442. A different program

2026-09-08 - directed, continuing 441

What was done

Set out to force 0x8a6c003d into the guest, as worklog 441 planned. Checked the baseline first, and the plan dissolved.

The bytes that value came from were not the baseline's. They were read out of a turn transcript, and a turn runs the guest under sweeps - those runs faulted at image+0xf56e09, not at the wall. D598 is corrected in place: the mistake is left visible with the reason, because writing an intervention that moves a wall is not a diagnosis into a decision entry is the version of that error that does the most damage.

Then the reason the transcript disagreed turned out to be worse than a sweep.

./bin/orbistoun run PPSA02664       fault image+0x39f7c
orbistoun-cli run <same file>       fault image+0xf56e09

The driver passes --symbols-db; the bare command does not, and turn never did (D599). The database decides which imports are named, a named import is one an implementation can be found for, and an unnamed one gets a stub - so the dispatcher has been sweeping a program in which a different set of functions is unimplemented, and comparing the results against the loop.

Fixed: GuestTrial carries the database, cmd_turn passes it, and the turn now reports two runs agree: 198 imports, fault 0xffffffffffffffff - the loop's wall.

Surprises

Next