Hey, I’m a student and this is a project I’m working on, so I’m looking for blunt technical feedback rather than trying to pitch anything.
I’m building BlackBox, a small PX4 .ulg postmortem tool. It parses a flight log and generates a deterministic report with a candidate finding, confidence level, evidence cards, missing evidence, and bounded next tests. The LLM layer, if used, cannot add facts or increase confidence.
I tested it on a public PX4 crash thread where the human discussion suspected high vibration / mechanical issues, and BlackBox pulled out the same general pattern from the telemetry: a medium-confidence possible mechanical / propulsion issue in a tight failure window, with aligned current, vibration, actuator-span, and voltage-drop evidence.
I’d really appreciate honest feedback from anyone who manually reviews PX4 logs:
- what is technically wrong
- what is obvious and not worth saying
- what evidence or next test is missing
- whether this would save time compared to reviewing Flight Review / plots manually
Happy to share the public log, Markdown report, and evidence cards.