Tested source: 43340b94a13f192c3b5d3c3112dbe87246790a44, Windows GNU/LLVM Rust/Cargo1.98.1. The implementation report defines acceptance and coverage limits. This package is branch-specific verification evidence, not a runnable release or an integrated acceptance claim.

Source/adapter review, including final boxed-snapshot delta and reviewer independence limits.
Native observations, including unsuccessful actions and recovery.
Native evidence review, independently inspected original images and exact bytes, with planner/fixture self-review limits.
Final report audit.
Captured checks: eleven closed attempts, three retained failures, eleven exact823-input source proofs, four deduplicated source capsules.
Native archive and standalone verifier: two closed native runs and current CLI preflight, original captures, helpers, source/oracle bytes, lifecycle and immutable readback receipts.

Run standalone verifiers using Python3.11+; they need no checkout, Cargo, Git or network and do not execute the product or archived helpers:

python -B checks/verify.py
python -B native/verify.py native

checks/evidence.zip: 13,144,614 bytes, SHA256 cc4703a21853988d5b833a817a9f6331a98fc4e3d117e74a8f85733ad57d4cc0. The checks verifier validates the logical-file map, nested captured-source archives, original result/stream/proof relationships and all recorded exits. Its PASS is consistency, not success of the three failed attempts. Exact Git-source comparison was separately performed when each gate closed; offline validation does not recreate Git trust.

native/native-evidence.zip: 6,942,333 bytes, SHA256 9eb28f970d256160238234279fa91ad60e404e8b11d8d7e6d55178999919e2b1, 305 members. Run1 contains 62 capture records and five exact snapshots; run2 contains 15 capture records and its exact reopened snapshot. Both runs and the current CLI preflight are explicitly accepted only within the report's scope.

The parent ran both standalone verifiers successfully. Separate direct audits compared all88 original checks-run files and all304 original native members byte for byte; the generated native configuration matched semantically. All six omitted GUI/CLI/runtime files were independently rehashed against their pins while still locally available. Original failed checks and native input/capture recovery attempts retain their actual outcomes.

The first staged-byte check detected Git newline conversion in the native manifest. A scoped .gitattributes rule now preserves this evidence directory verbatim; the manifest and archived originals were not normalized or rewritten.

The native archive excludes executable/DLL bytes, retaining their explicit build/copy/readback pins. Offline validation cannot rehash omitted binaries or independently infer which keys were sent. The parent action record, original captures, exact file states, lifecycle records and independent review jointly support the reported bounded workflows. Native Shift-click, focused Delete success, keyboard-only confirmation, broad accessibility and real App source-write failure are not established. Linux remains user-paused.