diff --git a/AGENTS.md b/AGENTS.md
index 9b930f1..dc69f45 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -790,6 +790,24 @@ Next instrument, and it needs a one-line model change rather than more guessing:
 100 ms, which is the same order as the app's whole life. Dropping that interval to a couple of milliseconds
 turns it into a real trace and would show where in the overlay the app stops.
 
+**The PC probe's interval is now a knob (UVK5_PC_PROBE_MS), and at 2 ms the overlay app's whole life is one tick.**
+
+100 ms was the same order as the thing being measured: an app that the launcher starts and that is gone again
+before the next sample. uvk5_pc_probe_interval_ms() reads UVK5_PC_PROBE_MS and falls back to 100, so every
+existing probe run is unchanged. At 2 ms, a launch gives 468 samples in about 0.94 s and **exactly one of them
+is inside the 4 KiB overlay** -- the app runs for roughly 2 ms, not 100.
+
+With a real trace in hand, two things are now settled and one is new. Settled: the loader copies and jumps (the
+single overlay sample), and a 16-byte app that calls nothing stays there forever. Settled the other way: the
+sector-cache overwrite is dead -- the flash log's **last** transaction for the run is the app's own code load
+(addr=103000 len=2408) and **nothing follows it**, no font read included. New: after the app dies the firmware
+sits in a loop inside its own flash (PCs around 0x08005118/0x08005122/0x08005248) with **zero flash reads and
+no key response at all** -- even F then 7 no longer opens the menu, and an explicit MENU down/up changes
+nothing. So the launcher (or the fault path it takes) is wedged, not merely waiting for a release.
+
+That is where the next round starts: the app dies within about 2 ms, the firmware never reads the flash after
+loading it, and what is left running is a tight loop in the firmware with no display or key activity.
+
 ## The keypad: two real bugs, both fixed
 
 The old note here said "keys reach the firmware but the UI does not react" and
diff --git a/AGENTS.zh-CN.md b/AGENTS.zh-CN.md
index 50a1cae..2060bfa 100644
--- a/AGENTS.zh-CN.md
+++ b/AGENTS.zh-CN.md
@@ -642,6 +642,22 @@ flash 探针是按顺序记录每一笔事务的，而那一次启动的**最后
 下一个仪器，只需要改模型一行而不是继续猜：PC 探针每 100 ms 采样一次，而应用整个生命期就是这个量级。把这个
 间隔降到几毫秒，它就变成真正的 trace，能看出应用**停在叠加区的哪个位置**。
 
+**PC 探针的间隔现在是可配置的（UVK5_PC_PROBE_MS）；调到 2 ms 之后，叠加应用整个生命期只有一个采样点。**
+
+100 ms 与被测对象同量级：启动器刚把应用带起来，下一个采样到来之前它就已经没了。uvk5_pc_probe_interval_ms()
+读 UVK5_PC_PROBE_MS、缺省仍是 100，所以以往任何一次探针运行都保持原样。调到 2 ms 后，一次启动在约 0.94 秒里
+得到 468 个样本，其中**恰好一个**落在那 4 KiB 叠加区内 —— 应用只活了约 **2 毫秒**，不是 100。
+
+有了真正的 trace，两件事就此确定，另有一件是新发现。确定的：加载器会拷贝并跳转（那一个叠加区样本），而且
+那个 16 字节、什么都不调用的应用能一直待在里面。反过来确定的：**扇区缓存覆盖说已经死了** —— 那次运行的 flash
+日志里**最后一条**就是应用自己的代码加载（addr=103000 len=2408），**其后什么都没有**，包括字体读取。新发现：
+应用死掉之后，固件**停在自身 flash 内的一个循环里**（PC 在 0x08005118/0x08005122/0x08005248 一带），**零次 flash
+读取、按键完全无响应** —— 连 F 再 7 都打不开菜单了，显式地 MENU down/up 也毫无变化。也就是说**启动器（或它走的
+故障路径）卡死了**，而不只是在等一次松开。
+
+下一轮就从这里开始：应用在约 2 ms 内死亡，固件在加载它之后**再也没有读过 flash**，而剩下的是一个没有显示、
+没有按键活动的紧循环。
+
 ## 键盘：两个真 bug，都已修复
 
 这里原来的笔记写的是"按键到达了固件但界面不反应"，并且归咎于机器模型。结果发现有**两个
diff --git a/qemu/py32f071.c b/qemu/py32f071.c
index 3da31c6..f4277ae 100644
--- a/qemu/py32f071.c
+++ b/qemu/py32f071.c
@@ -3609,6 +3609,19 @@ static uint32_t uvk5_sniff_app_offset(const char *path)
  * can be located without a debugger (a gdb attach stops the guest, which is the
  * one thing that must not happen while working out where it stopped).
  */
+/*
+ * How often the PC probe samples. 100 ms is fine for finding a hang, but an overlay
+ * app's whole life can be shorter than that -- the launcher takes the screen and the app
+ * is gone again before the next tick -- so UVK5_PC_PROBE_MS lets a run ask for a real
+ * trace instead of a few points. The default keeps every existing probe run identical.
+ */
+static int uvk5_pc_probe_interval_ms(void)
+{
+    const char *v = g_getenv("UVK5_PC_PROBE_MS");
+    int ms = v ? atoi(v) : 0;
+    return ms > 0 ? ms : 100;
+}
+
 static void uvk5_pc_probe_tick(void *opaque)
 {
     UVK5MachineState *s = opaque;
@@ -3620,7 +3633,7 @@ static void uvk5_pc_probe_tick(void *opaque)
             fclose(f);
         }
     }
-    timer_mod(s->pc_probe_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 100);
+    timer_mod(s->pc_probe_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + uvk5_pc_probe_interval_ms());
 }
 
 static void uvk5_boot_key_release(void *opaque)
@@ -3800,7 +3813,7 @@ static void uvk5_machine_init(MachineState *machine)
     uvk5_arm_boot_key(s);
     if (g_getenv("UVK5_PC_PROBE")) {
         s->pc_probe_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, uvk5_pc_probe_tick, s);
-        timer_mod(s->pc_probe_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 100);
+        timer_mod(s->pc_probe_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + uvk5_pc_probe_interval_ms());
     }
     qdev_connect_gpio_out_named(DEVICE(&s->soc.gpio[0]), "pin-out", 3,
                                 qdev_get_gpio_in_named(DEVICE(&s->flash),