diff --git a/scripts/kernel-test.sh b/scripts/kernel-test.sh
index 925520bcab..708ae6ba7f 100755
--- a/scripts/kernel-test.sh
+++ b/scripts/kernel-test.sh
@@ -34,7 +34,9 @@ VOL="${BUILD_VOL:-orcacad_kerneltest}"
 # Two pre-existing failures are excluded by default (see [known-broken] in
 # test_caddocument.cpp): one of them SIGABRTs inside the vendored solver and takes the
 # whole process down, so without this exclusion a green run is simply unreachable and the
-# suite stops after ~12 of 32 cases. CI runs everything and still reports both.
+# suite stops after ~12 of 32 cases. Both are also tagged [NotWorking], the ctest label
+# the mainline fork's CI excludes, so they no longer redden its Unit Tests job on every
+# commit; they stay tracked as snaporca-tkz and snaporca-kzy rather than as CI noise.
 TAGS="${TAGS:-[CadDocument]~[known-broken]}"
 HOST=""
 
diff --git a/tests/libslic3r/test_caddocument.cpp b/tests/libslic3r/test_caddocument.cpp
index 428a33bcd3..ca14a887f1 100644
--- a/tests/libslic3r/test_caddocument.cpp
+++ b/tests/libslic3r/test_caddocument.cpp
@@ -611,9 +611,15 @@ TEST_CASE("entity constraints: point-on-line positions a centre onto an axis", "
 // [known-broken]: the "tangent line to circle" SECTION below aborts inside the vendored
 // solver (slvs/dsc.h FindById, "Cannot find handle"). SIGABRT is fatal to the whole Catch2
 // process, so this one case takes the entire suite down with it and no later test runs.
-// Tagged so the delegated dev loop (scripts/kernel-test.sh) can exclude it and still reach
-// a green baseline; CI runs every test and keeps reporting it, so the bug stays visible.
-TEST_CASE("entity constraints: tangent/midpoint/symmetric/angle", "[CadDocument][known-broken]")
+// [known-broken] lets the delegated dev loop (scripts/kernel-test.sh) exclude it and still
+// reach a green baseline. [NotWorking] is the ctest label the mainline-based fork's CI already
+// excludes (run_unit_tests.sh passes -LE NotWorking, and catch_discover_tests registers Catch2
+// tags as ctest labels via ADD_TAGS_AS_LABELS) — so its Unit Tests job stops being red on every
+// single commit, which was training everyone to ignore it and would have hidden the next real
+// regression. The tag is inert on the Snapmaker fork, which has neither that runner nor
+// catch_discover_tests; the two files are kept identical because they are ported wholesale
+// between the forks. The bug stays visible in the tracker instead — see snaporca-tkz.
+TEST_CASE("entity constraints: tangent/midpoint/symmetric/angle", "[CadDocument][known-broken][NotWorking]")
 {
     using R = SketchPointRole;
     using T = SketchConstraintType;
@@ -962,9 +968,10 @@ TEST_CASE("extrude taper + up-to-face distance", "[CadDocument]")
 }
 
 // [known-broken]: pre-existing failure, the cut groove volume does not meet the asserted
-// threshold. Excluded from the delegated dev loop so a green run means "I broke nothing";
-// CI still runs and reports it.
-TEST_CASE("internal thread cuts a visible groove into the bore wall", "[CadDocument][known-broken]")
+// threshold. Excluded from the delegated dev loop so a green run means "I broke nothing",
+// and [NotWorking] excludes it from CI's ctest gate for the same reason as the case above.
+// Tracked in the issue tracker instead — see snaporca-kzy.
+TEST_CASE("internal thread cuts a visible groove into the bore wall", "[CadDocument][known-broken][NotWorking]")
 {
     using namespace Slic3r;
     SketchPlane xy = SketchPlane::XY();