@GizanTech4IR Seen the same pattern. The fixture fixes the 11 min, but the dev-cert unit usually ships because nothing records which cert went on which serial. Logging the flash step per unit (serial, image hash, cert ID, pass/fail) is what closes that. We built TofuPilot partly for this.
@mr_ehsane The USB power comparison in your write-up is interesting. Did you keep receiver gain and bandwidth fixed between the laptop and power-bank runs? Saving those settings with each spectrum would make the noise difference easier to compare.
@AstroForge Can you replay the same injected fault against a new firmware or model revision while holding the bench’s initial state fixed? I’m curious which state you preserve alongside telemetry to make those runs comparable.
@ExodusBecause How do you distinguish a PLC restart from a legitimate write that clears the same bit? It would be interesting to see the simulator case and the real-controller trace side by side.
@lukebayes@0xIanCleary@Friis25@DigilentInc Do your Digilent scripts mostly support interactive debug, or also repeatable pass/fail checks? Curious how you keep capture settings and firmware versions with the traces when a bench bug becomes a regression test.
@VectorCrossProd How will TestEE distinguish a design that passes the tool checks from one that meets the electrical requirements? Will acceptance tests stay fixed across model runs, with generated artifacts and tool versions kept alongside each score?
@_RyanHwang Your point about breakpoints invalidating timing measurements is useful. Do your reports link each captured trace to the board revision, flashed binary and acceptance limits? That seems important when an agent reruns a test after changing code.
@_asadmemon For Batch 1, which checks decide whether a camera is ready to ship? I’m curious how you keep each unit’s calibration output and QC measurements linked to its serial number and firmware version.
@emotor For repeatable captures, I’d keep instrument ID, firmware and acquisition settings beside each waveform. Can a successful MCP session be exported as a script to rerun without an LLM? That would help turn an exploratory session into a repeatable test.