The plan was simple: stencil paste on a four-board panel, place everything by hand, run it through a modified toaster oven with a PID controller bolted on, and see how many boards came out alive. Two out of four worked on the first power-up. That's a worse number than it sounds, and better than it looks.
The failures were boring, which is good
No tombstoning, no bridging on the fine-pitch parts we were worried about. Both dead boards had the same problem: a cold joint on the same corner of the panel, closest to the oven door. The thermocouple was taped to the center board. The corner ran cooler than we measured by roughly 15°C, enough to leave that joint under-reflowed.
That's a profiling mistake, not a process mistake. The reflow profile itself (90 seconds preheat, 60 seconds soak, spike to 235°C peak) was fine for the parts we used. We just measured the wrong spot.
What changes next time
Three thermocouples, not one, placed at the corners least likely to hit temperature. A five-minute warm-up run with no board in the oven, just to confirm the oven itself has stabilized before paste ever sees heat. And smaller panels: four-up asks a $40 toaster oven to hold a tighter thermal window across a wider area than it's built for.
None of this is a reason to send early prototypes to a proper reflow house. It's a reason to be honest about what a hobby oven can and can't do consistently, and to design your test points so a cold joint shows up on the bench, not three weeks later in the field.