PXT direct RGB444 fragments — actual host capture
Tested source: c1d03878589b173c51c8d72fb1e7e33719dd6774,
submitted in PR169.
The actual capture run37647759023
and job112882868540
passed. This records the focused host check, not completion of all PR gates or
an already landed merge.
The original artifact11494779371
reported a 2,447-byte ZIP. Its byte-preserved
JSON member has SHA256
82803444d5a6079388de7f924a451d5781b47abd1eb6ae4a95dc8952e2220a04.
The companion exact Microsoft MIT license has SHA256
dea9265341829002e2c23a7372393eb2ed6e26085fb623f38a4ba0af833f30a6;
the notice and immutable original link remain in the
harness guide.
Exact SHA-bound original method and board-ID predicate fragments compiled with Ubuntu g++13.3.0 for 32-bit host x86, inside the documented authored class/palette/buffer/transport boundary. Ten admission controls passed. Thirteen board-ID cases (nine accepted, four rejected) and the missing-SPI negative passed. All emitted bytes, CS release and transfer sizes matched:
| Authored input dimensions | RAMWR data bytes | Transfer sizes |
|---|---|---|
| 4×2 | 12 | 12 |
| 3×2 | 12 | 12 |
| 160×5 | 1,200 | 480,480,240 |
| 159×4 | 960 | 720,240 |
The odd 3×2 case emits eight slots for six declared pixels. Odd-width padding
and original batching are preserved, not normalized. A named corrupt-first-data
capture was rejected with RAMWR mismatch. A separate read-only artifact audit
checked all four literal byte sequences, source buffers, dimensions, transfer
extents, predicate results and license hash without importing source helpers.
This is not execution of full WDisplay, actual Image storage, its constructor, updateScreen branch dispatch, ARM/SAM/DMA/IRQ, deployed macros/CF2, module GM/aperture/LUT, physical timing, RTx, browser-WASM or an installed app. The real Image four-byte-aligned column stride remains a separate compatibility gate. This capture does not justify inventing an RGB444 identity LUT or closing P3/CP14.