2.6 KiB
Actual Process render feedback byte proof
This small package holds the four actual WGPU RGBA readbacks, adjacent JSON
metadata, lossless PNGs, original png-readback.json, encode-png.py and parent
observations.md from .local/process-feedback-review/capture-1. It requires
Python and Pillow; it never invokes Cargo, a GUI, or archived capture helpers.
Use a portable Process checks package containing the closed feedback-1 gate:
python -B prepare.py --root C:/ar-multiselect-9e8r6t --checks CHECKS_PACKAGE --check-only
python -B prepare.py --root C:/ar-multiselect-9e8r6t --checks CHECKS_PACKAGE --output FRESH_DIRECTORY
python -B FRESH_DIRECTORY/verify.py --checks CHECKS_PACKAGE
Run the checks package's own verifier too. Feedback references exact logical
feedback-1 receipt/helper/log/input pins, so the checks package may include other
runs and may be moved independently. No full source capsule is duplicated here.
Preparation refuses existing output and compares all 742 captured input bytes
to Git commit b15345123a. Verification checks the
fixed source map identity and every source-capsule member through the supplied
checks package. The original gate HEAD predates b153; its captured inputs, rather
than that HEAD, identify the rendered source. The relevant captured test and
Mounted renderer paths have explicit source pins.
Every PNG is decoded and required to equal its raw RGBA byte for byte at the recorded 1440x900 size. For OA and SA separately, before-press and after-motion must preserve complete source bundle, selection string/count, canvas, camera, scale and dimensions while pixels differ. The original PNG readback receipt, unique safe ZIP inventory and every archived file hash are checked. A matching checks receipt must record the exact passing test and capture output directory.
This is byte/provenance consistency, not native acceptance or independent image interpretation. Pixel files have no gate-time hashes: their linkage is the recorded output directory, create-only captured test code, passing test receipt and parent observations; package pins establish bytes at packaging time. Selection metadata stores the test's captured snapshot; unchanged current selection and empty history are separately asserted in the passing test. There is no history metadata, and this tool does not invent any. Source text in metadata is compared exactly between stages, not reparsed as a Project. The parent observations describe visible quality and scope limits. Encoding helpers and observations are preserved, not executed or treated as instructions.