2026-08-19 (late) - The loop, made runnable

The development cycle existed only in my head. Now it is ./orbistoun.sh sweep and WORKFLOW.md. D077-D078; 24 crates, 273 tests.

Surprises

• The call trace was volatile and nobody had noticed. Stderr only, dying with the process, after runs that take ten minutes. It contradicted a principle the project states explicitly, and it survived because every individual run looked fine.
• Three of four modules recorded nothing, because a faulting guest died before the trace was written. The fault path was the one that most needed it.
• The worker had no paths configured at all, so nothing would have been persisted anywhere regardless of which path wrote it. It is the only process that ever sees a trace.
• Databases were written with an empty suffix_hex whenever the shipped default was used, which makes them unloadable. Every trace printed hashes for names we had already found, and it read as the search having failed rather than the file being unusable.
• Documentation drifted in both directions at once: a verb that existed and was not in the help, and a command in the help that did not exist.

Outstanding

sceKernelDirectMemoryQuery is 99.9% of all guest calls across four executables. Implementing it is the next piece of work, and it is the first time the project has had a next piece of work chosen by measurement rather than by roadmap order.