Two nodes at 1 Mbit/s that sampled the bit in different places
Two controllers on one 1 Mbit/s segment sampling 125 ns apart, the time-quantum arithmetic behind it, and why 87.5 per cent is a default worth matching.
serial-bench / grep
Nine messages is a small enough pile to read end to end, but when I want the one about a sample point or a divider I would sooner type than scroll. This filters the list as you type - across subjects, summaries, buses and tags - and does it entirely in your browser. With scripts off the full list is right here and nothing is hidden.
Two controllers on one 1 Mbit/s segment sampling 125 ns apart, the time-quantum arithmetic behind it, and why 87.5 per cent is a default worth matching.
Seventeen nodes, 500 kbit/s and a missing 120 Ω resistor that failed long frames while short ones kept working.
A device stretching SCL after the address byte, and a master whose 25 µs timeout aborted the read halfway.
A 12 MHz crystal against an 11.0592 MHz one, a seven per cent divider error, and why the failures depended on the byte value.
Eleven devices and 1.8 kΩ pull-ups that passed at 100 kHz and corrupted the last ACK at 400 kHz.
A link that logged errors only in the morning, and the ground-potential measurement that led to an isolated transceiver.
Three ADCs in series and a chip-select edge released 90 ns too early for the farthest device's top bits.
A mode mismatch at 8 MHz that halved every value, and why slowing the clock made the wrong answer look plausible.
The first entry, and a confession: 24 MS/s against an 8 MHz clock, and four hours spent chasing a sampling artefact.