sorokin.oleg.edu
2df7f580aa
Replace the oversized header and reference rail with stable editor slots and a retained supporting drawer. Bind diagnostic, result and menu actions to their original publication and visit; preserve effective selections on navigation and dismiss Actions, Find and the drawer one surface per Escape. Retain exact Windows source captures, eight final gates (871 desktop passes), independent reviews, native selection/rename/recovery trials and the original failures. All 526 tested inputs and staged evidence were byte-checked before commit. Combined preview integration and packaging remain with issue 27; Linux remains paused.
2.8 KiB
Selection correction report: bounded evidence review
PASS — no factual mismatch found. Reviewed the versioned report and historical Find-only red/green evidence. No retained helpers, builds, tests or GUI actions were executed; no source was edited.
The manifest has exactly 25 payloads totaling 9,173,324 bytes. Every listed payload's actual size and SHA-256 matches. Four input maps each contain 525 files: red1→red2 changes only app/text_workspace/focus_selection_tests.rs; red2→green2 changes only app/source_symbols.rs; green1→green2 changes no input bytes. The retained source hashes and green stdout hash match the report.
Raw red1 stderr identifies the same before-Find child-index panic in all four tests; it is correctly rejected as regression proof. Red2 compiles and has two passing controls and two failures at selected-byte assertions (Some("") versus Some("oa_contact")). Its timestamps, duration and unchanged-input receipt match the report. Green1 has no compile line, reproduces those failures and is explicitly rejected. Its ZIP source entry retains the old 20:04:46 local modification time; green2's source entry is 20:16:32 with identical source bytes. Green2 stderr records fresh compilation and stdout records four passes; 03:16:36.205020Z–03:17:54.177014Z and 77.9678s support the stated rounded values. These observations support the reported cache-retry interpretation; they are not an independent rerun of Cargo's cache behavior.
The report properly separates this executed Find-only correction from later departure changes and their pending combined/native evidence. The distinction between a nonmutating live focus restoration and a canonical range retained on a committed departure matches the independent static review. CPU/App Content separation, injected Escape bridge message, lack of GPU/window/scroll proof, and lost Word/Line extension mode after real navigation are stated rather than silently accepted.
This review checked manifest payload hashes, input-map differences, raw receipts/logs and ZIP entry metadata. It did not execute archive builders or independently revalidate every decompressed source member. Historical inputs remain distinct from the currently running combined gate.
Bindings
- Report
C:/ar-text-9e8r6t/harness/text-selection-correction-20260913.md: SHA-256adfd9b5f11cf3ae6c65579f62ec90e529dd8dd569c4fc452e41b5b2994d21e61. - Experiment manifest: SHA-256
4580d8cbc2f0347f1ebc1448728af6ef8885f6ebbd4c2d2dbe0c41eac97fb283(4,401 bytes), binding all 25 checked payloads. - Green2 stdout: SHA-256
6f19f24e2fa37db4fc95c9b2d11b65e420299c6c6f90e4657e7e8e328b326fa0. - The separate durable departure static review is
departure-selection-review.md, SHA-256f872c511eab74ca05d35b97ad31ee1a879c42717e40a698959501b6f96546dac; it retains the five final formatted source hashes and the corrected fixture finding.