#1914 bench: community report, RTX 5080 Laptop 16 GB, IQ3_XXS, engine 0.1.42

Description

Results-only submission (no engine changes): `bench/results/2026-10-11-community-rtx-5080-laptop/`.

- Hardware: RTX 5080 Laptop GPU 16 GB (175 W limit per NVML), Ryzen 9 9955HX3D 16C/32T, 64 GB RAM, NVMe SSD, Windows 11, driver 582.05, AC power, desktop apps open.
- Software/model: Strata `61b3fb5`, engine 0.1.42 ready-made Windows release; Flash-Next IQ3_XXS, 131,072 context, `--kv int8 --kv-resident 32768`, `--spec 4`, MTP, expert cache and prefill auto.
- Sweep: 3 fresh runs each (no reused prefix, 256-token cap, streaming): ~3.5K prompt: prompt 2,205 tok/s, decode 116.3 tok/s, TTFT 1.62 s; ~28K: 2,668 / 107.1 / 10.49 s; ~109K: 2,604 / 103.4 / 41.89 s (medians, ranges in the README).
- Needles: 6/6 found at 32k and 128k (depths 10/50/90).
- Extra: a short structured-output (json_schema) workload note comparing engine 0.1.41 and 0.1.42.
- Limitations: actual prompt lengths are below the 4K/32K/128K targets (counts in the table); single engine session on a laptop, no clock or temperature capture.

Per-run JSON and the measurement script are included.