2026-08-30 - The work list could not see a system call
Having got elfldr as far as making real syscalls, the
obvious question was which one to implement next.
orbistoun-cli worklist is the list this project works from
- 358 entries ranked by call volume across sixteen runs - and syscall
649, the thing that actually stops the payload, was not on it. Nor could
it have been: the list totals imports, and a guest
reaching the kernel by number touches no stub.
The fact was being produced and thrown away, printed to stderr at the end of a run. Third time that exact shape has turned up: the sysctl names before D397, the unanswered paths before D387, and now this.
Recorded in the trace beside the imports rather than among them, and ranked by how many runs asked rather than by how often - because the recorder is a bitmap and a call count would be a number nobody measured.
The bit that would have gone wrong quietly
The traces directory is not wiped between versions, and
cmd_worklist skips a file it cannot parse with a
note rather than a failure. A new field without a serde default
would therefore have turned every existing trace into a skipped file and
left the work list silently counting only runs from today - still
printing a confident ranked table. Tested, and the test says why.