# Protected source destination diagnostic evidence (#46) Source: `0e83157e09d52e6152df91208e41693b62da2d46`, based on published `ea23e229260d7d050b3c3b88e77c3a851e936abe`. This packet was assembled locally before integration/publication. See [the contribution note](../../save-as-diagnostics-20261001.md). The packet preserves the original bytes of the first failed candidate diff, source hashes and E0308 log; corrected targeted tests; strict Clippy and GUI build logs; preparation/freeze/launch/oracle scripts; native plan, 16 captures, eight full byte-oracle receipts, baseline, launch/exit receipts and empty streams; full parser request/response envelopes; and final independent review. Donor build/extraction receipts establish the reused parser server's separate provenance. `commands.json` records parent-observed invocation scope; flags are not recoverable from build stdout logs alone. `source-files.json` identifies relevant Git source bytes. The desktop hash is `e9cce9355525ecdc59bfae12608b1e0272167acef5a86b25368ef547e08886ff`; runtime hash is `f45604a12e8bcd839fd86f667c1be1cc7c4f9da2b3c69ff381ac708f0e1ae7d4`. The reused server hash is `6cebacc117e46fdab023c6402a5cebd6929fb55993ae8ff6b89cd1c6c1170237`, from `715036e132677dd2bc2fb37ed316d93dc9987aec`; it was not rebuilt for #46. Excluded: the 119 MB desktop binary, runtime copies, live fixture directory copies, parser store, redundant failed log, unrelated package evidence and release assets. Complete source inputs survive in request envelopes; seven companion files are reconstructible from the pinned source and checked against all phase receipts. The independent review checked the live fixture/store files before exclusion; this packet's offline checker does not claim to repeat those live-file observations. From a repository checkout containing the pinned Git commit, run Python 3: ```powershell python harness/experiments/save-as-diagnostics-20261001/verify.py # After staging, additionally detect any Git newline conversion: python harness/experiments/save-as-diagnostics-20261001/verify.py --index ``` `evidence-files.json` inventories every packet file except itself. Scoped `.gitattributes` disables newline conversion for all descendants. The checker is read-only: it checks hashes, exact filename sets, all eight declared phase byte sets, full parser source/model envelopes and native exit receipts. It does not launch the application, rerun compilation, or re-execute a native test. Historical scripts retain original local paths and are evidence, not portable one-command replay. `prepare-native46.py` asserts the frozen source and copies the built executable; `launch-native46.py` restricts PATH and captures exit/logs. `oracle.py inject` and `restore` deliberately mutate the isolated B fixture. Do not run these scripts against an existing trial or interpret elapsed historical paths as available artifacts. For a new trial, build the pinned source in the configured Windows GNU LLVM environment and adapt paths into a fresh isolated directory, following the native plan and recording new receipts. Scope: Windows development build and bounded #46 workflows. Native desktop exit was normal (0), streams empty; owned parser termination was forced (1), not graceful-shutdown evidence. Persistent collision footer truncation remains, with pointer tooltip readable at normal/minimum sizes; keyboard tooltip access was not tested. Fresh destination companions were seven parent copies; 30 source files were application writes. No full-suite, new-platform, packaged #46 release or complete OA/SA acceptance is claimed. The earlier #39 release retains its old diagnostic wording; integration/publication is a separate parent-owned action.