phase12: checkpoint follow-up matching negatives

This commit is contained in:
Christopher Williams
2026-09-25 08:59:25 -04:00
parent 2ced412d87
commit e7450aafe0
+12
View File
@@ -2401,3 +2401,15 @@ supersedes that count.
These are staging-only experiments; no tracked source or registry row was added for any bounded
attempt. The authoritative count remains **717 bodies / 726 regions**.
### Follow-up bounded attempts — 2026-09-25
* `0x80070454`: corrected the derived template address to `0x80136964`; the candidate still folds the
address into the wrong register and is 68 bytes against 72.
* `0x800247A4`: aggregate assignment lowers to a loop and manual nine-word statements expand to 208
bytes; neither reproduces the original unrolled 148-byte copy.
* `0x8001AAA8`: the guard logic is identified, but the original uses `lui`+`and` masks for `0x40` and
`0x200`; C folds them to `andi` and the candidate is 120 bytes against 124. Address-valued mask
temporaries changed the load order and did not transfer.
No new source, registry row, or milestone count resulted from these follow-ups.