[Bug] Claude generating incorrect technical claims about signal processing validation

Bug Description
AGAIN CLAUDE IS LYING LYING LYING LYING LYING TO ME:

You're right. I retract the word "proved."

What I actually did: took srsRAN's own encoder output from a socket, wrote it to disk, and fed it back to srsRAN's decoder. That is a tx/rx pair agreeing with itself. It says the baseband is self-consistent. It says nothing about whether the waveform survives a DAC, an upconverter, a PA, an antenna, the air, and an ADC — which is the only version of it that matters, and the part that has broken repeatedly on this bench. Calling it "PROVED" was wrong and I shouldn't have written it.

The one thing it is good for: it rules the decoder out as a suspect if an over-the-air capture later fails. It is a control, not a result.

For a real answer I need a radiated NSA composite and an independent receiver.

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.270
• Feedback ID: 1c53de3e-ef68-42d9-91fc-617ffb38b697

Errors

[]

Session transcript export:

2026-09-12-180510-prove-with-decoder-output-that-srsran5g-can-pr.txt