diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index d20921328..68706f190 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -227,12 +227,13 @@ jobs:
             group: yield
             timeout_minutes: 20
             cargo_test_args: ""
-            # Both files share this one crate (issue #923, Phase 3 of the
+            # All three files share this one crate (issue #923, Phase 3 of the
             # statistical/yield epic #710, added `klt yield-sensitivity` to
-            # the same `klt_yield_native` extension `klt yield` uses).
-            test_paths: tests/test_yield.py tests/test_yield_sensitivity.py
+            # the same `klt_yield_native` extension `klt yield` uses; issue
+            # #2675 added the `klt yield-campaign` native end-to-end tests).
+            test_paths: tests/test_yield.py tests/test_yield_campaign.py tests/test_yield_sensitivity.py
             extension_module: klt_yield_native
-            no_skip_message: "klt_yield_native did not install -- tests/test_yield.py's and tests/test_yield_sensitivity.py's statistics tiers would have skipped"
+            no_skip_message: "klt_yield_native did not install -- tests/test_yield.py's, tests/test_yield_campaign.py's and tests/test_yield_sensitivity.py's statistics tiers would have skipped"
     defaults:
       run:
         working-directory: ${{ matrix.dir }}
diff --git a/tests/test_yield_campaign.py b/tests/test_yield_campaign.py
index 6496d6c55..f004c9843 100644
--- a/tests/test_yield_campaign.py
+++ b/tests/test_yield_campaign.py
@@ -22,11 +22,13 @@
 import importlib.util
 import json
 import os
+import shutil
 from pathlib import Path
 
 import pytest
 
 from helpers.subprocess_fakes import fake_completed
+from klayout_tools import _paths as paths_module
 from klayout_tools import sim, yield_campaign
 from klayout_tools.yield_analysis import _read_samples
 from klayout_tools.yield_campaign import CampaignError, run_campaign
@@ -77,7 +79,28 @@ def _base_spec(**overrides) -> dict:
     return spec
 
 
+def _stub_ngspice_discovery(monkeypatch) -> None:
+    """Issue #2675: `sim.py` resolves the `ngspice` binary (issue #2423)
+    *before* dispatching any corner, i.e. before the `subprocess.run` stubs
+    below take effect. Stub resolution to the bare name (mirrors
+    `test_sim.py`'s `_bare_name_ngspice_resolution`) and drop any host
+    `$KLT_NGSPICE_BINARY`, so these tests do not need a real install.
+    Production discovery is untouched; only called by tests that also stub
+    `subprocess.run`."""
+    real_which = shutil.which
+
+    def fake_which(cmd, *args, **kwargs):
+        if cmd == "ngspice":
+            return cmd
+        return real_which(cmd, *args, **kwargs)
+
+    monkeypatch.delenv("KLT_NGSPICE_BINARY", raising=False)
+    monkeypatch.setattr(paths_module.shutil, "which", fake_which)
+
+
 def _stub_subprocess_run(monkeypatch, *, log_text: str) -> None:
+    _stub_ngspice_discovery(monkeypatch)
+
     def fake_run(cmd, capture_output, text, timeout, cwd=None):
         log_path = cmd[cmd.index("-o") + 1]
         with open(log_path, "w", encoding="utf-8") as handle:
@@ -636,6 +659,7 @@ def _stub_subprocess_run_per_sample(monkeypatch, *, values: list[str]) -> None:
     """Like `_stub_subprocess_run`, but emits a different measured value per
     ngspice invocation, so a single campaign can mix trustworthy and
     implausible draws."""
+    _stub_ngspice_discovery(monkeypatch)
     calls = {"n": 0}
 
     def fake_run(cmd, capture_output, text, timeout, cwd=None):
@@ -722,9 +746,8 @@ def test_run_campaign_hands_yield_a_netlist_that_unwraps_to_a_string(
 ):
     """The regression, over the **real** `run_sim` -> `_read_samples` seam
     rather than a hand-built fixture -- only the native statistics core is
-    stubbed, so this runs everywhere (`test_yield_campaign.py` is not in
-    CI's `klt_yield_native` matrix leg, which is why the `requires_native`
-    end-to-end below cannot be the only cover for this).
+    stubbed, so this runs everywhere (the `requires_native`
+    end-to-end below also runs in CI's `klt_yield_native` matrix leg).
 
     `run_campaign` dumps `run_sim`'s live report verbatim to
     `sample-set.json` and hands the path straight to `run_yield`, so