# Repair/save marker correction: Windows evidence

The correction and rationale are in [the preserved gate README](gates/README.md); [independent review](independent-review.md) identified the journal-recovery seam and verified the final source. Root owns the associated native workflow and integration report in [the navigation evidence topic](../). This archive preserves exact bytes and contains no executable.

Captured source: parent eeeaa1262748800c4c9aaa1eca79a7de96757358 plus the three correction files. The intervening records-only commit 00a21d3322e39f1699e30e3bdaf0ba60acc19d04 did not change build inputs. `gates/inputs1.zip` contains all 500 captured input files, including the new five-test file; `gates/inputs1.json` records their hashes. Its ZIP SHA256 is 213fdb39ea7c3601b4baac674c3ab3ecc863ac44db1b092d922d3e24f26887b7.

Actual terminal results: full desktop 755 passed / 0 failed / 1 intentionally ignored; strict all-target desktop Clippy, scoped formatting/diff checks, and matching development GUI build all exited zero. Every gate checked the same input file set and bytes. Commands, toolchain/environment, separate raw streams and result receipts are preserved under `gates/`. No compiler/test failure occurred in this gate attempt.

The matching GUI is 113,206,272 bytes, SHA256 b183ef31f16b3986538b3f2aba8435eab45978635c47f99ca01be6e72b168848, identified by the actual Cargo artifact in `gates/gui-build-result.json`. Parent independently preserved its executable for native work. `gates/final-readback.json` is the exact historical private-directory inventory and includes that excluded `gui-repair.exe`; **this archive's manifest is the archive whitelist**. Binary identity is retained in receipts, not duplicated here.

The archive verifies copied gate files and review byte-for-byte, each ZIP member against its frozen manifest, and the 500 source files against the current workspace at archive creation. Later source changes do not invalidate the historical archive. `verify_archive.py` checks archive integrity without requiring the workspace to remain unchanged. It excludes only manifest/readback self-references from the payload hash list.

No GUI/native action, WSL/Linux action, release acceptance, or new #9 timing measurement is implied. The initial GUIv3 defect evidence remains in the neighboring historical archive. The first-save rollback reload-error hypothesis was corrected before execution; the tests distinguish real missing-entry reload from an explicitly injected error.