diff --git a/src/libstore/build/find-cycles.cc b/src/libstore/build/find-cycles.cc
new file mode 100644
index 000000000..2ac63522a
--- /dev/null
+++ b/src/libstore/build/find-cycles.cc
@@ -0,0 +1,328 @@
+#include "find-cycles.hh"
+
+#include "nix/store/store-api.hh"
+#include "nix/util/file-system.hh"
+#include "nix/util/archive.hh"
+#include "nix/util/base-nix-32.hh"
+
+#include <filesystem>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+namespace nix {
+
+// Hash length in characters (32 for base32-encoded sha256)
+static constexpr size_t refLength = StorePath::HashLen;
+
+// Maximum expected file path length for buffer carry-over
+static constexpr size_t MAX_FILEPATH_LENGTH = 1000;
+
+void scanForCycleEdges(const Path & path, const StorePathSet & refs, StoreCycleEdgeVec & edges)
+{
+    StringSet hashes;
+    std::map<std::string, StorePath> hashPathMap;
+
+    // Extract the store directory from the path
+    // Example: /run/user/1000/nix-test/store/abc-foo -> /run/user/1000/nix-test/store/
+    auto storePrefixPath = std::filesystem::path(path);
+    storePrefixPath.remove_filename();
+    std::string storePrefix = storePrefixPath.string();
+
+    debug("scanForCycleEdges: storePrefixPath = %s", storePrefixPath.string());
+    debug("scanForCycleEdges: storePrefix = %s", storePrefix);
+
+    // Build map of hash -> StorePath and collect hashes to search for
+    for (auto & i : refs) {
+        std::string hashPart(i.hashPart());
+        auto inserted = hashPathMap.emplace(hashPart, i).second;
+        assert(inserted);
+        hashes.insert(hashPart);
+    }
+
+    scanForCycleEdges2(path, hashes, edges, storePrefix);
+}
+
+void scanForCycleEdges2(std::string path, const StringSet & hashes, StoreCycleEdgeVec & edges, std::string storeDir)
+{
+    auto st = lstat(path);
+
+    debug("scanForCycleEdges2: scanning path = %s", path);
+
+    if (S_ISREG(st.st_mode)) {
+        // Handle regular files - scan contents for hash references
+        AutoCloseFD fd = open(path.c_str(), O_RDONLY | O_CLOEXEC);
+        if (!fd)
+            throw SysError("opening file '%1%'", path);
+
+        std::vector<char> buf(65536);
+        size_t rest = st.st_size;
+        size_t start = 0;
+
+        // Buffer to carry data between reads (for references spanning chunks)
+        std::vector<char> bufCarry(MAX_FILEPATH_LENGTH);
+        bool bufCarryUsed = false;
+        size_t bufCarrySize = 0;
+
+        while (rest > 0) {
+            auto n = std::min(rest, buf.size());
+            readFull(fd.get(), buf.data(), n);
+
+            debug("scanForCycleEdges2: read file %s: n = %lu", path, n);
+
+            // Check if we have carry-over data from previous iteration
+            if (bufCarryUsed) {
+                // Search in the overlap region (carry + start of new buffer)
+                size_t searchSize = std::min(bufCarrySize + n, MAX_FILEPATH_LENGTH);
+                std::vector<char> searchBuf(searchSize);
+
+                // Copy carry buffer
+                std::copy(bufCarry.begin(), bufCarry.begin() + bufCarrySize, searchBuf.begin());
+
+                // Copy start of new buffer
+                size_t newDataSize = searchSize - bufCarrySize;
+                std::copy(buf.begin(), buf.begin() + newDataSize, searchBuf.begin() + bufCarrySize);
+
+                // Search for hashes in the overlap region
+                for (size_t i = 0; i + refLength <= searchBuf.size();) {
+                    bool match = true;
+                    for (int j = refLength - 1; j >= 0; --j) {
+                        if (!BaseNix32::lookupReverse(searchBuf[i + j])) {
+                            i += j + 1;
+                            match = false;
+                            break;
+                        }
+                    }
+                    if (!match)
+                        continue;
+
+                    std::string hash(searchBuf.begin() + i, searchBuf.begin() + i + refLength);
+
+                    if (hashes.find(hash) != hashes.end()) {
+                        debug("scanForCycleEdges2: found hash '%s' in overlap region", hash);
+
+                        // Try to find the full path
+                        size_t storeDirLength = storeDir.size();
+                        if (i >= storeDirLength) {
+                            std::string targetPath = storeDir + hash;
+                            StoreCycleEdge edge;
+                            edge.push_back(path);
+                            edge.push_back(targetPath);
+                            edges.push_back(edge);
+                        }
+                    }
+                    ++i;
+                }
+
+                bufCarryUsed = false;
+            }
+
+            // Search in the main buffer
+            for (size_t i = 0; i + refLength <= n;) {
+                bool match = true;
+                for (int j = refLength - 1; j >= 0; --j) {
+                    if (!BaseNix32::lookupReverse(buf[i + j])) {
+                        i += j + 1;
+                        match = false;
+                        break;
+                    }
+                }
+                if (!match)
+                    continue;
+
+                // Found a potential hash
+                std::string hash(buf.begin() + i, buf.begin() + i + refLength);
+
+                if (hashes.find(hash) != hashes.end()) {
+                    debug("scanForCycleEdges2: found reference to hash '%s' at offset %lu", hash, start + i);
+
+                    // Try to reconstruct the full store path
+                    size_t storeDirLength = storeDir.size();
+                    std::string targetPath = storeDir + hash;
+                    std::string targetStorePath;
+
+                    // Check if we have storeDir + hash in the buffer
+                    if (i >= (size_t) storeDirLength
+                        && std::string(buf.begin() + i - storeDirLength, buf.begin() + i + refLength) == targetPath) {
+
+                        debug("scanForCycleEdges2: found store path prefix at offset %lu", start + i - storeDirLength);
+
+                        // Try to find the complete path by checking what exists on disk
+                        // We probe incrementally to find the longest existing path
+                        size_t testNameLength = refLength + 2; // Minimum: hash + "-x"
+                        size_t targetPathLastEnd = 0;
+                        bool foundStorePath = false;
+                        bool foundPath = false;
+
+                        for (; testNameLength < 255 && i + (size_t) targetPathLastEnd + (size_t) testNameLength <= n;
+                             testNameLength++) {
+                            std::string testPath(
+                                buf.begin() + i - storeDirLength, buf.begin() + i + targetPathLastEnd + testNameLength);
+
+                            struct stat testStat;
+                            if (stat(testPath.c_str(), &testStat) == 0) {
+                                debug("scanForCycleEdges2: found existing path: %s", testPath);
+
+                                if (!foundStorePath) {
+                                    // First match is the store path component
+                                    targetStorePath = testPath.substr(storeDirLength);
+                                    foundStorePath = true;
+                                }
+
+                                foundPath = true;
+                                targetPath = testPath;
+
+                                // Check if this is a directory (followed by '/')
+                                if (buf[i + targetPathLastEnd + testNameLength] == '/') {
+                                    debug("scanForCycleEdges2: path is a directory, continuing");
+                                    targetPathLastEnd += testNameLength;
+                                    testNameLength = 1; // Reset for next component
+                                    continue;
+                                }
+                            }
+                        }
+
+                        if (foundPath) {
+                            debug("scanForCycleEdges2: cycle edge: %s -> %s", path, targetPath);
+                        } else {
+                            // Couldn't find exact path, use store path + hash
+                            targetPath = storeDir + hash;
+                        }
+                    }
+
+                    StoreCycleEdge edge;
+                    edge.push_back(path);
+                    edge.push_back(targetPath);
+                    edges.push_back(edge);
+                }
+                ++i;
+            }
+
+            start += n;
+            rest -= n;
+
+            // Carry over the end of the buffer for next iteration
+            if (n == buf.size() && rest > 0) {
+                size_t carrySize = std::min(MAX_FILEPATH_LENGTH, n);
+                std::copy(buf.end() - carrySize, buf.end(), bufCarry.begin());
+                bufCarrySize = carrySize;
+                bufCarryUsed = true;
+            }
+        }
+    } else if (S_ISDIR(st.st_mode)) {
+        // Handle directories - recursively scan contents
+        std::map<std::string, std::string> unhacked;
+
+        for (DirectoryIterator i(path); i != DirectoryIterator(); ++i) {
+            std::string entryName = i->path().filename().string();
+
+#ifdef __APPLE__
+            // Handle case-insensitive filesystems on macOS
+            std::string name(entryName);
+            size_t pos = entryName.find(caseHackSuffix);
+            if (pos != std::string::npos) {
+                debug("removing case hack suffix from '%s'", path + "/" + entryName);
+                name.erase(pos);
+            }
+            if (unhacked.find(name) != unhacked.end()) {
+                throw Error(
+                    "file name collision between '%1%' and '%2%'", path + "/" + unhacked[name], path + "/" + entryName);
+            }
+            unhacked[name] = entryName;
+#else
+            unhacked[entryName] = entryName;
+#endif
+        }
+
+        for (auto & [name, actualName] : unhacked) {
+            debug("scanForCycleEdges2: recursing into %s/%s", path, actualName);
+            scanForCycleEdges2(path + "/" + actualName, hashes, edges, storeDir);
+        }
+    } else if (S_ISLNK(st.st_mode)) {
+        // Handle symlinks - scan link target for hash references
+        std::string linkTarget = readLink(path);
+
+        debug("scanForCycleEdges2: scanning symlink %s -> %s", path, linkTarget);
+
+        for (size_t i = 0; i + refLength <= linkTarget.size();) {
+            bool match = true;
+            for (int j = refLength - 1; j >= 0; --j) {
+                if (!BaseNix32::lookupReverse(linkTarget[i + j])) {
+                    i += j + 1;
+                    match = false;
+                    break;
+                }
+            }
+            if (!match)
+                continue;
+
+            std::string ref(linkTarget.begin() + i, linkTarget.begin() + i + refLength);
+
+            if (hashes.find(ref) != hashes.end()) {
+                debug("scanForCycleEdges2: found reference '%s' in symlink at offset %lu", ref, i);
+
+                // Try to extract full path from link target
+                std::string targetPath = storeDir + ref;
+
+                StoreCycleEdge edge;
+                edge.push_back(path);
+                edge.push_back(targetPath);
+                edges.push_back(edge);
+            }
+            ++i;
+        }
+    } else {
+        throw Error("file '%1%' has an unsupported type", path);
+    }
+}
+
+void transformEdgesToMultiedges(StoreCycleEdgeVec & edges, StoreCycleEdgeVec & multiedges)
+{
+    debug("transformEdgesToMultiedges: processing %lu edges", edges.size());
+
+    for (auto & edge2 : edges) {
+        bool edge2Joined = false;
+
+        for (auto & edge1 : multiedges) {
+            debug("comparing edge1 (size=%lu) with edge2 (size=%lu)", edge1.size(), edge2.size());
+
+            // Check if edge1.back() == edge2.front()
+            // This means we can extend edge1 by appending edge2
+            if (edge1.back() == edge2.front()) {
+                debug("appending: edge1.back()='%s' == edge2.front()='%s'", edge1.back(), edge2.front());
+
+                // Append all but the first element of edge2 (to avoid duplication)
+                for (size_t i = 1; i < edge2.size(); i++) {
+                    debug("  appending edge2[%lu] = %s", i, edge2[i]);
+                    edge1.push_back(edge2[i]);
+                }
+                edge2Joined = true;
+                break;
+            }
+
+            // Check if edge2.back() == edge1.front()
+            // This means we can extend edge1 by prepending edge2
+            if (edge2.back() == edge1.front()) {
+                debug("prepending: edge2.back()='%s' == edge1.front()='%s'", edge2.back(), edge1.front());
+
+                // Prepend all but the last element of edge2 (to avoid duplication)
+                for (int i = edge2.size() - 2; i >= 0; i--) {
+                    debug("  prepending edge2[%d] = %s", i, edge2[i]);
+                    edge1.push_front(edge2[i]);
+                }
+                edge2Joined = true;
+                break;
+            }
+        }
+
+        if (!edge2Joined) {
+            debug("edge2 is new, adding as separate multiedge");
+            multiedges.push_back(edge2);
+        }
+    }
+
+    debug("transformEdgesToMultiedges: result has %lu multiedges", multiedges.size());
+}
+
+} // namespace nix
diff --git a/src/libstore/build/find-cycles.hh b/src/libstore/build/find-cycles.hh
new file mode 100644
index 000000000..86f49c6a5
--- /dev/null
+++ b/src/libstore/build/find-cycles.hh
@@ -0,0 +1,69 @@
+#pragma once
+///@file
+
+#include "nix/store/store-api.hh"
+#include "nix/util/types.hh"
+
+#include <string>
+#include <deque>
+#include <vector>
+
+namespace nix {
+
+/**
+ * Represents a cycle edge as a sequence of file paths.
+ * Uses deque to allow efficient prepend/append when joining edges.
+ *
+ * Example: {"/nix/store/abc-foo/file1", "/nix/store/def-bar/file2"}
+ * represents a reference from file1 to file2.
+ */
+typedef std::deque<std::string> StoreCycleEdge;
+
+/**
+ * A collection of cycle edges found during scanning.
+ */
+typedef std::vector<StoreCycleEdge> StoreCycleEdgeVec;
+
+/**
+ * Scan output paths to find cycle edges with detailed file paths.
+ *
+ * This is the second pass of cycle detection. The first pass (scanForReferences)
+ * detects that a cycle exists. This function provides detailed information about
+ * where the cycles occur in the actual file system.
+ *
+ * @param path The store path to scan (e.g., an output directory)
+ * @param refs The set of potentially referenced store paths
+ * @param edges Output parameter that accumulates found cycle edges
+ */
+void scanForCycleEdges(const Path & path, const StorePathSet & refs, StoreCycleEdgeVec & edges);
+
+/**
+ * Recursively scan files and directories for hash references.
+ *
+ * This function walks the file system tree and searches for store path hashes
+ * in file contents, symlinks, etc. When a hash is found, it attempts to
+ * reconstruct the full store path by checking what files actually exist.
+ *
+ * @param path Current path being scanned
+ * @param hashes Set of hash strings to look for (32-char base32 hashes)
+ * @param edges Output parameter that accumulates found cycle edges
+ * @param storeDir The store directory prefix (e.g., "/nix/store/")
+ */
+void scanForCycleEdges2(std::string path, const StringSet & hashes, StoreCycleEdgeVec & edges, std::string storeDir);
+
+/**
+ * Transform individual edges into connected multi-edges (paths).
+ *
+ * Takes a list of edges like [A→B, B→C, C→A] and connects them into
+ * longer paths like [A→B→C→A]. This makes it easier to visualize the
+ * actual cycle paths.
+ *
+ * The algorithm is greedy: it tries to extend each edge by finding
+ * matching edges to prepend or append.
+ *
+ * @param edges Input edges to transform
+ * @param multiedges Output parameter with connected paths
+ */
+void transformEdgesToMultiedges(StoreCycleEdgeVec & edges, StoreCycleEdgeVec & multiedges);
+
+} // namespace nix
diff --git a/src/libstore/meson.build b/src/libstore/meson.build
index d1b3666cc..faab698dc 100644
--- a/src/libstore/meson.build
+++ b/src/libstore/meson.build
@@ -267,6 +267,7 @@ sources = files(
   'build/derivation-trampoline-goal.cc',
   'build/drv-output-substitution-goal.cc',
   'build/entry-points.cc',
+  'build/find-cycles.cc',
   'build/goal.cc',
   'build/substitution-goal.cc',
   'build/worker.cc',
diff --git a/src/libstore/unix/build/derivation-builder.cc b/src/libstore/unix/build/derivation-builder.cc
index d6abf85e3..bd00b7701 100644
--- a/src/libstore/unix/build/derivation-builder.cc
+++ b/src/libstore/unix/build/derivation-builder.cc
@@ -45,6 +45,7 @@
 
 #include "store-config-private.hh"
 #include "build/derivation-check.hh"
+#include "build/find-cycles.hh"
 
 #if NIX_WITH_AWS_AUTH
 #  include "nix/store/aws-creds.hh"
@@ -1473,43 +1474,100 @@ SingleDrvOutputs DerivationBuilderImpl::registerOutputs()
         outputStats.insert_or_assign(outputName, std::move(st));
     }
 
-    auto sortedOutputNames = topoSort(
-        outputsToSort,
-        {[&](const std::string & name) {
-            auto orifu = get(outputReferencesIfUnregistered, name);
-            if (!orifu)
-                throw BuildError(
-                    BuildResult::Failure::OutputRejected,
-                    "no output reference for '%s' in build of '%s'",
-                    name,
-                    store.printStorePath(drvPath));
-            return std::visit(
-                overloaded{
-                    /* Since we'll use the already installed versions of these, we
-                       can treat them as leaves and ignore any references they
-                       have. */
-                    [&](const AlreadyRegistered &) { return StringSet{}; },
-                    [&](const PerhapsNeedToRegister & refs) {
-                        StringSet referencedOutputs;
-                        /* FIXME build inverted map up front so no quadratic waste here */
-                        for (auto & r : refs.refs)
-                            for (auto & [o, p] : scratchOutputs)
-                                if (r == p)
-                                    referencedOutputs.insert(o);
-                        return referencedOutputs;
+    std::vector<std::string> sortedOutputNames;
+
+    try {
+        sortedOutputNames = topoSort(
+            outputsToSort,
+            {[&](const std::string & name) {
+                auto orifu = get(outputReferencesIfUnregistered, name);
+                if (!orifu)
+                    throw BuildError(
+                        BuildResult::Failure::OutputRejected,
+                        "no output reference for '%s' in build of '%s'",
+                        name,
+                        store.printStorePath(drvPath));
+                return std::visit(
+                    overloaded{
+                        /* Since we'll use the already installed versions of these, we
+                           can treat them as leaves and ignore any references they
+                           have. */
+                        [&](const AlreadyRegistered &) { return StringSet{}; },
+                        [&](const PerhapsNeedToRegister & refs) {
+                            StringSet referencedOutputs;
+                            /* FIXME build inverted map up front so no quadratic waste here */
+                            for (auto & r : refs.refs)
+                                for (auto & [o, p] : scratchOutputs)
+                                    if (r == p)
+                                        referencedOutputs.insert(o);
+                            return referencedOutputs;
+                        },
                     },
-                },
-                *orifu);
-        }},
-        {[&](const std::string & path, const std::string & parent) {
-            // TODO with more -vvvv also show the temporary paths for manual inspection.
-            return BuildError(
-                BuildResult::Failure::OutputRejected,
-                "cycle detected in build of '%s' in the references of output '%s' from output '%s'",
-                store.printStorePath(drvPath),
-                path,
-                parent);
-        }});
+                    *orifu);
+            }},
+            {[&](const std::string & path, const std::string & parent) {
+                // TODO with more -vvvv also show the temporary paths for manual inspection.
+                return BuildError(
+                    BuildResult::Failure::OutputRejected,
+                    "cycle detected in build of '%s' in the references of output '%s' from output '%s'",
+                    store.printStorePath(drvPath),
+                    path,
+                    parent);
+            }});
+    } catch (std::exception & e) {
+        debug("cycle detected during topoSort, analyzing for detailed error report");
+
+        // Scan all outputs for cycle edges with exact file paths
+        StoreCycleEdgeVec edges;
+        for (auto & [outputName, _] : drv.outputs) {
+            auto scratchOutput = get(scratchOutputs, outputName);
+            if (!scratchOutput)
+                continue;
+
+            auto actualPath = realPathInSandbox(store.printStorePath(*scratchOutput));
+            debug("scanning output '%s' at path '%s' for cycle edges", outputName, actualPath);
+
+            scanForCycleEdges(actualPath, referenceablePaths, edges);
+        }
+
+        if (edges.empty()) {
+            debug("no detailed cycle edges found, re-throwing original error");
+            throw;
+        }
+
+        debug("found %lu cycle edges, transforming to connected paths", edges.size());
+
+        // Transform individual edges into connected multi-edges (paths)
+        StoreCycleEdgeVec multiedges;
+        transformEdgesToMultiedges(edges, multiedges);
+
+        // Build detailed error message
+        std::string cycleDetails = fmt("Detailed cycle analysis found %d cycle path(s):", multiedges.size());
+
+        for (size_t i = 0; i < multiedges.size(); i++) {
+            auto & multiedge = multiedges[i];
+            cycleDetails += fmt("\n\nCycle %d:", i + 1);
+            for (auto & file : multiedge) {
+                cycleDetails += fmt("\n  → %s", file);
+            }
+        }
+
+        cycleDetails +=
+            fmt("\n\nThis means there are circular references between output files.\n"
+                "The build cannot proceed because the outputs reference each other.");
+
+        // Add hint with temp paths for debugging
+        if (settings.keepFailed || verbosity >= lvlDebug) {
+            cycleDetails +=
+                fmt("\n\nNote: Build outputs are kept for inspection.\n"
+                    "You can examine the files listed above to understand the cycle.");
+        }
+
+        // Throw new error with original message + cycle details
+        BuildError * buildErr = dynamic_cast<BuildError *>(&e);
+        std::string originalMsg = buildErr ? std::string(buildErr->msg()) : std::string(e.what());
+        throw BuildError(BuildResult::Failure::OutputRejected, "%s\n\n%s", originalMsg, cycleDetails);
+    }
 
     std::reverse(sortedOutputNames.begin(), sortedOutputNames.end());
 
@@ -1848,12 +1906,61 @@ SingleDrvOutputs DerivationBuilderImpl::registerOutputs()
     /* Register each output path as valid, and register the sets of
        paths referenced by each of them.  If there are cycles in the
        outputs, this will fail. */
-    {
+    try {
         ValidPathInfos infos2;
         for (auto & [outputName, newInfo] : infos) {
             infos2.insert_or_assign(newInfo.path, newInfo);
         }
         store.registerValidPaths(infos2);
+    } catch (BuildError & e) {
+        debug("cycle detected during registerValidPaths, analyzing for detailed error report");
+
+        // Scan all outputs for cycle edges with exact file paths
+        StoreCycleEdgeVec edges;
+        for (auto & [outputName, newInfo] : infos) {
+            auto actualPath = store.toRealPath(store.printStorePath(newInfo.path));
+            debug("scanning registered output '%s' at path '%s' for cycle edges", outputName, actualPath);
+
+            scanForCycleEdges(actualPath, referenceablePaths, edges);
+        }
+
+        if (edges.empty()) {
+            debug("no detailed cycle edges found, re-throwing original error");
+            throw;
+        }
+
+        debug("found %lu cycle edges, transforming to connected paths", edges.size());
+
+        // Transform individual edges into connected multi-edges (paths)
+        StoreCycleEdgeVec multiedges;
+        transformEdgesToMultiedges(edges, multiedges);
+
+        // Build detailed error message
+        std::string cycleDetails = fmt("Detailed cycle analysis found %d cycle path(s):", multiedges.size());
+
+        for (size_t i = 0; i < multiedges.size(); i++) {
+            auto & multiedge = multiedges[i];
+            cycleDetails += fmt("\n\nCycle %d:", i + 1);
+            for (auto & file : multiedge) {
+                cycleDetails += fmt("\n  → %s", file);
+            }
+        }
+
+        cycleDetails +=
+            fmt("\n\nThis means there are circular references between output files.\n"
+                "The build cannot proceed because the outputs reference each other.");
+
+        // Add hint with temp paths for debugging
+        if (settings.keepFailed || verbosity >= lvlDebug) {
+            cycleDetails +=
+                fmt("\n\nNote: Build outputs were kept for inspection.\n"
+                    "You can examine the files listed above to understand the cycle.");
+        }
+
+        // Throw new error with original message + cycle details
+        BuildError * buildErr = dynamic_cast<BuildError *>(&e);
+        std::string originalMsg = buildErr ? std::string(buildErr->msg()) : std::string(e.what());
+        throw BuildError(BuildResult::Failure::OutputRejected, "%s\n\n%s", originalMsg, cycleDetails);
     }
 
     /* If we made it this far, we are sure the output matches the
diff --git a/tests/functional/multiple-outputs.nix b/tests/functional/multiple-outputs.nix
index 2c9243097..87a29b19c 100644
--- a/tests/functional/multiple-outputs.nix
+++ b/tests/functional/multiple-outputs.nix
@@ -83,6 +83,8 @@ rec {
     '';
   };
 
+  # Test for cycle detection with detailed error messages
+  # This creates multiple cycles: a→b→c→a and a→c→b→a
   cyclic =
     (mkDerivation {
       name = "cyclic-outputs";
@@ -92,10 +94,22 @@ rec {
         "c"
       ];
       builder = builtins.toFile "builder.sh" ''
-        mkdir $a $b $c
-        echo $a > $b/foo
-        echo $b > $c/bar
-        echo $c > $a/baz
+        mkdir -p $a/subdir $b/subdir $c/subdir
+
+        # First cycle: a → b → c → a
+        echo "$b/subdir/b-to-c" > $a/subdir/a-to-b
+        echo "$c/subdir/c-to-a" > $b/subdir/b-to-c
+        echo "$a/subdir/a-to-b" > $c/subdir/c-to-a
+
+        # Second cycle: a → c → b → a
+        echo "$c/subdir/c-to-b-2" > $a/subdir/a-to-c-2
+        echo "$b/subdir/b-to-a-2" > $c/subdir/c-to-b-2
+        echo "$a/subdir/a-to-c-2" > $b/subdir/b-to-a-2
+
+        # Non-cyclic reference (just for complexity)
+        echo "non-cyclic-data" > $a/data
+        echo "non-cyclic-data" > $b/data
+        echo "non-cyclic-data" > $c/data
       '';
     }).a;
 
diff --git a/tests/functional/multiple-outputs.sh b/tests/functional/multiple-outputs.sh
index f703fb02b..b115af202 100755
--- a/tests/functional/multiple-outputs.sh
+++ b/tests/functional/multiple-outputs.sh
@@ -91,9 +91,22 @@ nix-build multiple-outputs.nix -A a.first --no-out-link
 
 # Cyclic outputs should be rejected.
 echo "building cyclic..."
-if nix-build multiple-outputs.nix -A cyclic --no-out-link; then
+if cyclicOutput=$(nix-build multiple-outputs.nix -A cyclic --no-out-link 2>&1); then
     echo "Cyclic outputs incorrectly accepted!"
     exit 1
+else
+    echo "Cyclic outputs correctly rejected"
+    # Verify error message mentions cycles
+    echo "$cyclicOutput" | grepQuiet "cycle"
+
+    # Enhanced cycle error messages were added in 2.33
+    if isDaemonNewer "2.33"; then
+        echo "$cyclicOutput" | grepQuiet "Detailed cycle analysis"
+        echo "$cyclicOutput" | grepQuiet "Cycle 1:"
+        # The error should mention actual file paths with subdirectories
+        echo "$cyclicOutput" | grepQuiet "subdir"
+        echo "Enhanced cycle error messages verified"
+    fi
 fi
 
 # Do a GC. This should leave an empty store.