetch
+    3.10 = py310-coverage-lmoments-doctest, notebooks
+    3.9 = py39-coverage-sbck-doctest, lint, docs
+
 [testenv:lint]
 description = Run code quality compliance tests under {basepython}
 skip_install = True
@@ -65,16 +73,15 @@ allowlist_externals =
 ;deps =
 ;extras =
 
-[testenv:notebooks_doctests{-coverage,}]
-description = Run notebooks and doctests with pytest under {basepython}
+[testenv:notebooks{-prefetch,}]
+description = Run notebooks with pytest under {basepython}
 commands =
     pytest --no-cov --nbval --dist=loadscope --rootdir=tests/ --ignore=docs/notebooks/example.ipynb docs/notebooks
-    pytest --rootdir=tests/ --xdoctest xclim
+commands_post =
 
 [testenv:offline{-prefetch,}{-coverage,}]
 description = Run tests with pytest under {basepython}, preventing socket connections (except for unix sockets for async support)
 commands:
-    prefetch: xclim prefetch_testing_data
     python -c 'print("Running offline tests with positional arguments: --disable-socket --allow-unix-socket --m \"not requires_internet\"")'
     python -c 'print("These can be overwritten with: tox -e offline -- -m \"some other marker statement\"")'
     pytest --disable-socket --allow-unix-socket {posargs:-m 'not requires_internet'}
@@ -99,14 +106,11 @@ passenv =
     XCLIM_*
 extras = dev
 deps =
-    # FIXME: Remove when numba 0.59.0 is released
-    numba: numba==0.59.0rc1
-    numba: llvmlite==0.42.0rc1
     coverage: coveralls
     upstream: -rrequirements_upstream.txt
     sbck: pybind11
     lmoments: lmoments3
-    notebooks_doctests: lmoments3
+    notebooks: lmoments3
 install_command = python -m pip install --no-user {opts} {packages}
 download = True
 commands_pre =
@@ -115,11 +119,13 @@ commands_pre =
     xclim show_version_info
     python -m pip check
     xclim --help
+    prefetch: xclim prefetch_testing_data
 commands =
     prefetch: xclim prefetch_testing_data
-    doctest: pytest --no-cov --rootdir=tests/ --xdoctest xclim
     pytest {posargs}
+    doctest: pytest --rootdir=tests/ --xdoctest xclim
 commands_post =
     coverage: - coveralls
 allowlist_externals =
     git
+    xclim
diff --git a/xclim/__init__.py b/xclim/__init__.py
index ccc4ad28f..e33337d19 100644
--- a/xclim/__init__.py
+++ b/xclim/__init__.py
@@ -16,7 +16,7 @@
 
 __author__ = """Travis Logan"""
 __email__ = "logan.travis@ouranos.ca"
-__version__ = "0.48.3-dev.5"
+__version__ = "0.48.3-dev.6"
 
 
 _module_data = _files("xclim.data")
diff --git a/xclim/ensembles/_robustness.py b/xclim/ensembles/_robustness.py
index 00e9f1b1b..9c1603336 100644
--- a/xclim/ensembles/_robustness.py
+++ b/xclim/ensembles/_robustness.py
@@ -91,11 +91,15 @@ def robustness_fractions(  # noqa: C901
                 The weighted fraction of valid members showing significant change.
                 Passing `test=None` yields change_frac = 1 everywhere. Same type as `fut`.
         positive :
-                The weighted fraction of valid members showing positive change, no matter if it is significant or not.
+                The weighted fraction of valid members showing strictly positive change, no matter if it is significant or not.
         changed_positive :
-                The weighted fraction of valid members showing significant and positive change (]0, 1]).
+                The weighted fraction of valid members showing significant and positive change.
+        negative :
+                The weighted fraction of valid members showing strictly negative change, no matter if it is significant or not.
+        changed_negative :
+                The weighted fraction of valid members showing significant and negative change.
         agree :
-                The weighted fraction of valid members agreeing on the sign of change. It is the maximum between positive and 1 - positive.
+                The weighted fraction of valid members agreeing on the sign of change. It is the maximum between positive, negative and the rest.
         valid :
                 The weighted fraction of valid members. A member is valid is there are no NaNs along the time axes of `fut` and  `ref`.
         pvals :
@@ -106,7 +110,7 @@ def robustness_fractions(  # noqa: C901
     The table below shows the coefficient needed to retrieve the number of members
     that have the indicated characteristics, by multiplying it by the total
     number of members (`fut.realization.size`) and by `valid_frac`, assuming uniform weights.
-    For compactness, we rename the outputs cf, cpf and pf.
+    For compactness, we rename the outputs cf, pf, cpf, nf and cnf.
 
     +-----------------+--------------------+------------------------+------------+
     |                 | Significant change | Non-significant change | Any change |
@@ -115,9 +119,11 @@ def robustness_fractions(  # noqa: C901
     +-----------------+--------------------+------------------------+------------+
     | Positive change | cpf                | pf - cpf               | pf         |
     +-----------------+--------------------+------------------------+------------+
-    | Negative change | (cf - cpf)         | 1 - pf - (cf -cpf)     | 1 - pf     |
+    | Negative change | cnf                | nf - cnf               | nf         |
     +-----------------+--------------------+------------------------+------------+
 
+    And members showing absolutely no change are ``1 - nf - pf``.
+
     Available statistical tests are :
 
     {tests_doc}
@@ -213,10 +219,16 @@ def robustness_fractions(  # noqa: C901
     n_valid = valid.weighted(w).sum(realization)
     change_frac = changed.where(valid).weighted(w).sum(realization) / n_valid
     pos_frac = (delta > 0).where(valid).weighted(w).sum(realization) / n_valid
+    neg_frac = (delta < 0).where(valid).weighted(w).sum(realization) / n_valid
     change_pos_frac = ((delta > 0) & changed).where(valid).weighted(w).sum(
         realization
     ) / n_valid
-    agree_frac = xr.concat((pos_frac, 1 - pos_frac), "sign").max("sign")
+    change_neg_frac = ((delta < 0) & changed).where(valid).weighted(w).sum(
+        realization
+    ) / n_valid
+    agree_frac = xr.concat((pos_frac, neg_frac, 1 - pos_frac - neg_frac), "sign").max(
+        "sign"
+    )
 
     # Metadata
     kwargs_str = gen_call_string("", **test_params)[1:-1]
@@ -233,7 +245,7 @@ def robustness_fractions(  # noqa: C901
                 test=str(test),
             ),
             "positive": pos_frac.assign_attrs(
-                description="Fraction of valid members showing positive change.",
+                description="Fraction of valid members showing strictly positive change.",
                 units="",
             ),
             "changed_positive": change_pos_frac.assign_attrs(
@@ -242,12 +254,25 @@ def robustness_fractions(  # noqa: C901
                 units="",
                 test=str(test),
             ),
+            "negative": neg_frac.assign_attrs(
+                description="Fraction of valid members showing strictly negative change.",
+                units="",
+            ),
+            "changed_negative": change_neg_frac.assign_attrs(
+                description="Fraction of valid members showing significant and negative change. "
+                + test_str,
+                units="",
+                test=str(test),
+            ),
             "valid": valid_frac.assign_attrs(
                 description="Fraction of valid members (No missing values along time).",
                 units="",
             ),
             "agree": agree_frac.assign_attrs(
-                description="Fraction of valid members agreeing on the sign of change. Maximum between pos_frac and 1 - pos_frac.",
+                description=(
+                    "Fraction of valid members agreeing on the sign of change. "
+                    "Maximum between the positive, negative and no change fractions."
+                ),
                 units="",
             ),
         },