diff --git a/src/cli/install/download.zig b/src/cli/install/download.zig
index 66ecd03..131ffd7 100644
--- a/src/cli/install/download.zig
+++ b/src/cli/install/download.zig
@@ -78,6 +78,17 @@ pub fn progressBridge(ctx: *anyopaque, bytes_so_far: u64, content_length: ?u64)
     bar.update(clamped);
 }
 
+/// Sleeps for `ms` between retries. Returns false when the caller's
+/// stop signal cancels the sleep, true otherwise. std.Io cancellation
+/// is single-shot per task, so a swallowed Canceled would silently
+/// consume the request and the loop would keep retrying.
+fn cancellableBackoff(io: std.Io, ms: u64) bool {
+    std.Io.sleep(io, std.Io.Duration.fromNanoseconds(@intCast(ms * std.time.ns_per_ms)), .awake) catch |e| switch (e) {
+        error.Canceled => return false,
+    };
+    return true;
+}
+
 /// Download a bottle and commit to store. Runs in a worker thread.
 /// `http_pool` is shared across all download workers — each worker
 /// borrows a client for the duration of a single blob download and
@@ -152,7 +163,7 @@ pub fn downloadWorker(
                 break;
             }
             if (dl_attempt + 1 < max_attempts) {
-                std.Io.sleep(io, std.Io.Duration.fromNanoseconds(@intCast(retry_delays_ms[dl_attempt] * std.time.ns_per_ms)), .awake) catch {};
+                if (!cancellableBackoff(io, retry_delays_ms[dl_attempt])) break;
             }
         }
     }
@@ -692,3 +703,55 @@ test "MaterializeResult.kegPath reflects keg_path_len after write" {
     r.keg_path_len = path.len;
     try std.testing.expectEqualStrings(path, r.kegPath());
 }
+
+// Patches an inner Io's vtable so `sleep` reports cancellation on the
+// configured call index; non-canceled sleeps return immediately so the
+// retry-table delays don't pad the test runtime.
+const CancelSleepProbe = struct {
+    var vtable: std.Io.VTable = undefined;
+    var sleep_calls: usize = 0;
+    var cancel_at: usize = 1;
+
+    fn wrap(inner: std.Io, cancel_at_call: usize) std.Io {
+        vtable = inner.vtable.*;
+        vtable.sleep = sleepMaybeCanceled;
+        sleep_calls = 0;
+        cancel_at = cancel_at_call;
+        return .{ .userdata = inner.userdata, .vtable = &vtable };
+    }
+
+    fn sleepMaybeCanceled(_: ?*anyopaque, _: std.Io.Timeout) std.Io.Cancelable!void {
+        sleep_calls += 1;
+        if (sleep_calls >= cancel_at) return error.Canceled;
+    }
+};
+
+test "cancellableBackoff returns true when sleep completes normally" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    // Zero-ms request keeps the test deterministic without exercising the
+    // real clock; a successful sleep must report true so the retry loop
+    // moves on to the next attempt.
+    try std.testing.expect(cancellableBackoff(threaded.io(), 0));
+}
+
+test "cancellableBackoff returns false when sleep is cancelled" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    const cancel_io = CancelSleepProbe.wrap(threaded.io(), 1);
+    try std.testing.expect(!cancellableBackoff(cancel_io, 100));
+    try std.testing.expectEqual(@as(usize, 1), CancelSleepProbe.sleep_calls);
+}
+
+test "cancellableBackoff propagates cancellation when called repeatedly" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    // Cancel the third call: the prior two complete and report true,
+    // pinning that the helper isn't inadvertently latched after the
+    // first non-cancelled sleep.
+    const cancel_io = CancelSleepProbe.wrap(threaded.io(), 3);
+    try std.testing.expect(cancellableBackoff(cancel_io, 0));
+    try std.testing.expect(cancellableBackoff(cancel_io, 0));
+    try std.testing.expect(!cancellableBackoff(cancel_io, 0));
+    try std.testing.expectEqual(@as(usize, 3), CancelSleepProbe.sleep_calls);
+}
diff --git a/src/core/services/supervisor.zig b/src/core/services/supervisor.zig
index d536963..98a13f2 100644
--- a/src/core/services/supervisor.zig
+++ b/src/core/services/supervisor.zig
@@ -407,7 +407,12 @@ pub fn followLog(
     while (!interrupted()) {
         const n = f.readPositionalAll(io, &buf, offset) catch return SupervisorError.IoFailed;
         if (n == 0) {
-            std.Io.sleep(io, std.Io.Duration.fromNanoseconds(@intCast(poll_ns)), .awake) catch {};
+            // Cancelled sleep means SIGINT (or another stop signal) reached
+            // the io subsystem before `interrupted()` flipped; bail rather
+            // than busy-spin until the flag catches up.
+            std.Io.sleep(io, std.Io.Duration.fromNanoseconds(@intCast(poll_ns)), .awake) catch |e| switch (e) {
+                error.Canceled => break,
+            };
             continue;
         }
         writer.writeAll(buf[0..n]) catch return SupervisorError.IoFailed;
@@ -518,3 +523,126 @@ test "followLog flushes appended bytes between polls" {
     try testing.expect(std.mem.indexOf(u8, aw.written(), "appended\n") != null);
     try testing.expectEqual(@as(usize, 3), FollowProbe.calls);
 }
+
+// Patches an inner Io's vtable so `sleep` reports cancellation on the
+// configured call index; non-canceled sleeps return immediately so the
+// 200 ms poll cadence doesn't pad the test runtime.
+const CancelSleepProbe = struct {
+    var vtable: std.Io.VTable = undefined;
+    var sleep_calls: usize = 0;
+    var cancel_at: usize = 1;
+
+    fn wrap(inner: std.Io, cancel_at_call: usize) std.Io {
+        vtable = inner.vtable.*;
+        vtable.sleep = sleepMaybeCanceled;
+        sleep_calls = 0;
+        cancel_at = cancel_at_call;
+        return .{ .userdata = inner.userdata, .vtable = &vtable };
+    }
+
+    fn sleepMaybeCanceled(_: ?*anyopaque, _: std.Io.Timeout) std.Io.Cancelable!void {
+        sleep_calls += 1;
+        if (sleep_calls >= cancel_at) return error.Canceled;
+    }
+};
+
+test "followLog breaks when sleep reports cancellation on the first poll" {
+    testIoInit();
+    defer testIoDeinit();
+
+    const dir = "/tmp/malt_supervisor_follow_cancel";
+    std.Io.Dir.cwd().deleteTree(test_io, dir) catch {};
+    try std.Io.Dir.createDirAbsolute(test_io, dir, .default_dir);
+    defer std.Io.Dir.cwd().deleteTree(test_io, dir) catch {};
+
+    const log_path = dir ++ "/sample.log";
+    {
+        const f = try std.Io.Dir.createFileAbsolute(test_io, log_path, .{ .truncate = true });
+        defer f.close(test_io);
+        try f.writeStreamingAll(test_io, "seed\n");
+    }
+
+    var aw: std.Io.Writer.Allocating = .init(testing.allocator);
+    defer aw.deinit();
+
+    const cancel_io = CancelSleepProbe.wrap(test_io, 1);
+
+    FollowProbe.reset();
+    // High stop_at: only sleep cancellation should end the loop. If the loop
+    // ever spins past Canceled, FollowProbe.calls will exceed 1.
+    FollowProbe.stop_at = 16;
+
+    try followLog(cancel_io, testing.allocator, log_path, 0, &aw.writer, FollowProbe.cb);
+
+    try testing.expectEqual(@as(usize, 1), CancelSleepProbe.sleep_calls);
+    try testing.expectEqual(@as(usize, 1), FollowProbe.calls);
+}
+
+test "followLog breaks when sleep reports cancellation after several idle polls" {
+    testIoInit();
+    defer testIoDeinit();
+
+    const dir = "/tmp/malt_supervisor_follow_cancel_late";
+    std.Io.Dir.cwd().deleteTree(test_io, dir) catch {};
+    try std.Io.Dir.createDirAbsolute(test_io, dir, .default_dir);
+    defer std.Io.Dir.cwd().deleteTree(test_io, dir) catch {};
+
+    const log_path = dir ++ "/sample.log";
+    {
+        const f = try std.Io.Dir.createFileAbsolute(test_io, log_path, .{ .truncate = true });
+        defer f.close(test_io);
+        try f.writeStreamingAll(test_io, "seed\n");
+    }
+
+    var aw: std.Io.Writer.Allocating = .init(testing.allocator);
+    defer aw.deinit();
+
+    // Cancel the fourth sleep — three idle polls happen first, pinning that
+    // the catch arm fires regardless of how many normal sleeps preceded it
+    // and that the loop doesn't latch on the prior non-cancelled returns.
+    const cancel_io = CancelSleepProbe.wrap(test_io, 4);
+
+    FollowProbe.reset();
+    FollowProbe.stop_at = 16;
+
+    try followLog(cancel_io, testing.allocator, log_path, 0, &aw.writer, FollowProbe.cb);
+
+    try testing.expectEqual(@as(usize, 4), CancelSleepProbe.sleep_calls);
+    try testing.expectEqual(@as(usize, 4), FollowProbe.calls);
+}
+
+test "followLog flushes data appended before sleep cancellation lands" {
+    testIoInit();
+    defer testIoDeinit();
+
+    const dir = "/tmp/malt_supervisor_follow_cancel_flush";
+    std.Io.Dir.cwd().deleteTree(test_io, dir) catch {};
+    try std.Io.Dir.createDirAbsolute(test_io, dir, .default_dir);
+    defer std.Io.Dir.cwd().deleteTree(test_io, dir) catch {};
+
+    const log_path = dir ++ "/sample.log";
+    {
+        const f = try std.Io.Dir.createFileAbsolute(test_io, log_path, .{ .truncate = true });
+        defer f.close(test_io);
+        try f.writeStreamingAll(test_io, "seed\n");
+    }
+
+    var aw: std.Io.Writer.Allocating = .init(testing.allocator);
+    defer aw.deinit();
+
+    // Append data on the first interrupt-check, then cancel the second
+    // sleep so the flushed bytes must already be in the writer when the
+    // loop bails.
+    FollowProbe.reset();
+    FollowProbe.append_path = log_path;
+    FollowProbe.append_bytes = "tail-bytes\n";
+    FollowProbe.append_at = 1;
+    FollowProbe.stop_at = 16;
+
+    const cancel_io = CancelSleepProbe.wrap(test_io, 2);
+
+    try followLog(cancel_io, testing.allocator, log_path, 0, &aw.writer, FollowProbe.cb);
+
+    try testing.expect(std.mem.indexOf(u8, aw.written(), "tail-bytes\n") != null);
+    try testing.expectEqual(@as(usize, 2), CancelSleepProbe.sleep_calls);
+}
diff --git a/src/net/client.zig b/src/net/client.zig
index 8511580..bdf4447 100644
--- a/src/net/client.zig
+++ b/src/net/client.zig
@@ -377,8 +377,13 @@ pub const HttpClient = struct {
                 if (classifyStatus(resp.status)) |dl_err| {
                     if (isTransientError(dl_err) and attempt < max_retries) {
                         resp.allocator.free(resp.body);
-                        // Sleep is backoff jitter; interruption just retries sooner.
-                        std.Io.sleep(self.io, std.Io.Duration.fromNanoseconds(@intCast(retry_delays_ms[attempt] * std.time.ns_per_ms)), .awake) catch {};
+                        // Cancellation is single-shot per task in std.Io —
+                        // swallowing it here means the caller's stop signal
+                        // is consumed by the backoff and never reaches the
+                        // next request, so propagate it as the result.
+                        std.Io.sleep(self.io, std.Io.Duration.fromNanoseconds(@intCast(retry_delays_ms[attempt] * std.time.ns_per_ms)), .awake) catch |e| switch (e) {
+                            error.Canceled => return error.Canceled,
+                        };
                         attempt += 1;
                         continue;
                     }
@@ -386,8 +391,9 @@ pub const HttpClient = struct {
                 return resp;
             } else |err| {
                 if (attempt < max_retries) {
-                    // Sleep is backoff jitter; interruption just retries sooner.
-                    std.Io.sleep(self.io, std.Io.Duration.fromNanoseconds(@intCast(retry_delays_ms[attempt] * std.time.ns_per_ms)), .awake) catch {};
+                    std.Io.sleep(self.io, std.Io.Duration.fromNanoseconds(@intCast(retry_delays_ms[attempt] * std.time.ns_per_ms)), .awake) catch |e| switch (e) {
+                        error.Canceled => return error.Canceled,
+                    };
                     attempt += 1;
                     continue;
                 }
@@ -712,3 +718,59 @@ test "shouldFireIdleWatchdog: cancellation fires regardless of elapsed time" {
     try std.testing.expect(shouldFireIdleWatchdog(0, 0, 30, 600, true));
     try std.testing.expect(shouldFireIdleWatchdog(1, 1, 999_999, 999_999, true));
 }
+
+// Patches an inner Io's vtable so `sleep` reports cancellation on the
+// configured call index; non-canceled sleeps return immediately so the
+// retry-table delays don't pad the test runtime. `cancel_at = 1` cancels
+// the first sleep, `cancel_at = N` returns from the prior N-1 sleeps and
+// trips the Nth.
+const CancelSleepProbe = struct {
+    var vtable: std.Io.VTable = undefined;
+    var sleep_calls: usize = 0;
+    var cancel_at: usize = 1;
+
+    fn wrap(inner: std.Io, cancel_at_call: usize) std.Io {
+        vtable = inner.vtable.*;
+        vtable.sleep = sleepMaybeCanceled;
+        sleep_calls = 0;
+        cancel_at = cancel_at_call;
+        return .{ .userdata = inner.userdata, .vtable = &vtable };
+    }
+
+    fn sleepMaybeCanceled(_: ?*anyopaque, _: std.Io.Timeout) std.Io.Cancelable!void {
+        sleep_calls += 1;
+        if (sleep_calls >= cancel_at) return error.Canceled;
+    }
+};
+
+test "doGetWithRetry surfaces sleep cancellation on the first backoff" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    const cancel_io = CancelSleepProbe.wrap(threaded.io(), 1);
+
+    var http = HttpClient.init(cancel_io, std.process.Environ.empty, std.testing.allocator);
+    defer http.deinit();
+
+    // 127.0.0.1:1 fast-fails with ECONNREFUSED so the retry-sleep is reached
+    // on the first attempt; without the fix, every backoff swallows Canceled
+    // and the call resolves with the connect error after burning all retries.
+    const result = http.get("http://127.0.0.1:1/nothing-listens-here");
+    try std.testing.expectError(error.Canceled, result);
+    try std.testing.expectEqual(@as(usize, 1), CancelSleepProbe.sleep_calls);
+}
+
+test "doGetWithRetry surfaces sleep cancellation on a later backoff" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    // Cancel the third sleep — the prior two return normally, exercising
+    // the retry counter alongside the catch arm so we know the propagation
+    // isn't tied to attempt 0 only.
+    const cancel_io = CancelSleepProbe.wrap(threaded.io(), 3);
+
+    var http = HttpClient.init(cancel_io, std.process.Environ.empty, std.testing.allocator);
+    defer http.deinit();
+
+    const result = http.get("http://127.0.0.1:1/nothing-listens-here");
+    try std.testing.expectError(error.Canceled, result);
+    try std.testing.expectEqual(@as(usize, 3), CancelSleepProbe.sleep_calls);
+}
diff --git a/src/ui/progress.zig b/src/ui/progress.zig
index b323ba8..ce96bc6 100644
--- a/src/ui/progress.zig
+++ b/src/ui/progress.zig
@@ -39,8 +39,14 @@ fn nowNs() i128 {
     return std.Io.Clock.real.now(pkg_io).toNanoseconds();
 }
 
-fn sleepNs(ns: u64) void {
-    std.Io.sleep(pkg_io, std.Io.Duration.fromNanoseconds(@intCast(ns)), .awake) catch {};
+/// Sleeps for `ns` nanoseconds against the package io. Returns false when
+/// a cancellation request reached the io subsystem before the sleep
+/// finished, so spin/poll callers can bail rather than swallow the signal.
+fn sleepNs(ns: u64) bool {
+    std.Io.sleep(pkg_io, std.Io.Duration.fromNanoseconds(@intCast(ns)), .awake) catch |e| switch (e) {
+        error.Canceled => return false,
+    };
+    return true;
 }
 
 /// Braille-based spinner frames, shared by ProgressBar and Spinner.
@@ -561,7 +567,7 @@ pub const Spinner = struct {
         while (!self.stop_flag.load(.acquire)) {
             self.drawFrame(frame);
             frame +%= 1;
-            sleepNs(100 * std.time.ns_per_ms);
+            if (!sleepNs(100 * std.time.ns_per_ms)) break;
         }
     }
 
@@ -672,3 +678,62 @@ test "restoreTerminal is callable without an active MultiProgress" {
     restoreTerminal();
     restoreTerminal();
 }
+
+// Patches an inner Io's vtable so `sleep` reports cancellation on the
+// configured call index; non-canceled sleeps return immediately so the
+// 100 ms spinner cadence doesn't pad the test runtime.
+const CancelSleepProbe = struct {
+    var vtable: std.Io.VTable = undefined;
+    var sleep_calls: usize = 0;
+    var cancel_at: usize = 1;
+
+    fn wrap(inner: std.Io, cancel_at_call: usize) std.Io {
+        vtable = inner.vtable.*;
+        vtable.sleep = sleepMaybeCanceled;
+        sleep_calls = 0;
+        cancel_at = cancel_at_call;
+        return .{ .userdata = inner.userdata, .vtable = &vtable };
+    }
+
+    fn sleepMaybeCanceled(_: ?*anyopaque, _: std.Io.Timeout) std.Io.Cancelable!void {
+        sleep_calls += 1;
+        if (sleep_calls >= cancel_at) return error.Canceled;
+    }
+};
+
+test "sleepNs returns true when the sleep completes normally" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    const prev_io = pkg_io;
+    pkg_io = threaded.io();
+    defer pkg_io = prev_io;
+
+    try std.testing.expect(sleepNs(0));
+}
+
+test "sleepNs returns false when sleep is cancelled" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    const prev_io = pkg_io;
+    pkg_io = CancelSleepProbe.wrap(threaded.io(), 1);
+    defer pkg_io = prev_io;
+
+    try std.testing.expect(!sleepNs(100));
+    try std.testing.expectEqual(@as(usize, 1), CancelSleepProbe.sleep_calls);
+}
+
+test "sleepNs propagates cancellation when called repeatedly" {
+    var threaded: std.Io.Threaded = .init(std.testing.allocator, .{});
+    defer threaded.deinit();
+    const prev_io = pkg_io;
+    pkg_io = CancelSleepProbe.wrap(threaded.io(), 3);
+    defer pkg_io = prev_io;
+
+    // Two normal sleeps complete and report true; the third trips Canceled,
+    // pinning that the helper isn't latched after the first non-cancelled
+    // return and matches the spin-loop's per-tick break contract.
+    try std.testing.expect(sleepNs(0));
+    try std.testing.expect(sleepNs(0));
+    try std.testing.expect(!sleepNs(0));
+    try std.testing.expectEqual(@as(usize, 3), CancelSleepProbe.sleep_calls);
+}