[rsasaki0109]

Camera confidence-mask resizing recalculates the same horizontal nearest-neighbor index for every row. Compute the exact columns once per call, and use x directly for equal widths. Keep full confidence validation, strict >0.5 threshold, independent transferable output, worker protocol, model, explicit readbacks and Rust tracking. Temporary indices need at most 640 logical bytes with no retained cache.

Validation passes 92 JS controls on both parent and candidate, including all 102,400 horizontal size pairs, single-pixel supports, Float32 boundaries, unsampled invalid values and transfer ownership. All six asset and 86 actual WASM/model browser controls pass, including 42 complete browser mask comparisons.

Retain the original four-process timing reversals. Two prospective in-process interleaved surveys show lower medians in 17 of 20 pairs, with all three higher pairs and every raw sample published. The common camera sizes shorten by about 10-23% in these Node observations; physical-phone, GPU, AI and end-to-end performance remain unqualified.

The source-bound receipt preserves all 11 terminal guarded commands and a diagnostic prelaunch guard rejection. Reuse the current tested WASM and existing vendor hardlinks under a 128MiB owned-root cap, with no Rust compilation, cache growth or new data/model acquisition. See docs/CAMERA_MASK_RESIZE.md and docs/receipts/2026-10-22_camera_mask_resize.json.