443. Re-derived

2026-09-08 - directed, continuing 442

What was done

Made the correct invocation the easy one. orbistoun-cli turn needs --symbols-db to measure the program the loop measures, and a flag somebody has to remember is what cost a session's verdicts (D599). ./bin/orbistoun turn <title> now resolves the title and passes the database exactly as run does. Fixing the tool made it capable of being right; this makes being right the default path.

Re-derived PPSA03416's turn against the right symbol set (D600):

• The starred libc::memcpy answered the code the guest followed finding is gone. It was the headline of D583 - the thing CheckRepeats caught as drift - and it does not occur when the guest gets the symbols the loop gives it. The check was right that it was noise and wrong about the kind: a configuration artefact, not run-to-run variation. D583's conclusion stands.
• two runs agree: 193 imports, fault 0xa0, where the same turn previously reported a disagreement.

The determinism work landed, on one signal

imports, five runs	193 every time
the same, four runs two iterations ago	192, 192, 193, 192
mapping sequence, three runs	1 to 2 of 47 identical

So the logical clock, the single region for guest-visible handles, and the configuration fix together made the coarse signal stable - and the mapping sequence is still not.

CheckRepeats compares where the guest died and how far it got, and both are stable now. D583 wrote this case down before it happened: a run can repeat on those two and vary in something no step reads. The anticipated case arrived, which is worth recording, because a caveat nobody revisits is indistinguishable from one nobody needed.

Surprises

• The bogus finding was worse than "noise". It was a stable, reproducible artefact of a configuration nobody knew was different - so re-running it would have reproduced it every time and confirmed it.
• The mapping sequence diverges at the first mapping, not at the thread creation D582 identified around mapping seventeen. Something earlier moved and nothing here says what.

Next

• Re-derive the rest of this session's turn conclusions. One down; the axis sweeps, the placeholder hunts and the read-structure results are each a boot away.
• Why the mapping sequence diverges immediately.
• The clean title's wall, which is platform work with no shim in it.