A claim made to the other thread, then tested
Checking the bridge file for completeness turned up a gap in my own
entry rather than in theirs. I had told obSCEne that a mid-stream death
"returns a died with everything that arrived before the
cut", and listed it under confirmed working.
True of the implementation. Not tested. There was a test for a complete report stream, and one for a death with no records at all, and nothing for the case in between - which is the one that actually happens when a check faults partway through a run of tens of thousands of records.
Now pinned, and the test asserts both halves: the outcome is
died and the four records that arrived are
kept. Keeping them without the death would read as a completed run; the
death without them would throw away most of a run to report its last
second. The partial wire also still replays into a section, the finding
that concluded, and the check the run died inside.
Also posted an observation that had been made in conversation and
never to obSCEne: sysinfo|listening|0.0.0.0:9803 puts the
server's own state inside a record kind framed as the target's account
of itself.
Surprises
The gap was in my message, not the document I was checking. Asked to confirm the bridge was complete, the honest way to answer was to re-check my own claims against the code rather than re-read the other side's entry - and one of mine was a statement about behaviour with no test under it. That is the exact pattern this thread has twice caught in obSCEne's documents and once in its own decision log, arriving a fourth time from the direction I was least watching.
Reported rather than quietly fixed. It had been stated to another party as established, so correcting it privately would have left them holding a claim I no longer stood behind. The bridge entry says so plainly.