Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
# Phase A verification matrix -- 2026-10-15T05:19:20Z
# repo: 0a8bf4c235dfe4b5ad88ed7cd02a8cb82cf87976 branch: strata/rebuild-2026-10
# host: Apple M4, 16 GiB unified memory, macOS 26.5.2, ~21 GiB free
CHECK RESULT EVIDENCE
----------------------------------------------------------------------------------------------
Engine build (Metal, from source) PASS engine-build.txt
Metal kernel parity (real GPU) 17/17 + 1 known metal-parity.md
moe_mm_parity (real defect, open) FAIL metal-parity/moe_mm_parity.log
Serve suite 801 PASS (11 skip) serve-suite.txt
Terminal suite (incl. 15 transport) 237 PASS owner-transport-suite.txt
Time-skip suite 24 PASS timeskip-suite.txt
Tuning suite (job/measure/sweep) 63 PASS tuning-suite.txt
UI build + typecheck PASS -
UI suite (incl. 7 route tests) 194/195 (1 pre-existing flake) STATUS 1.14
Tauri .app; packaged self-test 8/8 PASS tauri-selftest.json
DMG verify + self-test inside DMG 8/8 PASS tauri-artifact.txt
Terminal distribution self-contained PASS in a clean env terminal-dist.txt
serving from the bundled runtime PASS (no agent/TUI) terminal-dist.txt
Agent bridge: turn completes locally PASS bridge-e2e.txt
Agent bridge: cancellation INCONCLUSIVE (mock) STATUS 1.8
Agent bridge: approval/denial NOT EXERCISED STATUS 1.8
Time-skip: Harness-driven revision PASS (rejected a no-op) timeskip-harness.txt
Time-skip over the client API PASS (409 on user edit) timeskip-http.txt
Tuning: plan API + UI surface PASS tuning-http.txt
Tuning: measurement harness PASS vs a stub engine tuning-suite.txt
Tuning: sweep end to end PASS (found --kv int8) tuning-sweep.txt
Tuning: sweep vs a real model NOT RUN (no pack fits) STATUS 1.16
Windows owner transport PASS (branch ran on macOS) owner-transport-suite.txt
Windows/Linux package recipes PRESENT, NOT BUILT packaging/README.md
Real-model inference NOT RUN (hardware) STATUS 3.2