diff --git a/extra/sqtt/test_timing.py b/extra/sqtt/test_timing.py
index f0f51516c3..898f4de91d 100644
--- a/extra/sqtt/test_timing.py
+++ b/extra/sqtt/test_timing.py
@@ -14,7 +14,7 @@ from tinygrad.uop.ops import UOp, Ops, KernelInfo, AddrSpace
 from tinygrad.engine.realize import CompiledRunner
 from tinygrad.device import Device, ProfileDeviceEvent
 
-from extra.sqtt.roc import decode, InstExec, PrgExec
+from extra.sqtt.roc import decode, WaveExec
 
 dev = Device[os.environ["DEV"]]
 
@@ -36,7 +36,7 @@ def asm_kernel(instrs:list[str], l:int=1, g:int=1) -> Tensor:
 def save_sqtt():
   # clear the old traces
   dev.profile_events.clear()
-  sqtt:dict[PrgExec, list[InstExec]] = {}
+  sqtt:dict[str, list[WaveExec]] = {}
   yield sqtt
   # decode sqtt
   if os.environ["DEV"] == "AMD":
diff --git a/tinygrad/viz/serve.py b/tinygrad/viz/serve.py
index 8a4207f3fa..d30496f349 100755
--- a/tinygrad/viz/serve.py
+++ b/tinygrad/viz/serve.py
@@ -215,9 +215,9 @@ def load_sqtt(profile:list[ProfileEvent]) -> None:
   if not rctx.inst_execs: return err("EMPTY SQTT OUTPUT", f"{len(sqtt_events)} SQTT events recorded, none got decoded")
   steps:list[dict] = []
   for name,waves in rctx.inst_execs.items():
-    if (r:=ref_map.get(name)): name = ctxs[r]["name"]
-    steps.append({"name":name, "depth":0, "query":f"/render?ctx={len(ctxs)}&step={len(steps)}&fmt=counters",
-                  "data":{"src":trace.keys[r].ret.src if r else name, "lang":"cpp"}})
+    prg = trace.keys[r].ret if (r:=ref_map.get(name)) else None
+    steps.append({"name":prg.name if prg is not None else name, "query":f"/render?ctx={len(ctxs)}&step={len(steps)}&fmt=counters",
+                  "depth":0, "data":{"src":prg.src if prg is not None else name, "lang":"cpp"}})
 
     # Idle:     The total time gap between the completion of previous instruction and the beginning of the current instruction.
     #           The idle time can be caused by: