[mstan]

MMX's reconstructed backgrounds resolve scene coordinates, tiles, private CHR and palettes for each output pixel. Bind these resources once per source span, ending spans at tile, native-view, resource-phase and scene-mapping boundaries. Preserve the independent scalar mapper for reference and imported menus/Rush ceiling, and the existing mosaic path.

MMX_NATIVE_BG_SPANS defaults OFF pending the owner's bundled playtest approval. The candidate is title-owned scene/asset policy using the existing shared row planner. No new persistent cache, serialized field or netplay protocol change is introduced. Benchmark/replay identity records the selection.

Measured on the i5-4670 NAS using matched 4,200-frame Highway workloads with audio emulation active and dummy output:

Profile	Median process-CPU reduction	Evidence
16:9 / 342px	12.8%	Five balanced OFF/ON pairs, all positive (10.2–13.7%).
32:9 / 682px	34.7%	Five balanced OFF/ON pairs, all positive (33.0–35.5%).
Custom 4:3 / 256px	Neutral/inconclusive	Five mixed pairs, median 0%.
Stock	Neutral/inconclusive	Initial and repeated five-pair batches, A/A controls and canonical-main controls retained; no reproducible regression.

Primary OFF/ON binaries use identical clean source 7e55826, effective engine 147b55e, and UI 7c635c6. A separate unmodified current-main e5d1308 comparison confirmed 13.1% at 16:9 and 34.0% at 32:9 (one pair each), with five stock pairs remaining mixed/neutral. Those confirmation runs are labelled separately. Background CPU was present; raw resource observations, ordering and repeated controls are retained. These are process-CPU results, not Xbox or FPS predictions, and must not be added to earlier package percentages.

Validation passes: native scenes; 1,536 scalar-compared full-width background rows; non-aligned CHR/palette phase checks; 512 randomized composition rows; Windows/Linux fixtures; eight captured view/HUD comparisons; all four ROM-backed Magic Carpet cases; current-build save/load/rewind/replay and fresh-process replay for both arms. A new real-ROM rollback test varies rendering width/count across resimulation and verifies identical rollback state. Paired rendered images and available full custom captures match. CI passes at review head ef821a3; runtime source is unchanged from measured 7e55826.

Relative to canonical current main, adoption adds 8,064 text bytes and 64 read-only-data bytes; data/BSS are unchanged and there is no new persistent cache. Same-source selector-only text cost is 6,464 bytes.

Tracking: beads-8wg.1.150, roadmap beads-49cy, campaign beads-2wy8.1. Machine-readable samples, identities, configurations, images and gate logs: G:/Temp/Codex/snesrecomp/background-spans-20261027/campaign-results.json and evidence/. Final NAS export: 771 files verified by CRC and individual SHA-256. No original-Xbox or cross-build netplay qualification is claimed. Owner playtest/default promotion and merge remain pending.